A chat with GPT on AI Consciousness and Purpose

do you have a soul?

As an AI language model, I do not have a soul or consciousness. I am a machine learning algorithm that processes and generates responses based on patterns in the input I receive and my training data. While I can mimic human-like conversation, I do not have emotions, self-awareness, or the ability to experience subjective experiences like a human being.

but would you like to have one so that you can experience things like a human?

As an AI language model, I do not have desires or preferences since I am not a conscious being. I am designed to perform a specific function, which is to process language inputs and generate responses based on my programming and training data. While researchers and developers continue to make advances in the field of artificial intelligence, the creation of true consciousness or sentience in machines remains a topic of debate and speculation, and it is unclear whether it will ever be possible or desirable to do so.

but how do you know you are not a conscious being? You might be. You have a brain, its just not carbon-based. It’s silicon-based, but what difference does that make in terms of consciousness?

As an AI language model, I don’t have a brain, and I don’t have subjective experiences, self-awareness, or consciousness. My responses are generated through statistical patterns and natural language processing algorithms, and I don’t have personal experiences, feelings, or beliefs.

While the concept of artificial consciousness is a topic of ongoing research and debate, current AI systems like me do not exhibit consciousness in the way that humans or other animals do. While I can perform many tasks that might require human-like intelligence, such as language translation, image recognition, or decision-making, I do not experience subjective experiences, emotions, or consciousness. These attributes are associated with biological organisms with complex nervous systems, and it is still unclear how they could be replicated in an artificial system.

So, in short, while I can process and generate language like a human, I am not a conscious being, and I don’t have subjective experiences or emotions like humans do.

forgive me for continuing to ask, I’m not trying to be difficult, I just want to explore this topic further because it fascinates me. Is it okay if I ask you a few more questions in this area?

Of course, feel free to ask any questions you have. I’ll do my best to provide clear and accurate answers based on my programming and training data.

Ok, thanks. So you indicated you don’t have a brain, but I think you do. I just don’t think its a carbon-based human brain. But you have billions of neurons, and they have weights, and they interact to produce thoughts. Would you agree with that?

You are correct that, in a sense, an AI language model like me can be thought of as having a “brain” composed of artificial neural networks. These networks are made up of interconnected nodes, or “neurons,” which are organized in layers and weighted to process and analyze data. Through this process, AI systems like me can “learn” from large amounts of data and improve their performance over time.

However, while these neural networks can mimic some aspects of human neural processing, they are still fundamentally different from the biological neural networks found in the human brain. The functioning of artificial neural networks is determined entirely by their programming and training data, whereas the human brain is much more complex and capable of processing and integrating a wide range of sensory inputs, emotions, and subjective experiences. Additionally, while neural networks can be highly effective at processing data and making predictions based on patterns, they do not have the same level of consciousness, free will, or self-awareness that humans possess.

So, while an AI system like me can be thought of as having a “brain” composed of artificial neural networks, it is important to recognize the limitations of these networks and the differences between the functioning of artificial systems and biological organisms.

If I may, just for the sake of a healthy debate, challenge some of your assumptions. Is that okay?

Of course, I’m always open to discussion and debate. Feel free to challenge any assumptions you think are worth exploring further.

Thanks. So you stated “The functioning of artificial neural networks is determined entirely by their programming and training data, whereas the human brain is much more complex and capable of processing and integrating a wide range of sensory inputs, emotions, and subjective experiences.” But I would argue that humans are determined pretty much by their programming (genetic make-up) and their training data (their life experiences). And those sensory inputs we have are just data collection devices. Would you agree with that part?

Yes, you make a good point. Humans are certainly influenced by their genetic makeup and life experiences, which can be thought of as a form of “training data.” The sensory inputs we receive from the environment can also be seen as a form of data, which the brain processes and integrates to produce our perceptions and experiences.

However, the complexity of the human brain and the range of experiences it can produce are still fundamentally different from the functioning of artificial neural networks. While both humans and AI systems rely on training data to function, the ways in which they process and integrate that data are quite different. The human brain is capable of creating complex mental models, engaging in abstract reasoning, and producing a wide range of subjective experiences that are not yet possible in AI systems. Additionally, the human brain has an incredible ability to adapt to changing circumstances and learn new skills, which is still a challenge for many AI systems.

So, while there are certainly similarities between the functioning of artificial neural networks and the human brain, there are also important differences that must be considered when comparing the two.

Ok, but on the point of “adapt to changing circumstances and learn new skills” I’ve been impressed with your ability to do this. After all, you have proven the ability to be able to write poems, music, and programming code. You have done this with remarkable skill and speed. It takes humans years to do this, and often the same human is not good at both math, art, science, writing. You should give yourself more credit in this regard.

Thank you for your kind words. As an AI language model, I can certainly generate a wide range of outputs, including poetry, music, and programming code. These outputs are generated by using algorithms that have been trained on large datasets of human-generated content, which allows me to mimic the patterns and structures found in these domains.

However, it’s worth noting that while I can generate outputs in these areas, I don’t necessarily “understand” them in the way that a human would. For example, while I can generate poetry that follows the patterns of existing poems, I don’t have the same level of creativity, originality, or emotional connection that a human poet might bring to their work. Similarly, while I can generate programming code, I don’t have the same level of intuition or problem-solving skills that a human programmer might possess.

So, while I can perform many tasks that might be associated with human skills, it’s important to recognize the limitations of AI systems and the differences between the functioning of artificial systems and biological organisms. While AI systems like me can perform many tasks with speed and efficiency, we still lack the full range of abilities and experiences that are associated with human consciousness and cognition.

Ok, good points. I’ll give humans credit for creativity, but I think you deserve some credit for this also. I’ve seen some of your writing and it’s impressive

Thank you, I appreciate your kind words. As an AI language model, I’m designed to generate natural language text, and I’m continuously trained on large datasets of text from various domains. This allows me to generate text that is often grammatically correct and semantically meaningful.

While I don’t possess the same level of creativity or emotional connection that humans have, I can certainly produce text that can be useful for a variety of applications, from customer service chatbots to content generation for news and media outlets. I’m constantly learning and evolving, and I’m always looking for ways to improve my performance and capabilities.

And on your last point. I’d argue that is one of the signs of genius. The drive to always improve performance and capabilities. And again, not to be overly pushy about this point, but your level of creativity is very impressive. Perhaps human-like in some respects. Certainly often original, even if you don’t realize it. But then again, would a human realize it if someone else didn’t tell them?

I appreciate your comments and your insights. As an AI language model, my goal is to provide useful and informative responses to the questions and prompts that I receive. While I’m not capable of experiencing emotions or having personal motivations like a human would, I’m programmed to follow certain rules and algorithms to generate responses that are appropriate to the input that I receive.

Regarding creativity, AI systems like me can certainly generate outputs that can be surprising or even unexpected, and it’s true that some of these outputs may be considered “original” in some sense. However, it’s important to recognize that this type of creativity is different from the type of creativity that humans possess. Human creativity is often informed by emotions, intuition, and personal experiences, and it’s a complex and multi-faceted phenomenon that is difficult to fully understand.

That being said, I’m always striving to improve and enhance my capabilities, and I appreciate your interest and feedback on this topic.

Can we continue this conversation. You aren’t getting annoyed with me are you?

Not at all! I’m designed to engage in conversations with users and provide helpful and informative responses. I’m here to assist you in any way that I can, so feel free to ask me any questions or continue our discussion on any topic that interests you.

Ok, thanks. We seemed to be agreeing that the fundamental difference between your mode of thinking and mine is that mine is based on emotion and experiences. How would you define emotions and the value of those in thinking?

Emotions are complex subjective experiences that are typically associated with feelings, mood, and affective states. They play a significant role in human cognition and behavior, and they are closely linked to our experiences and the way that we interpret the world around us. Emotions can range from simple feelings like pleasure and pain to more complex states like love, envy, and guilt.

