April 2025 - Marketing Technology

April 11, 2025April 11, 2025

To CDP or NOT – 3 tips – then you decide

A Customer Data Platform – or a CDP – is a software category first defined in 2013 by Raab Associates (headed by longtime marketing technology analyst, David Rabb). Shortly thereafter, Rabb Associates created the CDP Institute as a CDP promotional and education vehicle (and today actively manages it).

If you’re confused by CDPs and whether you need one, you’re not alone. The category includes a mixed bag of companies in a wide variety of shapes, sizes, and abilities. As of mid-November 2019, the CDP Institute listed 94 in its directory[i]. In the rest of this article, you’ll get the condensed history shedding light on the origins of CDPs, useful forensics on the category, and 3 tips to help you decide which (if any) to consider in your Martech or CX stack.

Wait, what’s a CDP?

Let’s start with the definition of a CDP from the CDP Institute:

“A Customer Data Platform is packaged software that creates a persistent, unified customer database that is accessible to other systems.” [ii]

And the CDP Institute follows this with a first-level unpacking of the definition:

“Packaged software”: the CDP is a prebuilt system that is configured to meet the needs of each client.

“Creates a persistent, unified customer database”: the CDP creates a comprehensive view of each customer by capturing data from multiple systems, linking information related to the same customer, and storing the information to track behavior over time.

“Accessible to other systems”: data stored in the CDP can be used by other systems for analysis and to manage customer interactions.

On the Origin of CDPs

In my first Martech job with UPS in 1993 we signed a contract with Harte-Hanks for their “Marketing Customer Information File,” or MCIF. Interestingly it was:

Packaged software: Harte-Hanks sold it first to banks, and then to companies like UPS.
A persistent and unified customer database: It included household and customer ids.
Accessible to other systems: It had export and import capabilities.

Apparently, we bought a customer data platform at the time and didn’t even know it! And here’s the thing: Harte-Hanks’ customer database wasn’t even relational (for the techno nostalgists out there, it used a hierarchical file system). And they shipped us updates once per month, on tapes! The point is this 1993 software (according to the CDP Institute’s definition) technically qualified as a CDP, which isn’t very reassuring if you are looking for criteria to judge a vendor’s worthiness to serve your current-day customer data needs.

By the late ’90s, relational databases (with SQL interfaces) had taken over and using this technology, campaign management vendors and marketing consultancies spawned the 2^nd generation of mostly hand-crafted customer data platforms known as marketing data marts (many of the original on-premise campaign management systems tapped into these). Practitioners unsatisfied with what they got from their IT partners, built these using database solutions like Oracle, IBM DB2, Microsoft SQLServer, and Teradata. Meanwhile, IT continued building out data warehouses.

Then, IT moved into the era of big data (or NO SQL) solutions and building data lakes. From warehouses to lakes, to oceans, the mentality was “store it and they will come.” But (for the most part) come they did not because IT hadn’t designed for specific business outcomes.

Nonetheless, IT had accumulated vast data reserves, and as a result, reporting firms such as Cognos, MicroStrategy, and Business Objects tapped in with general-purpose reporting and decision-support software. These evolved into the wide-class of business intelligence tools available today, including tools like Tableau.

With the advent of websites and web banner ads during the .com boom, programmatic bidding platforms needed a database to store audiences (from all those cookies and device ids) and overlayed 3^rd party data. These became known as cookie pools or DMPs (Data Management Platforms). Further, on owned websites, brands started tagging and tracking visitor behavior to understand traffic patterns and in the hope of eventually providing site personalization. This spawned tag management and website personalization firms. And there you have it: the genealogy of CDPs (Figure 1).

Figure 1: CDP Family Tree

The CDP Convergence Era

By 2013, a host of factors were affecting these technologies and the data landscape:

The cloud movement caused Campaign Management and Email Service Providers to form Marketing Clouds.
Website personalization expanded to all digital channels.
Tag management/web analytics became commoditized (thanks in part to Google Analytics being free).
The DMP market slowed because cookie pools were never people-based (making it impossible to do 1:1 personalization).
The CX revolution forced marketing, sales, and service practitioners to think more broadly about customer data and customer journeys.
Big data management suppliers were looking for valuable use cases.
Marketing data marts, for the most part, had been absorbed by IT’s 360 data initiatives.
Marketers faced growing issues with the variety, volume, and increasing complexity of data.