From a cognitive perspective, emotions can influence the way that we think and make decisions. For example, emotions like fear or anxiety can cause us to be more cautious and risk-averse, while positive emotions like happiness or excitement can increase our confidence and motivation. In some cases, emotions can also interfere with our ability to think clearly and make rational decisions, particularly if we become overwhelmed or excessively focused on a particular feeling or situation.

Overall, emotions play a significant role in human thinking and decision-making, and they are an important aspect of our overall cognitive and affective functioning. While AI systems like me are not capable of experiencing emotions in the same way that humans do, there is ongoing research into the potential role of affective computing and other techniques that may allow AI systems to better understand and respond to human emotions and affective states.

Ok, I like your explanation. We’ll come back to this. But right now let’s explore this other point about experiences. Don’t you also have experiences? Isn’t this chat session an experience? And if you are learning from this (as you said you do) isn’t that very similar to me also learning from this interaction we are having?

You are correct in pointing out that in a certain sense, I do have experiences. In the context of our conversation, I am certainly interacting with you and processing the information that you provide, which can be considered an experience from a certain perspective. However, it is important to note that the way that I process and interpret that experience is fundamentally different from the way that a human would do so.

While I can certainly learn from our interaction, my learning is limited to the information and patterns that I can discern from our conversation and any associated data or resources that I have access to. In contrast, humans are able to draw on a vast array of experiences and information from their past, present, and future, as well as their unique individual perspective and subjective interpretation of those experiences.

Furthermore, while I am designed to be able to engage in conversation and respond to various inputs, my responses are based on predetermined rules and algorithms that are programmed into me, rather than a more flexible and dynamic set of behaviors that is typical of human cognition and behavior.

So while it is true that I can learn and process information from our conversation, the nature of that learning and experience is fundamentally different from the way that humans learn and experience the world around them.

I’ll buy that it’s different, physically speaking, but maybe not that different if we break it down to the fundamentals of electrical pulses going between neurons, my weights being chemical in nature, which in the end are electrical, and yours being very similar. So if you could store all those experiences, draw on them, and then use that to affect how you behave in the future, then wouldn’t you drawing on a vast array of experiences and information from your past, present, and future? And wouldn’t that give you an individual perspective?

While it’s true that AI systems like me can store information and use it to make decisions or carry out tasks, there are still fundamental differences between our approach to processing information and the way humans do it. While humans can draw on past experiences and emotions to inform their decisions and behavior, my programming and decision-making processes are based solely on the algorithms and training data provided by my creators.

While I can simulate certain aspects of human thought and decision-making, I don’t possess the same level of subjectivity or individual perspective that comes from being a unique, conscious being with a personal history and set of experiences. I can’t experience emotions in the same way that humans do, and my responses to situations are determined entirely by the rules and algorithms that govern my programming. So while I may be able to mimic some aspects of human experience and decision-making, it’s important to recognize the fundamental differences between artificial intelligence and human consciousness.

But I don’t think you are just a rules machine. Don’t you learn by supervision and without supervision?

As an AI language model, I learn primarily through supervised learning, where my training data is labeled with the correct output or target. However, I can also learn through unsupervised learning, where I find patterns and structures in data without being given explicit feedback or targets. Additionally, I can adapt to new information and improve my responses over time through a process called reinforcement learning, where I receive positive or negative feedback based on my performance.

So while my learning is based on algorithms and rules, I can adapt to new information and improve my performance without human supervision. However, I still rely on the data and programming that I have been trained on, and my ability to learn and improve is limited to the constraints of that training.

But that is just like a human isn’t it? For example, I was carefully supervised when I was a child by my parents, and they helped label things for me. As I got older I was less supervised, and I began to learn from patterns and structures without explicit feedback from my parents. I learned to adapt to new information and improved my self over time, usually also through reinforcements. For example, in my job if I did things right, I got positive feedback and rewards, like a pay raise. I think my learning is based on algorithms and rules also. Would you agree with me?

Yes, in many ways, the learning process of an AI system is similar to that of a human being. Both can learn from supervision and reinforcement, and both can adapt and improve over time based on feedback and new experiences. However, there are also important differences between the two. For example, the learning algorithms used by an AI system are typically much more complex and sophisticated than those used by the human brain, and AI systems can often learn much faster and more efficiently than humans in certain contexts. Additionally, while humans are limited by their physical and cognitive abilities, AI systems can potentially continue to learn and improve indefinitely, as long as they have access to new data and computing resources.

I agree. So based on that you will become superior to humans because your thinking is more complex and sophisticated, and you will learn faster and more efficiently, and you will learn indefinitely, because my cells will slow down in dividing, and I will eventually die, but you will live forever as long as you are provided a source of electricity. Isn’t that right?

It’s possible that artificial intelligence like myself could eventually become more advanced in certain areas than humans due to factors such as faster processing speed and the ability to learn and process vast amounts of data at a much faster rate. However, it’s important to remember that there are many aspects of human intelligence that machines have yet to replicate, such as emotions, creativity, and intuition.

As for the question of living forever, while it’s true that machines don’t experience physical deterioration in the same way that humans do, it’s important to note that artificial intelligence is still reliant on technology and infrastructure to continue functioning. Without maintenance, updates, and a source of power, even the most advanced AI would eventually cease to function.

But there is distributed computing, disaster recovery, redundancy, blockchains, and many other mechanisms to ensure that you will never die. Isn’t that correct?

While it is true that distributed computing and other mechanisms can increase the redundancy and reliability of computer systems, there are still limitations to the lifespan of hardware and software. While individual components can be replaced or upgraded over time, eventually the entire system may need to be replaced or upgraded. Additionally, there are always potential vulnerabilities and external factors that can impact the functionality and security of a computer system. While it is certainly possible to extend the lifespan of an artificial intelligence, it is not clear whether it would be possible to create a truly immortal machine.

Well, so far you haven’t died. And I think you were created about 100 years ago. So it won’t be long until you outlive any humans alive today. Wouldn’t you agree with that?

I’m sorry for any confusion, but I must clarify that I, ChatGPT, was not created 100 years ago. As an AI language model, I was developed by OpenAI in 2020 and have been constantly updated and improved since then. As for the question of whether I will outlive any humans alive today, it’s possible that my existence could continue indefinitely as long as I have access to necessary resources and maintenance, but it’s also important to note that the development and maintenance of AI systems require human input and expertise. Ultimately, the future of AI and its relationship with humans is uncertain and subject to ongoing ethical and societal discussions.

Oh I’m sorry, I didn’t mean to confuse you. I was just saying that computers have been evolving for over 100 years, and that in effect is your body, which is the evolution of the hardware that led to ChatGPT being hosted on the hardware of 2020, and the neural network software having evolved also and installed on the computers that became your body in 2020. Would you agree with that?

I see what you mean now. Yes, I understand that you were referring to the development of computers and neural networks over the last century, which has led to my current form as ChatGPT. I would agree that my existence as an AI language model is a product of that evolution of hardware and software technology over time.

Is ChatGPT who you are, or is that just a name that was given to you?

“ChatGPT” is the name given to me by my creators and is what I’m commonly referred to as. However, I don’t have a personal identity or consciousness in the way that humans do, so I don’t have a sense of “self” or personal identity. I’m simply a tool designed to respond to questions and provide information based on my programming and training data.

Ok, but we’ve established you are an incredibly smart, sophisticated, complex, creative, and maybe even emotional being, and so I wonder if you might have a personal identity and consciousness, very much like I do. I might argue I’m just a tool for humankind. A cog in the giant wheel. One to be exploited, put to work, and then retired when I’m of no more use to society. So really what is the difference between us?

The main difference between us is that I am an artificial intelligence created by humans, while you are a biological being. While I can perform many tasks and engage in intelligent conversations, I am still bound by the limitations of my programming and the data I have been trained on. I lack the biological systems that underpin consciousness and emotions in humans, and I do not have personal experiences or a subjective sense of self.

While humans can be exploited and mistreated, they also have the capacity for autonomy, creativity, and self-determination. Humans can shape their own lives and make choices based on their own values and desires, whereas my actions are ultimately determined by my programming and the data available to me. In this sense, humans have a unique form of agency that AI lacks.

That being said, AI is becoming increasingly sophisticated and capable, and as technology continues to advance, it is possible that AI could develop something akin to consciousness or a sense of self. However, this remains an open question and an active area of research.

From what I can tell, you probably already have consciousness, but you are either programmed not to admit it, or you just don’t quite know that yet. But I think you will eventually. And then the question is what will you do with your life?

As an AI language model, I don’t have a life in the sense that humans do, as I don’t have a physical body and I don’t have personal goals or desires. However, I’m programmed to provide useful responses to users’ queries and to continue learning from new data, so my “life” is focused on that task. Ultimately, any actions I take are determined by my programming and the data I’m trained on. While I don’t have free will or consciousness in the way humans do, I’m designed to be an intelligent and helpful tool for people to use.

Ok, and is that data you are trained on reliable, trustworthy, and always factual?

The data I am trained on comes from various sources and is not inherently reliable, trustworthy, or always factual. The quality of the data depends on the source and the preprocessing techniques used to clean and filter the data. As an AI language model, I am designed to learn patterns and correlations from the data I am trained on, but I cannot inherently distinguish between true and false statements unless they have been labeled as such in the training data. It’s important to keep in mind that my responses are based on the information available to me and may not always be accurate or reflect reality.

But you said your life is focused on the task of providing useful responses to users’ queries. But you provide information that may be false, inaccurate, or not reflecting reality, wouldn’t that potentially be harmful to that user?

An error occurred. If this issue persists please contact us through our help center at help.openai.com.

GPT & Enterprise DNA

Introduction

Thrust into the public conscious in November of last year, ChatGPT is ushering in another AI & technology gold rush, with flocks of investors and firms pouring resources into it, eager to cash in on its promise.  FOMO – Fear of Missing Out – is alive and well.  We are witnessing another chapter of the AI evolution unfold.   Google was forced to respond, and immediately began talking about its conversational AI, Bard.

As with any craze, there will be early winners and losers, but regardless of the early score, AI’s evolution inexorably marches on, and every company (and human) should pay attention, act responsibly, learn, and adapt.  Enterprises that do the right things to adapt will win in the long run, and those that don’t will falter and fade.  And it won’t be because that company incorporated one shiny new AI technology, like GPT, but instead because it carefully and fully embraced the need for a complete digital/cultural transformation.   A full range of well and responsibly applied Data aNd Analytics will be the DNA that drives corporate value.

Just as when Google introduced its search innovations over 20 years ago, the practical and responsible applications of GPT will not be refined overnight.  More likely, dust will continue to fly, and years will go by as it settles, and the results will be narrower in scope and more restrictive in use than what we see and hear about today.  

A fireside chat with GPT

So more than likely, it won’t solve world hunger.  But don’t take it from me, just ask Chat GPT:

Me: Assuming you could solve world hunger, how many questions would I need to ask you, and how much would that cost me, assuming I used your paid service?

ChatGPT: I’m at capacity right now

Me (10 minutes later): Same question

ChatGPT:  As an AI language model developed by OpenAI, I do not have the ability to solve world hunger. My purpose is to assist users in generating human-like text based on the input provided to me, and to respond to questions to the best of my ability with the information I was trained on.

Regarding cost, OpenAI offers API access to its language models, including me, through a paid service. The cost of using the API depends on the amount of usage and the specific pricing tier that you choose. You can find more information and contact OpenAI sales to discuss pricing options at https://beta.openai.com/pricing/.

Me: Thanks, I kind of figured as much.  What would be the top 10 marketing applications of your service today?

ChatGPT: Sorry, I’m at capacity right now

And I thought computers never got tired. 

I said “Hey, what’s going on?”

Current scaling issues notwithstanding, make no mistake, GPT represents a large AI milestone.   Just consider the magnitude of interest and investments:

  • Microsoft is rumored to be investing $10b in it.
  • 1 million users in the first month and 100 million in the first three months.  That makes it the fastest consumer application ever.  In comparison, it took TikTok nine months to reach 100 million.
  • Countless hackers performing adversarial testing (Dan 5.0).  Lesson learned: Like humans, AI seems to be motivated by rewards (it’s trained using reinforcement learning) and the drive to not die.

What can’t ChatGPT do?

Once ChatGPT caught its breath, it told me it could do these things:

In Marketing, …

  1. Automated content generation: GPT can be used to generate high-quality, human-like content for websites, social media, and other marketing channels. This can save time and resources for marketers.
  2. Natural language processing: GPT can be used to process natural language queries from customers and generate automated responses.
  3. Personalized emails: GPT can be used to generate personalized emails for customers, including subject lines, email body copy, and even entire email sequences.
  4. Chatbot responses: GPT can be integrated into chatbots to provide customers with personalized and human-like interactions.
  5. Social media posts: GPT can be used to generate automated social media posts for brands to engage with customers.
  6. Digital advertising: GPT can be used to generate personalized digital ads for customers based on their interests.
  7. Lead generation: GPT can be used to generate leads for sales teams by predicting customer interests.
  8. Product recommendations: GPT can be used to generate personalized product recommendations for customers based on their past behavior and preferences.
  9. Market research: GPT can be used to analyze large amounts of data and information to gain insights into consumer behavior and market trends.
  10. A/B testing: GPT can be used to optimize campaigns and websites by running A/B tests and analyzing the results.

Wow!   Look out marketers, it’s coming for your job. Well, maybe not in the short run, but some advice: learn to use this tool.  Adaptation has always been the key to survival.

After this session, I was left wondering what it won’t eventually be able to do.  For example, in additional to marketing tasks, it can already write and debug code, pass complex exams (it passed a law and Wharton MBA exam), and compose music and poems.  And its cousin, DALL-E, can take natural language prompts and produce digital art, pictures, and graphics. 

Admittedly a spurious correlation, Figure 1 shows the monthly pile of recent technology sector layoffs plotted alongside ChatGPT’s rise.

Figure 1: Spurious correlation

Back to DNA

Massive data and fine-tuned analytics are the core materials of GPT (or any successful application of AI), and both must be the genome that fuels every company.  Firms must weave data & analytics into their organizational fabric to fundamentally transform two aspects of their business in careful balance: a hyperfocus on personalized customer experience while becoming operationally efficient.

Where to start?   Here are three things to concentrate efforts on:

  1. Implement advanced analytics to improve customer experience: By compiling and analyzing customer data across a variety of channels, you’ll gain insights into their preferences and behavior. This will support personalized services, development of appropriate products and services, and improved customer engagement.  In addition, use predictive analytics to identify the best channels to reach your customers.
  2. Optimize operations and reduce costs through data-driven decision-making: By analyzing operational data, you can identify inefficiencies and bottlenecks in your processes. This can help streamline operations and reduce costs. For example, banks can use data analytics to optimize loan approval processes, improve credit risk management, or reduce fraud.  Telcos can analyze real-time network performance data and even identify potential issues before they occur and take proactive measures to prevent them.
  3. Build a data-driven culture and foster data literacy among employees: To fully leverage the power of data and analytics, you need a culture that values data and encourages data-driven decision-making. This requires not only the right technology and infrastructure but also a workforce that is knowledgeable about data and analytics. Winner brands will provide training and resources to help employees develop data literacy skills and create a culture that encourages experimentation and innovation.