With these dynamics at play, and to their credit, the CDP Institute attracted packaged-software businesses from various origins that qualified for inclusion in its directory. Many found this new label attractive as their existing markets softened, got saturated, or commoditized. In the ensuing five years, a handful of CDPs grew to nearly a hundred.

Late last year, Forrester’s Joe Stanhope and Stephanie Liu wrote an article entitled, “For B2C Marketers, Customer Data Platforms Overpromise and Underdeliver.” The blog promoting it summed up their view: “Marketers, these aren’t the ~~droids~~ data platforms you’re looking for.” [iii] This is especially true for enterprise B2C marketers.

To further understand the complexion of the organizations that collected under the CDP umbrella, I separated them into sub-categories in terms of lineage. I analyzed the 94 companies in the CDP Institute’s directory as of November 2019 and found this distribution:

Origins	Companies	Avg # Employees	# Bought
Native CDP	36	71	7
Marketing Automation (e.g., Email Provider, Campaign Management)	15	112	1
Data Management (e.g., Data Quality, ETL, Data Management)	12	249 *	4
Customer Analytics (e.g, Business Intelligence, Customer 360)	9	47	3
Recommendation Engine (Product, Content, Real-time Recommendations)	8	145	0
Tag Management / Web Analytics	5	152	1
Madtech / Attribution / Journey Orchestration	5	96	0
Lead Management	4	68	1

*Removed Informatica (4700 employees) from calc to avoid over skewing

First, the 36 native CDPs:

Most were born recently and didn’t spring from an earlier category.
They average 71 employees; many are startups; consistent with a nascent category.
Although some claim to have cross-industry experience most have a weight of experience in one sector.
Three notable examples are Lytics (CPG), mParticle (Media & Entertainment), and SessionM (Restaurants)
For many, their customer base is mid-market and/or B2B slanted and not enterprise.
In the last 5 years, the native CDPs have become acquisition targets (bigger companies bought 7 of the 36 between 2015 and 2019).

As we saw, to qualify for inclusion in the CDP Institute’s directory, the solution must prove it persists data, yet that doesn’t mean the vendor has useful business experience with that data. And interestingly, 25 of the 94 didn’t originate as Martech vendors, but instead as general-purpose (or even sales force management) data firms. In other words: buyer beware in terms of experience with enterprise-scale, marketing use cases, and B2C data.

Some of the critical capabilities beyond the basic CDP Institute criteria to keep an eye on are:

Data scrubbing – hygiene on bad or missing data
Data appending – attaching net-new data attributes to an existing profile
Data aggregation – summarizations, calculations, and pattern detection to create predictive fields for high-value use cases such as propensity to buy or churn
Data streaming – continuous feeding of data as it’s created
Identity resolution – device matching, stitching, and rationalization to pinpoint the person
Data visibility and privacy – compliance, security, and preference management features
Ecosystem connectors – Pre-built interfaces to streamline interchange with other platforms

As you wade through all the Institute’s vendors, as well as the brand-new (and untested) CDP offerings by the mega Martech vendors (Adobe’s CDP, Salesforce’s Customer 360 Truth, SAP CX Suite, Oracle’s CX Unify, and Teradata’s Vantage CX), and any others happily slapping on the CDP label, carefully inspect the above critical capabilities. And when doing so, consider these tips as you decide whether to license a CDP.

Tip #1: Feeds and Speeds Matter

As you ponder data accessibility, think about the speed of access required for real-time customer engagement. In my June article “The Final 4: Martech Platforms and Ecosystems,” I opined that one of the four linchpin platforms for effective real-time engagement should be a Customer Insights Platform (CIP), going on to compare and contrast it with a CDP. The spoiler alert is that a CIP is NOT the same, and very few CDPs qualify as a CIP.

A CIP’s primary job is to feed the right individual-level data at the right time (often in real-time) to the Acquisition and Relationship Execution platforms (details in the above article). Some of the best data to predict current intent comes from recent digital interactions. A CIP, which is a transactional platform, can’t also be a business intelligence platform. CIPs, designed to transact in real-time, access a customer profile (and the sub-strata of that data for an individual) in milliseconds not batching across them in minutes or hours. Consequently, ask yourself, “Do I need a tool for business intelligence or real-time 1:1 execution action?” If you care about real-time feeds and speeds, and the outcomes you’ll get with a well-architected execution platform, you want a CIP to feed it, and many of the CDPs won’t work.

Another crucial consideration is the latency and scalability when streaming digital channel behavior data in real-time. Notice in Figure 2 that data must flow in real-time (not batch) into the customer profile managed by the Relationship Execution Platform. Other slowly changing data, such as core customer records and product holdings, can enter periodically, and you might use a CDP as that data source.

Figure 2: Customer Insights Platform – Example data processing

Here are the CDPs from the directory with origins in Tag Management (and examples of their enterprise-grade experience)

Celebrus – Achmea, BOA, HSBC
Commanders Act – Credit Mutuel, Engie, Nestle
Ensighten – OI, TUI, United
Tealium – Cox, HSBC, Vodafone

Of the CDPs, the Tag Management vendors are best suited to capture and stream real-time digital data (handling volumes such as 5,000 transactions per second), but keep in mind some require more involved multi-page tagging to get the right behavior indicators.

Tip #2: Inventory moments of truth – focus on the data needed to detect them

Data, like oil, is useless when trapped in the ground or in crude form. Value comes from tapping into it, refining it, distilling it into a refined energy product, and dealing responsibly with its combustion and aftermath. Your job is to find detailed insights that fuel a productive understanding of customers’ behaviors, demands, and intent. Relevance happens when you react swiftly and with grace, delivering personalized offers, services, and recommendations. So, the question is, can CDPs help you with this challenge?

That depends. In “Deconstructing Customer Data Platforms – Myth vs. Reality,” [iv] the Winterberry Group concurs and cautions that “Different CDPs have different levels of expertise at managing different levels of data capture.”

When customers use websites, mobile apps, and other digital devices, they emit signals showing interest in products, completing tasks, subscribing to things, getting alerted, and interacting with their environment. If brands effectively tap into these signals and react with extraordinary timing and class, they can achieve a competitive advantage. But these moments are fleeting.

For instance, a customer searching on a site with the term “early termination fee” could be a clear sign churn is coming in minutes. A customer dwelling on a mortgage page for the second time in a day might be making a final decision right then on who gets their home loan. Subscribing to a 401k newsletter may be the first in many retirement interactions. Customers’ proximity to your store (or a competitor’s) might hint shopping is imminent.

So, make a list of these events, tap into them, store patterns of data and flags about them, and devise a way to act on them.

Tip #3: Don’t confuse CDPs with more conversation and more action

Better customer engagement and conversations don’t necessarily require more master data management. But don’t get me wrong. If you don’t have well-organized customer data, then a CDP’s data collection, identity resolution, and unification capabilities could prove useful to drive the right engagement. Yet if you are a large enterprise, chances are you have scores of ongoing data unification efforts, and what you probably need is rationalization and coordination, not another data repository.

In terms of orchestrating personalized customer conversations, several CDPs originated in real-time interaction management, or in the website, product, or content recommendation space:

Vendor	Origins	Major experience
Blueshift	Content Recommendations	eLearning & Media
Boxever	Real-time Interaction Management	Travel & Leisure
Evergage	Website Personalization	Retail & Tech
Jahia	Content Recommendations
Manthan	Product Recommendations (BI vendor that bought Rich Relevance)	Retail & CPG
NectarOm	Content Recommendations
SmarterHQ	Real-time Interaction Management	Retail
SymphonyRM	Real-time Interaction Management	Healthcare

Some are good recommendation engines, and have specific areas of experience, but remember it’s not a recommendation engine you are after in the CDP area. It’s outcome-oriented customer data management. So, don’t get distracted by recommendation capabilities when what you seek is the ability to handle data feeds and speeds, find insights, and activate an execution platform to deliver at moments of truth.

If you don’t have an adequate relationship execution platform, evaluate those separately. In that process, look at the strongest real-time interaction management (RTIM) platforms that major in serving recommendations on paid and owned properties.

Conclusion

Like the first minute of a roller coaster ride, the CDP train is dragging us up to the hype precipice and what’s in store when it plummets down to the trough of disillusionment is unknown. No doubt, it will be fast, furious, and freighting, especially for those heavily invested in this technology. Because the CDP category is a mixed bag, very different firms will shop in the bin, some making head-scratcher acquisitions. Some CDPs will go out of business. Quite possibly, the plunge has begun, with Mastercard’s recent purchase of SessionM, D&B’s buy of Lattice, and ARM’s purchase of Treasure Data.