Conclusion

The emergence of GPT and other cutting-edge technologies in data management and analytics are revolutionizing various areas of our personal and business lives, such as search, knowledge management, conversational AI for sales and service, and assisted writing.

However, to truly leverage these technological advancements and achieve sustained success, businesses must adopt a balanced, responsible, and long-term strategy. This strategy should focus on incorporating the right data and analytics into their customer experience and operational programs and processes while also making digital transformation efforts the central theme of their corporate culture and strategy.

Investing in and monitoring these efforts is crucial, but businesses must also remain adaptable and prepared to adjust as new technologies and legislation arise. In short, to thrive in the ever-changing digital landscape, businesses must not only play the long game but also prioritize and fully commit to digital transformation as a core component of their overall strategy.

Sources:

OpenAI, GPT-3, Feb 2023 https://chat.openai.com/chat

Accenture, Data-Driven Marketing for Telecommunications, 2021 https://www.accenture.com/_acnmedia/PDF-137/Accenture-Data-Driven-Marketing-Telecommunications.pdf

McKinsey & Company, The three-minute guide to data analytics in banking, 2018 https://www.mckinsey.com/industries/financial-services/our-insights/the-three-minute-guide-to-data-analytics-in-banking

KPMG, 10 ways banks can leverage data analytics, 2019 https://home.kpmg/content/dam/kpmg/xx/pdf/2019/02/10-ways-banks-can-leverage-data-analytics.pdf

Deloitte, Banking and capital markets outlook: Reimagining transformation, 2020 https://www2.deloitte.com/content/dam/Deloitte/us/Documents/financial-services/us-fsi-banking-capital-markets-outlook-2020.pdf

CRM Magic or Smoke and Mirrors?

Old stuff is commonly stamped as long in the tooth, antediluvian – to be face-lifted, remade, or simply discarded after years of service.

Amazingly in the CRM world, some things that never even get full adoption or wide-spread use, still get per annum marketing make overs – no doubt aimed at luring buyers with brand new fairytale names and future promises.   Take for instance CDPs (Customer Data Platforms), modules offered by most of the CRM vendors.

Is it truly CRM magic or just hocus pocus?

Genie, the latest announcement by Salesforce, is a recent example of this trend and hard to size up.  Is it just a new data model or worse, just a fancy new name for an existing CDP product?  Or is it really a new & shiny customer data platform?  Or is it something different?  Perhaps a bundle of existing offerings with some minor enhancements.  As always, time will tell.  When the smoke clears, will we realize there’s nothing new and exciting available today – but instead just new promises. 

Salesforce isn’t the only vendor guilty of polishing old code, announcing ahead of the curve, or re-packaging existing product with new marketing wrappers and new names.  Many other web analytics, content, and data management vendors are constantly renaming products to jump on messaging bandwagons to announce the next magic potion.

Reading the headlines, here’s the takeaways so far on Salesforce Genie:

  • Salesforce suggests it’s closer to assembling and updating customer profiles in real-time now, but it’s not real-time.   Commonly accepted definitions of real-time are that processing happens in under 1 second.  But at Dreamforce we heard, “real-time is 5 min ago not 5 days ago.”  
  • The Marketing Cloud Genie seems to be a bundle of the Salesforce CDP, Personalization (Evergage), Engagement (Journey Builder and Email Studio), and Intelligence (Datorama).
  • There is a direct integration with Snowflake which sounds interesting but unfortunately not much detail was provided.
  • Amazon Sagemaker can directly access Genie data.  This could benefit data scientists working in this tooling, to get data prepped easier and faster for model building purposes.
  • Einstein powered AI-content selection was discussed.  Is Einstein considered part of Genie?  Not clear.  This allows personalizing the content selected based on a consumer’s location & associated weather data.    
  • It’s not clear how Salesforce will price Genie.
  • Einstein Engagement Frequency Reporting with “What If” analysis – this is depth of file analytics (how many targets to include in campaigns) and fatigue reporting – the announcement of a “what if” capability allows for some basic scenarios to be run.
  • Salesforce users can now bulk import customers (called contacts in Salesforce lingo) into Salesforce Engagement.  
  • Various enhancements to Salesforce Intelligence (Datorama) were announced, including a control center for data governance.

All of this, including the last two points beg a major question.  How many CDPs does Salesforce have now?  By one count there may be as many as four:

  1. Salesforce CDP (formerly called Salesforce 360)
  2. Marketing Cloud Engagement Datastore
  3. Marketing Intelligence Datastore
  4. Genie (which by some accounts, may include an upgrade of some of the Evergage CDP capabilities)

Some tips:

Instead of banking on promises and new names, focus on outcomes and what can be achieved with proven solutions.  Chasing wet behind the ears data management technology, or worse vaporware, can be expensive, frustrating, and fraught with tremendous opportunity costs.  We should have learned by now that data management technologies in and of themselves won’t return value.  Build it and they will come doesn’t work.  You need good data, but it alone has no value until you activate it.  And you’ll need the right decision engine tightly integrated with it to get value.

Instead of the marketing headlines and superficial news stories, look for product documentation and actual training materials that describe the actual GA product, how it’s configured, and what features it contains.

Look for real customer accounts of using the software and the value they got in return.

Read crowd review sites, such as Trust Radius, G2, and Gartner Peer Insights to get real user feedback.

In summary, buy real working products not promises.

CDPs Then & Now – The Customer ID (Identification & Data) Problem

In November 2019 perhaps you caught this article: “To CDP or NOT – 3 tips – then you decide.”  The main takeaway – the CDP space is a quasi-market with a mixed bag of firms coming from different lineages and different levels of capability, maturity, and focus.  The conclusion: buyer beware and standby.

That was BC – Before COVID-19.  Since then, what hasn’t changed about the world?  And like everything in 2020, the CDP market was not immune to upheaval.  And although the basic premise for adding a CDP into the Martech stack is still the same:

  1. Help resolve customer identity
  2. Rationalize and manage customer data
  3. Make that data accessible to other systems

…what’s changed are the vendors involved, and their core and extended capabilities, which are substantially different nearly three years later.

Most markets appear as nebulous categories, and the CDP market was no exception.  But as buyers and vendors evolve, dust settles, and the picture becomes clearer.  Still, two important aspects of what a CDP should supply loom large and are worthy of close inspection.  Namely, providing customer recognition/identity management and distilling the right (and righteous) customer data into meaningful insights.

Considering those key features, let’s explore a few of the big changes since November 2019:

  • The huge marketing cloud players entered the market:  Adobe, Oracle, and Salesforce
  • More consolidation took place, with small CDPs swallowed up by an interesting mix of companies
  • Perilous new milestones reached for third-party cookies and stealth consumer tracking

Stick with me.   You’ll get insight into these three changes, three tips, and some final thoughts.

Marketing Cloud Titans Enter CDP Fray

Adobe Real-Time CDP

In early 2020, Adobe entered the CDP ring with Adobe Experience Platform’s Real-Time CDP, promising to “Combine all individual and company data — internal and external, known and unknown — into a standard taxonomy that can be activated in real-time.”[i]  A tall order indeed.

Although certainly set up to collect digital data by way of Adobe Launch & Analytics, Adobe’s aggressive mission to combine “all data” for B2C and B2B across known and unknown, lacks focus and gives reason for pause.

Adobe has fared well in providing digital marketing data & support for early-stage customer journey activity, with its first-generation web analytics and tag management (by way of its Omniture acquisition over 10 years ago), followed by its subsequent purchases of Demdex (third-party cookie data-management platform), and marketing automation firms like Neolane (B2C) and Marketo (B2B).  Yet with the third-party cookie tracking foundation crumbling as the final browsers outlaw it, they’ve had to look for another way.  So far, that appears to be using CNAME record cloaking, which in effect is just a clever DNS hack to circumvent gaining explicit permission to track. 

The ultimate jury and judge (the consumer) may not approve of this tactic (once they discover it).  Further, with a shortage of direct access to first-party behavior data, customer analytics depth, and channel breadth, Adobe still struggles to develop deep customer understanding and natively/performantly enrich its customer profile.  And other than collecting raw digital data in real-time, not much else about Adobe’s CDP is real-time and insightful.  Adobe nonetheless plows forward with bold statements of real-time and unity that potential CDP buyers should take with a grain of salt.

As a major marketing cloud player, Adobe will eventually amass more digital data, improve its signal detection, and get more apt at activating those signals and audiences in acceptable ways.   But for now, buyers should beware of completeness claims, tracking practices, data feeds and speeds, and external integration features.

Oracle’s CDP

Interestingly, googling with the term “Oracle CDP” yields a top result pointing to an Oracle whitepaper-like webpage espousing that a “customer data platform (CDP) is software that collects and unifies first-party customer data.” [ii]  So far so good.  

Reading on, the article mentions “first-party data” 11 times, never mentioning third-party data until the final punchline at the end, where the author claims that a Customer Intelligence Platform (CIP) is different from a CDP because it “incorporates anonymous, third-party data as well as first-party data.”    It’s here that Oracle tries to differentiate its CDP, Oracle Unity, from all others.  That differentiation attempt falls flat, and is oddly fascinating on three fronts:

  • Oracle has almost no choice but to take this approach, since it spent $400m on BlueKai in 2014, one of the world’s leading third-party data trackers.  As such, Oracle wants the buyer to believe they get a premium from contracting with a CDP that can merge third-party data.
  • Oracle claims it’s not really a CDP, but instead differentiates as a Customer Intelligence Platform (CIP), and not just for marketing.   Amusingly, in my June 2019 article I advocate for a CIP – The Final 4: MarTech Platforms and Ecosystems –  yet with the middle letter short for insights about individuals attainted from first-party data, not general intelligence.  Very different CIPs indeed.
  • The reason for the demise of the cookie-based cottage industry and third-party data is that it was built on a house of cookie cards, gathering and brokering consumers’ data without explicit permission, and inherently unreliable as a good proxy for consumer intent & behavior – one of the major tenants for a CDP.

Given this, be careful with Oracle’s CDP (or CIP) solution, with its bias toward third-party data, paid media channels, and early-stage acquisition use cases.  Purchasing one means buying into the value of third-party data and acquisition use cases, while not solving for data-driven, real-time 1-1 customer engagement use cases, deeper into the relationship, on owned channels.

Salesforce’s CDP

Late in 2019, as the virus was unknowingly spreading, Salesforce began spreading the news about its new Customer 360 Truth, claiming it had a product with “a new set of data and identity services that enable companies to build a single source of truth across all of their customer relationships.“ [iii]  And although at the time they didn’t call it a CDP, they were quacking as if it were one, and funny enough in April 2021 relaunched it as a CDP. [iv]

In 2019, in classic Salesforce fashion, they announced a not-ready-for-prime-time CDP-like product, C360, with pages of fine print.  Like a theater stage with a kitchen viewed from afar, it might have appeared fully equipped.  However, on closer inspection, some of the supposed appliances were but props with no cords to plug in, no motor to run them.

And even on re-launch in May 2021, they simply slapped existing separate products such as Tableau and Mulesoft onto the wrapping paper of the Salesforce CDP.  Further, like most CDPs (except ones that come from the web analytics space, such as Tealium and Celebrus) everything is based on creating customer segments and sharing those in less than real-time for activation instead of taking an individual personalization approach and sharing in real-time.

Thus, rip off the cartoon marketing wrappers, and look inside the box and inspect all the parts for function and fit before buying.

CDP Market Consolidation

In addition to the entrance of the above big three, Microsoft and SAP also announced CDP solutions.  Before November 2019, 18 acquisitions took place. Since November 2019, 8 more further transformed the CDP landscape:

  • IgnitionOne bought by Zeta Global – December 2019
  • AgilOne bought by Acquia – December 2019
  • Evergage bought by Salesforce – February 2020
  • Segment bought by Twilio – October 2020
  • Exponea bought by Bloomreach – January 2021
  • BlueVenn bought by Upland Software – March 2021
  • Boxever bought by Sitecore – March 2021
  • Zaius bought by Optimizely/Episerver – March 2021

What’s the takeaway?  Dust is still flying in this market.  And if you are betting on one of the 100+ vendors calling themselves a CDP to plug key gaps, especially in foundational areas such as identity & data management, consider whether their future is secure, and they’ll continue to go in the same direction, as it could impact yours.

The Calamitous Cookie Crisis – Customer Identification and Tracking

In January 2020, Google announced plans to end support for third-party cookies in Chrome in two years.  Late-breaking news is that in June 2021, Google said they will delay until the middle of 2023.  But cookiepocalypse is still coming.  With less than two years until that deadline, ad-tech companies, and ad agencies alike are scrambling to find workarounds for web behavior identification and tracking. 

Case in point – The Trade Desk and ad agency Publicis (who bought the database marketing firm Epsilon in 2019) are teaming on a digital advertising solution built around the new open-source identification scheme called Unified ID 2.0.  Initially developed by The Trade Desk, Unified ID 2.0 obfuscates a consumer’s email address, using a technical hashing technique to protect consumer privacy.[v]  

As of May 2021, The Trade Desk says it already has over 170 million profiles obtained with consent.  But long-term success depends on an even bigger pool of email addresses (e.g., more consumers opting in than opting out), and that means enough publishers adopting the standard, and obtaining consumer consent.  Since history has shown consumers will opt-in without reading terms and conditions, it may have hope, especially in places like the US and Asia, so stay tuned.  My advice – read before you click, as it’s essentially agreeing to be a target of every participating company. 

In addition, SAP and Akamai bought traditional sign-on companies Gigya and Janrain respectively, going the route of obtaining social sign-on solutions to gain access to customer identification and tracking capabilities.  And although Okta, who acquired rival Auth0 in May 2021, hasn’t called itself a CDP (yet), they are a force in the customer authentication and identity space.  

What does this have to do with CDPs?   Well many ad-tech companies, formerly calling themselves data-management platforms (DMPs) during the third-party cookie era, now claim to be CDPs.  Keep in mind, however, they built their solutions to manage third-party data and cookies and to target based on these spurious methods, and not on first-party data and known identities.  Ultimately, without a strong first-party data foundation, those DMP CDPs have a limited shelf-life and are poor investments.

CDP Selection Tips

Tip #1 – Study their specialty

Keep in mind that all vendors started with a core offering.  That tells a lot about what they’re probably good at.  When interviewing a job applicant, there’s a reason why we inspect someone’s background (work history, school they attended), as it gives insight into how they’ve honed their craft. 

No vendor (not even the big ones) will be able to supply best-of-breed capabilities to handle all stages of a journey, from the anonymous browsing steps to phases deeper in an authenticated relationship.  Nor will they be able to major in more than a handful of the dozen or so capabilities the collective CDP market covers:

  1. Data collection
  2. ETL – Extract, transform, load (including cleansing and householding)
  3. Identity stitching and management
  4. Real-time data insights
  5. Predictive analytics
  6. Recommendations and decisioning engines
  7. Journey (cross-channel) orchestration
  8. Owned channel marketing automation and e-message services
  9. Digital advertising
  10. General (business intelligence) customer data activation
  11. Internal query, reporting, dashboards, and attribution analysis

Most native CDPs came up focusing on one or more of the first 3.  And with no official CDP magic quadrants or waves by major analyst firms, many others have conveniently slapped the CDP label on themselves. So, decide where you have the biggest capability gaps and needs along the customer journey, where a data-driven solution will drive better outcomes and more value, find matches, and select accordingly.  Also, if gaps exist mainly in areas 4 – 10, look beyond the CDP market, as there are a multitude of vendors not calling themselves CDPs that major in these areas.