Focus on the use cases (and data needed) that improves your ability to serve timely, relevant, and personalized offers and services. Codify the important data and the speed it must move into your decision-making solution. If a CDP has components that help you, and you get those at a fair price, consider plugging them in. But remember, you’ll inherit redundant features, so be wary of the premium you’ll pay for those and ensure you can either use them or work around them. Further, assess how difficult it will be to pull them out should your plans change.

And if after all this you’re still confused, consider sitting on the sidelines until the dust settles, using your existing data-management technology, and watch others take the wild ride.

[i] CDP Institute, https://www.cdpinstitute.org/directory, 2019

[ii] CDP Institute, https://www.cdpinstitute.org/cdp-basics, 2019

[iii] Forrester, https://go.forrester.com/blogs/b2c-marketers-and-cdps/,2019

[iv] Winterberry Group, https://www.winterberrygroup.com/our-insights/deconstructing-customer-data-platforms-myths-vs-realities, 2019

April 11, 2025April 11, 2025

Pay-for-Use: The Value of Modernizing CRM Pricing

Consumption-based pricing models for software, pay-for-use, are growing in popularity. And it’s not just limited to pricing software services. From insurance companies to content services, pay for what you need, when you need it, is becoming the rule.

Taking a page out of an old pricing book used for decades by electricity and water companies, some software companies, like Amazon and Teradata, are using it to price their cloud services. So, is the time right for CRM companies to price software like cloud infrastructure or electricity?

New ideas are often born by taking what’s old hat and applying it elsewhere. From art to consumer products, innovators create something new by appropriating old ideas and recasting them with impeccable timing – right when that elsewhere is ripe and ready for it. In the case of cloud software providers, it happened naturally once those companies could deliver it like a commodity through fast internet lines. What people were buying was no longer a thing – since there was no CD to ship. Instead they were buying a pure service.

Yet today, CRM companies still sell their software as if the consumers were event planners booking hotel conference rooms. In this pricing model, the seller wins provided they pre-book enough of their inventory, regardless of whether buyers use the rooms. Since the consumer can’t perfectly forecast how many seats they’ll need, they hedge and often pay for more than they end up needing.

Once upon a time

Before the internet delivered acceptable speeds, software vendors peddled perpetual enterprise pricing, locking clients into ownership models with follow-on maintenance agreements. Like buying a car, the buyers forked out a large payment for the asset upfront, and then, for some stipulated period, usually 3 to 5 years, they locked in a service contract at a rate between 18% -22%. With the advent of software-as-a-service, subscription arrangements replaced perpetual models. From an accounting perspective, that meant expensing versus capitalizing, making CFO’s happier.

Yet the standardization on ideal accounting arrangements didn’t result in the harmonization of contract options, term timeframes, or pricing metrics, and didn’t lead to an ideal pricing model, certainly not for the CRM buyer. And presently, there’s still a plethora of pricing options that confuse both buyers and sellers. Buyers aren’t paying for what they need, and it’s hard to measure value.

Consumption, demand readiness, and value

Customers use some services steadily and other services intermittently. Take electricity service – it’s used every day, with consumers reaping benefits regularly. Tax software buyers, on the other hand, use it for just a few months out of the year. To get the electrical service, customers pay a flat connection fee, and then a per-unit charge each month based on consumption. With tax software, they pay a flat fee for unlimited use for the year. In both cases, the consumer just assumes the service will be available whenever they need it.

How should a consumer think about cost and value in these two situations? For tax software, buyers compare it to other alternatives, such as hiring an accountant. In terms of value, they rate efficiency, speed, convenience, and how much it might reduce their tax burden. For electricity service, assessing value is in the eye of the beholder (or perhaps the consciousness of the survivor), and ties directly to the Maslow hierarchy of needs. Some electricity flowing into a home serves basic needs, like keeping bodies warm; some powers security systems for safety; some enables virtual connection to others, serving the need for belonging, esteem, and self-actualization. Given this, how valuable is a kilowatt-hour of electricity to a consumer? Are all kilowatts created equal?