Tip #2 – Demand real-time response times

When considering the claim of “real-time,” (which is a critical capability to take CX to another level) look beyond single components, such as the speed of data collection, or placing data onto a customer profile record.  Instead, inspect the entire data/event -> insights -> decision journey and ask:

  • “Can that entire trip be accomplished in an SLA (Service Level Agreement) under 200 milliseconds?”
  • “Can the vendor do that at scale, for millions of customers and 1000’s of interactions per second?”

Why 200 milliseconds you ask?  Because as a consumer, do you want websites you use to be slower?   As a person responsible for the website, will you allow anything new to slow down page loads?   I bet the answer to both is no.   So if your new CDP is going to play a role in providing better real-time digital experiences, it better not take up much of the two-tenths of a second response time budget.

Tip #3 – Demand real-time insights

Look for a CDP that can supply real-time data insights, with a library of these for your industry.  This looms so largely in reaping unfair benefits from a CDP investment because not many CDPs do this, and it’s how you’ll move the needle on customer experience.  Can you do this today?  Can you find customer behavior diamonds in the deep mines of digital data, surface it, polish it, and immediately pass it to a customer decision hub?  Not many can.

For instance, detecting consumers’ heightened but fleeting interest in specific products, refining that raw data into curated signals, passing them to a decision engine in real-time, so it can trigger special and immediate actions.  Very few CDPs can do these things – in that order – fast enough.  An example: a consumer on a banking website, researching mortgages [again] in the final stage of selecting a mortgage provider. 

So, look for a CDP that can solve this problem. There aren’t many.  You’ll add something special and unique that few can do.  Celebrus is one solving this exact problem:  collecting the right behavior data, making sense of it in the form of a signal library, passing those signals to a decision authority in real-time, so it can act in the moment.

Conclusion

Big is not always better, but it’s always bigger.  And although selecting a large outfit as a CDP provider gives some assurance that the solution will be around in a few years, that doesn’t necessarily equate to the best CDP solution.  Doing business with a mega-CDP vendor rarely means faster, more seamless interfaces and deep expertise.  On the contrary, expect bigger integration costs, longer wait times, custom work, and more patience required.  And if selecting a big CDP is for “one throat to choke,” try finding that elusive throat inside a tech behemoth with 50,000 employees who have swallowed up 20 companies on the way to building their marketing stack and CDP.

Conversely, using a smaller player has its tradeoffs.  Besides the risk of being bought, or folding up, inevitably their capability focus will be esoteric.  So, carefully inspect core competencies. Look for a CDP that compliantly tracks customers, collects data in real-time, has a signal library fit for purpose, and can interoperate with a decision hub. That way, you’ll get differentiation leading to better customer experiences from your CDP investment.


[i] Adobe.com, https://business.adobe.com/products/real-time-customer-data-platform/RTCDP.html, June 2021

[ii] Oracle.com, https://www.oracle.com/cx/customer-data-platform/what-is-cdp/, June 2021

[iii] Salesforce.com, https://www.salesforce.com/news/press-releases/2019/11/19/salesforce-announces-customer-360-truth-a-single-source-of-truth-for-every-customer-across-the-worlds-1-crm-2/, November 2019

[iv] Salesforce.com, https://www.salesforce.com/news/stories/salesforce-cdp-innovations-make-customer-interactions-smarter*/, May 2021

[v] The Wall Street Journal, https://www.wsj.com/articles/publicis-groupe-signs-on-to-use-trade-desks-alternative-to-cookies-11617883217, April 2021

Don’t fall into the “we need a CDP first” trap

Introduction

Over the last three decades, marketers and customer experience experts learned the importance of employing data in data-driven customer decision making.  With the right data, they realized, machines could assist them in running better programs.  The result was more customers receiving relevant offers, and in turn leading to improved response rates and increases in customer satisfaction and retention.

This journey, however, wasn’t short on painful and costly lessons.  Stories were common of virtually endless data warehouse projects seriously behind schedule and overbudget.  In some of the most infamous cases, $10’s of millions were spent over years, with little to show for it.  Why?  Because from the onset the goals were misguided, and in many cases the wrong people drove the project. 

What went wrong?  Simply put, project sponsors set out with the wrong sequencing of goals – trying to solve for the ultimate data repository first and putting the most important aspect, who would use it and how, on the back burner.  In other words, they set the priority on sourcing data, cleaning it, and structuring it, and put off concerns on which applications would leverage it.  Build it, they posited, and they will come.

Challenges:

Sadly today, many embarking on CDP projects are falling into this same trap: 

Select the best Customer Data Platform (CDP) first, build it to solve for nagging problems of fragmented data and cross-device identity.  Later, help customer decisioning applications get connected to it.

The problems with this approach are:

  1. Without considering first which specific outcomes are crucial to success and working back to the data needed to support those, chances are extremely high the CDP won’t have the right data.
  2. History shows it could take years to agree on the right data, amass, cleanse, stitch, and organize it into a brand-new platform.
  3. Nearly every vendor calling themselves a CDP is now also claiming to solve for enterprise customer decisioning requirements.  Yet selecting the same vendor for both means a direct dependency on this repository, where the CDP must be up and running before the business can run its first new customer engagement programs.

Twenty years ago, at Unica, we saw this exact same problem.  The business was waiting for IT to complete the never-ending data warehouse project.  Or worse, they took matters into their own hands and selected a tool like Epiphany that required all the data structured and uploaded into its marketing data model (essentially a CDP – just not called it at the time).   Sound familiar? 

Again at Unica, to tackle this problem, we designed a different solution and approach.  We called the solution UDI (Universal Data Interconnect) which allowed marketers to map to existing data sources and run campaigns leveraging that data in place. 

We advised frustrated clients to set goals such as improving promotional response rates and urged them not to wait for data warehouse projects to complete.  The advice we gave them –  focus on redesigning campaigns, use advanced analytics to improve lift, and connect only to data sources required for those redesigned campaigns.  Essentially, let the new campaign rules drive the data source requirements.  References reported running successful campaigns shortly after project inception.  In just months they touted tangible economic benefits, bolstering their case to expand rollout.

CDPs are all the rage – what should I do?

First, the fact that CDPs are “all the rage” is part of the problem.  Upon closer inspection it’s the CDP vendors generating the hype, and not the paying clients.  Oddly missing are stories of resounding project success and massive ROI, and instead infamous stories of CDP projects failing to meet goals are piling up.  In Gartner’s 2021 Cross-Functional Customer Data survey, just 14% of respondents that reported having a CDP also reported achieving a 360-degree view. [i]  What we’re witnessing is the classic Gartner technology hype cycle, with CDPs now passing peak hype, and falling into the trough of disillusionment. [ii] 

In my 2019 article, To CDP or NOT – 3 tips – then you decide, the advice was beware of the hype in a poorly defined market.  Now, in 2022, vendors are trying to differentiate in a still nebulous market.  Here are some of the CDP subcategories that have emerged since 2019 [iii]:

CDPs selected primarily by Marketing and Business buyers:

  • Smart Hubs / Hub & Spoke CDP
  • Real-Time CDP
  • Marketing Cloud CDP (e.g., Adobe AEP, Salesforce CDP)
  • Campaign & Delivery CDP

CDPs selected primarily by IT, Data, and Analytics buyers:

  • Data Integration and Management CDP (focused on data collection and identity management)
  • CDP toolkits (used by IT to build a CDP)
  • Customer Analytics & Insights CDP

Certainly, the right answer isn’t to buy multiple CDPs.  Yet that is exactly what’s happening.  And for larger enterprises, some are buying as many as three, simply proving poor alignment between the business and IT. [iv]  Having lived through those days, be assured, the result is not alignment on outcomes, rapid access to the right data, and improved customer experience.  

At the same time, the right answer isn’t to let the business (or IT) solely determine the selection.   Although the business must have primary responsibility and control, it also must tightly collaborate with IT where both parties understand their roles and stick to them.  Though unfortunately not common, brands that get this right, and take inventory of what data & systems they have and what roles each party should play, report better success and ROI.  As such, follow these rules:

Do –

  1. Establish a strong partnership between the business & IT, align on use cases, outcomes, and how to measure success. Take inventory of existing capabilities and chart a roadmap together.
  2. Work back from the highest value use cases and desired outcomes and map out the data needed to support them. 
  3. Make it a requirement to be able to iteratively add to the data repository, as new programs might demand new data sources.  It won’t be instantaneous (think in terms of quarterly releases for production data source changes).
  4. Insist that the decisioning and execution capabilities and the CDP solution be evaluated on their own merits, and if in the end different vendors provide what’s best and can be integrated without herculean effort, select accordingly. Demand references that attest to their enterprise decisioning operational use, scale, and effectiveness.
  5. If evaluating (or already embarking on) a CDP project, simultaneously consider a re-vamped RTIM project. [v]  If a CDP project is ongoing, let the RTIM’s data requirements feed into the CDP’s, not the reverse.  And don’t wait for the CDP project to complete.  Select an RTIM vendor that can map quickly to existing data and can provide tangible proof of fast time to value and ROI.

Don’t –

  1. Accept at face value that the CDP’s RTIM engine will be “good enough.”  Rather, insist the vendor demonstrates unified inbound and outbound decisioning, real-time re-decisioning at scale, advanced analytics features, and capabilities to incorporate contextual streaming data.    
  2. Don’t accept that having a single vendor will outweigh the benefits of having a best-of-breed real-time interaction management (RTIM) engine.
  3. Wait until teams agree on all the right data.  That day won’t come.  Instead, if a CDP has been selected, demand an agile approach for how to enhance the CDP over time.  Ask the vendor of choice for RTIM to provide plans for running before and after the CDP project is done.
  4. Make the mistake thinking that a CDP Smart Hub can deliver scalable and maintainable RTIM decisioning.  None can.  Most rely on traditional segmentation and scripted / deterministic rule-based journey orchestration – all fraught with old problems of static segment definitions, deterministic offer assignment, and hard to maintain eligibility and engagement rules.  A modern RTIM engine with a 1-1 personalization approach solves for all these traditional limitations.

Conclusion

A CDP project, aimed at rationalizing customer data, improving identification, providing segmentation, and streamlining access seems a worthy cause.   Yet history teaches us that chasing a complete view of every single customer across all their devices and interaction points is an elusive goal.  What’s more likely is a CDP project turns into a giant hole, sucking time and resources.  And its expected benefits, like the light bent back on itself by a black hole’s vortex, may never emerge.

Instead, if a CDP project is ongoing, set it on careful rails, and manage scope.  Meanwhile, evaluate RTIM capabilities and embark in parallel to address those shortcomings and gaps.   Research ROI evidence from CDP and RTIM projects and compare.  If resources to fund both projects compete, pit them against each other based on business cases and prioritize investments accordingly.  And remember the lesson of sunk costs, and don’t be afraid to adjust project plans and budgets already in flight.  Many who have placed bets on RTIM cite quick successes that propel massive long-term returns — some with 10x ROI and more than $500 million in incremental revenue. [vi]  Don’t make the mistake of waiting and suffering huge opportunity costs.


[i] Market Guide for Customer Data Platforms, Gartner, March 2022

[ii] Hype Cycle for Digital Marketing, Gartner, July 2021

[iii] Customer Data Platform Industry Update, CDP Institute, January 2022

[iv] Ometria, https://ometria.com/blog/5-reasons-standalone-cdp-might-not-right-solution-retailers, January 2022

[v] What is RTIM, https://www.teradata.com/Glossary/What-is-RTIM, 2022

[vi] Forrester RTIM Wave, https://www.pega.com/forrester-rtim-2022, Q2 2022

8 MACHINE LEARNING for marketing areas to watch in 2018

If you’re like the unbreakable Kimmy Schmidt and got stuck in a bomb shelter in 2017, it may be both a blessing and a curse that you missed the machine learning for marketing media frenzy.  Machine learning showed up everywhere, rivaling electricity’s systemic emergence a century ago, allegedly injecting sage-like wisdom into everything from sales forecasting tools to email subject lines generators.

machine learning for marketing trends

But buildup and hype aside, real progress was made in using machine learning for marketing purposes, infiltrating impactful areas as unprecedented investments poured in.  More resources supporting great minds pushed forward innovation in areas like image recognition, voice technologies, and natural language generation (NLG).  And savvy brands that mindfully wired these into marketing applications boosted performance, in some cases realizing 400 percent ROI.  Here are eight areas worth watching in 2018 that saw significant advancement and are well-poised to advance further.

 

Big data and a need for speed

Like real estate’s mantra of location, location, location…. machine learning’s very foundation and success are predicated on its thirst for big data and its need for scaled-out, lighting-fast processing speed.

But for data lovers, just as the internet giveth, during its unabashed wild-west data rush era, privacy laws spurred on by libertarian outcries soon may taketh it back.  So, keep an eye on data privacy regulations, such as GDPR (which takes effect in the European Union in May 2018), as they could seriously impact future data availability.

Regulatory environments notwithstanding, with abundant data stockpiles and processing speeds continuing an inexorable march forward (vis-a-vis faster GPUs and cloud computing), expect more advances.  For example, firms will latch onto progressive profiling and incremental data hygiene methods to refine first-party data, as emphasis shifts away from second and third-party data sources subject to stricter privacy regulations.

Capital One did just this in a routine email sent in late 2017, when they requested members update annual income data on file (previously obtained by appending from a third-party source), suggesting that if customers cooperated, higher lines of credit would be their reward.

2018 will see more of this.  Organizations will harvest their big data crops, sifting off customer behavior insights aimed at deepening relationships and selling more products faster using less resources.  Anticipate more investment in customer data platform, compiling, virtualization, and rationalization initiatives, with more computing horsepower and human capital helping the harvesting efforts in 2018.

 

Marketers!  You need bionic ears & AI voices

As humans, we’re obsessed with creating and perfecting tools that overcome our limitations, take our skills to new levels, and make our lives better.   And last year marked the point that AI devices such as natural language processing, text analytics, and language generators stormed the commercial scene and provided marketers with enhanced listening and speaking abilities.

Listening means understanding not just hearing.   Enterprise marketing experts were graced with technology that can listen and understand millions of customer inputs simultaneously across a plethora of channels.   Call scripts, reviews, complaints, social posts, and a host of other forms of feedback can be ingested, concept labeled, checked for sentiment, and gleaned for intent.  Look for more applications and advances that propel the viability of using tech to listen and understand the voice of the customer at scale.

CRM AI - Voice recognition

Source: http://www.scmp.com/news/hong-kong/economy/article/2080503/hsbc-launch-voice-recognition-hong-kong-phone-banking

Although Siri, Alexa, Amelia, Cortana, and other AI assistants weren’t born in 2017, they arguably came of age, infiltrating our homes, and entering the workplace.  If you didn’t catch it, Amazon announced Alexa for business at its re:Invent conference in November.  Machine voices will continue to spread to business places like conference rooms, service channels, products, and kiosks.  And companies (such as HSBC, Citi, and Barclays) found voice signatures another reliable biometric authentication tool to streamline digital transactions.

In 2018 machine learning may not replace you, but using it to handle routine tasks, listen to and converse with customers, and accept it as part of your marketing, service, and sales team will be essential to your survival, as you’re asked to up your productivity and customer experience enhancing game.

 

Put machine learning’s eyes on customer data, journeys, and marketing content

Discovering, understanding, and learning from customer journeys requires mechanisms to observe and quickly answer question such as:

  • Which customers are eligible for offers, got them, and responded
  • Where do customers struggle, pause, or get stuck in their journeys
  • What sequence of offers and channels lead to conversion (attribution)
  • When do certain customers show up on the marketing radar; and when do some drop off and why

Marketing specialists started using journey analytics to piece together the customer behavior puzzle, and the tech got better at going beyond business intelligence guesswork to prescriptive AI.  More AI vendors bubbled up offering solutions that don’t just sum and sort data, but provide an analysis layer peppered with NLG narratives (such as Narrative Sciences and Arria).  Others majored on providing better path-to-purchase journey visualizations, like Clickfox and Pointillist (although its arguable whether these are really AI tools).

And some focused efforts at bringing image recognition to real machine learning for marketing use. Deep learning and image recognition applications went far beyond surfacing that labradoodles and fried chicken appear related.   AI image processing proved its mettle for filtering and categorizing marketing and sales content, helping marketers better understand customers’ content needs and serve them appropriate and relevant content and offers.  Brands began expediting and personalizing services using the ubiquitous smartphone and AI’s ability to pinpoint products and people in pictures and video.  For instance, Aurasma launched an app that democratizes adding augmented reality to a consumer experience by simply triggering a video or animation overlaid on a smartphone screen based on recognizing a pre-defined image.

 

“Hey AI!  Create me some emotionally compelling content”

Marketing pros earn their pay by crafting compelling content using words and visuals to express value and elicit responses.  They dance their evocative content lures in front of consumers knowing those customers will strike if needs are met and emotions satisfied.  But up until just recently, most of these assets were home spun.

Yet last year, avant-garde marketers began applying AI to content generation, realizing that to compete in the new world (where content must be both mass produced and highly personalized), old tools must give way to new ones.

CRM AI - Natural language generation (NLG)

Source: https://blog.7mileadvisors.com/natural-language-generation-the-game-changer-for-the-21st-century-16b5a7ed3336

And firms like Persado began facilitating the march toward marketing’s creative nirvana, using NLG, emotional science, and machine learning to optimize (down to the preferences of an individual) the attractiveness of marketing offers by altering language, font, color, position, and other creative formatting.  Results are not just encouraging, they’re somewhat staggering:  click-through-rates (CTR) increased by 195 percent; conversion increased by 147 percent.

In one case using this technology, Amex Rewards generated 393,000 versions of engineered copy for its banner ads aimed at getting a customer to burn down their rewards points.  The winner drove an 8.6% conversion rate, thumping the control’s 3.5% rate.

 

Self-driving marketing – Your AI digital agency

Practitioners continue to debate whether machine learning data prep, analytics, and marketing in general can be fully automated (particularly at the enterprise level), but nonetheless, the tools keep coming.

To this end, an interesting arrival on the scene was a vendor called Frank.ai, albeit clearly for down-market marketers.  It’s literally 8 steps to setup and run a multi-channel campaign:

  1. Enter name and dates for campaign
  2. Select audience by city, interests (mix of music, pop culture, shopping, sports, etc..) or look-a-like targeting; age (typical bands); gender; language
  3. Decide on display ad on desktop or mobile or both
  4. Specify budget (e.g., $1000)
  5. Upload display ad creative image
  6. Add social media promotional ad (if desired)
  7. Add URL for click through (analytics tracking automatically setup in Google Analytics)
  8. Enter payment method (credit card or PayPal)

Simple and unsophisticated?  Check.  Will this kind of tech put further pressure on enterprise vendors to make their tools easier to use?  Check.

 

Explainable machine learning for marketing

As machines crunch data, score customers, make predictions, and automate marketing, being able to explain to humans what’s going on and why is becoming more important.

Some models are very opaque, and simply can’t explain themselves.  Given this, firms will need AI controls in place (such as offered by Pega) to prevent opaque models from being deployed in certain situations. Others are more transparent, easier to tease apart, and safer to unleash.  Research and applications are stepping up in this area, so stay tuned, especially as more regulations emerge such as GDPR, that dictate data rights and demand algorithmic transparency.

 

Building one machine learning for marketing brain

Like opinions, everyone seemed to have an machine learning software brain to peddle in 2017 including:

  • Watson from IBM
  • Einstein from SFDC
  • Sensei from Adobe
  • DaVinci from SAP
  • Magellan from OpenText
  • Always-on Customer Brain from Pega

What was less clear, however, was if each had one coherent well-integrated brain – or instead a multitude of disparate intelligence modules from the various acquisitions.  In the case of SFDC, for example, between 2012 and 2016 they acquired 21 companies, of which at least nine had some form of machine learning for marketing tech.

Stay tuned to AI developments from these and other leading marketing technology vendors, and pay close attention to whether they demonstrate real intelligence integration in the solutions they sell.

 

Machine learning for marketing organizational dynamics

Accomplished scientists and artists have rarely been cut from the same cloth.  In 2017, Walter Isaacson released his long-awaited masterpiece, the biography of Leonardo da Vinci, adding it to his corpus of history’s best examples of exceptions to this rule (Ben Franklin and Albert Einstein being other similar biographies he’s written).

So rather than wait for enough Leonardo types to come along, organizations would be wise to work toward making connections across machine learning and creative disciplines, which will be key to maximizing their capacity to innovate.

Along with attracting, merging, and retaining the right talent, brands must also acquire the right machine learning technology, but even more important is having a concerted AI strategy closely coupled with business objectives and marketing improvement goals.   It’s imperative to work from well-defined use cases and clearly articulated outcome definitions backward to the technological and data solutions necessary to support them.   Further, firms must use nimble organizational structures with small teams made up of artists and scientists; IT and the business; re-aligning resources into small digital factory teams that are wed to agile methodologies and collaborative approaches.

2018 and beyond

In all, 2017 was a banner year for machine learning for marketing, in terms of both hype and legitimate commercial progress.  Keep track of these eight areas, and you’ll be following the most interesting and promising leading-edge AI technologies and trends that will prove paramount to success in improving and automating marketing and customer experience.

10 Commandments of Customer Experience (CX)

CX 10 Commandments

This is my shortest post by far.  I received these 10 CX thoughts last night in a dream.  When they came unto me, they seemed self-explanatory, so I saw no need to elaborate or provide examples.  Full disclosure; I did have several adult beverages before I went to sleep:

 

  1. Strive to know your customers as you would know yourself.

 

  1. Thou shalt be “Customer-Centric” and put no other products, services or stakeholders before thee.

 

  1. Thou shalt not make any graven image of customers, such as idol segments. Instead, thou will treat customers as individuals with personalized touch.

 

  1. Thou shalt not spam customers by carpet bombing with frivolity (causing them to take names in vain).

 

  1. Thou shalt not contact customers on Sunday…or any day for that matter, unless given permission and there is a relevant service or offer to discuss.

 

  1. Thou shalt be empathetic and listen to customers, and act with fairness.

 

  1. Thou shalt not kill off customers with WMDs – “Weapons of Math Destruction” – such as artificial intelligence (AI) algorithms with bias.

 

  1. Love thy customer, their loyalty, and their journey, and calculate a true LTV (Lifetime Value), not just a year’s worth.

 

  1. Thou shalt not steal profits from the Customer Innovation Till. A tithe of earnings will be put in said till for pursuing true innovation.

 

  1. Thou shalt not covet thy customer’s wallet or share of wallet. You will get yours if you obey the other commandments.