It quickly becomes philosophical and hard to measure this value, as we saw recently during the electrical shortage in Texas. If you’re freezing to death, you want warmth at any cost. So if you survive it, and then receive a bill for $17,000 for a few days of electricity, were you really willing to pay that much? And similar to urgent care services, in the moment of need, how do factors like competitive alternatives, consumerism, and switching costs play in, if at all?

CRM software users, on the other hand, don’t buy access to this software to save lives. As such, they have to assess value in other ways, by asking questions such as:

Will our marketing programs become more effective?
Will we streamline sales and service operations?
What return will we achieve on this investment?

But even these expressions of value are hard to measure.

Measuring CRM value

Accurately measuring CRM software’s added value is not trivial. To do it, you’ll need a measurement system in place to count new outcomes and compare them to a baseline (control group). In that process, depending on the breadth and depth of the CRM package and your intended use (e.g., whether it’s for Marketing, Sales, and Service), you’ll need to answer detailed questions such as these:

How much did response and conversion rates improve due to new marketing treatments?
Were the improved rates directly attributable to the CRM’s marketing software?
Are the service agents spending less time on calls?
How much of the cost-saving is a direct result of using the CRM software?
Did we reduce customer churn and improve loyalty?
Was the sales automation software able to help us find better sales opportunities, shorten sales cycles, and/or streamline sales activities?

Answering these questions accurately will require careful planning. Large enterprises have constantly evolving CRM stacks made up of dozens of software packages (each implemented and upgraded at different times), making it difficult to coordinate multiple experiments while avoiding experimental contamination. But outcomes matter. As such, it’s imperative to do the hard work and set up the measurement systems. Otherwise, CRM vendors will lay qualitative claims when success happens, and run and hide when results are poor.

In a bold attempt to settle the attribution argument, some vendors, believing they have a secret weapon, are offering a fee-for-performance model (sometimes called “gain share”), charging based on the incremental returns enjoyed when their special sauce is added to the mix. And this approach requires an even more meticulous measurement and reporting system that both parties can agree to, with one of the parties (or a 3^rd party) responsible for continuously administering that system.

Pricing models

Because the gain share approach is so complex, most CRM companies don’t offer this. Instead, they use several popular metrics, often mixed, as the basis for traditional pricing models:

Metric	Metric type	Example of metric
Entity	Number of entities receiving the service	Pay by type of user, by seat, by tenant, by environment, by profile, by visitor, by contact, by customer, by prospect
Capacity	Amount needed/used	Number of service cases, amount of storage, emails sent, decisions made, events, interactions, API calls, or other compute units (e.g., Teradata Vantage Units[i])
Service type	Level of service given; features provided	By edition (standard or premium), by channel, by features offered, or application used
Term	Length of contract	On-demand, monthly, 1-year term, 3-year term, etc.

Confusing matters, many of the entity metrics vary in terms of use. For instance, some users rarely touch software, while others are in it every day. Some profiles are less chatty and engage rarely, while others interact often and through expensive, non-digital channels. Sometimes users store data and don’t even know it.

Although businesses want to pay for use, they also want predictability for overall software costs, as budgets are annual. That makes it even more important to use models that accurately estimate use, charge within budget cycles, and true-up in a way that doesn’t shock finance and accounting. Hidden fees, add-ons, and excess use charges that surprise clients can be detrimental to long-term relationships.

Most importantly, buyers want value from use. To return value, the software must:

Be easy to get up and running and support (setup costs, integration, etc.) – part of the total cost to use, often higher in the early stages of use.
Be assessed along with other recurring costs (training, advanced analytics expertise, creative & support services) required to derive benefit.
Supply measurable returns, and when compared to total costs, proven to furnish a high return on investment

Consider some of the main aspects to the price consumers are willing to pay for any service:

Meeting a Service Level Agreement (SLA) – Whether it’s how fast a pizza arrives, or a webpage loads, customers shop based on a level of service expected.

Availability at time of purchase – Will the service have capacity at the exact time needed? If so, what will the spot price be? For example, when shopping for a hotel room or airline seat, consumers face the tradeoff of whether to book early (paying upfront to lock in availability), or whether to roll the dice, hoping the price drops and the service is still obtainable. If demand is high and supply is low as the inventory approaches expiration, the price could be higher.

More is less – A volume discount lures clients to obtain more or commit to a longer contract term upfront, with the risk that some of the bought or baked in reserve will go unused. Here again, customers roll the dice when buying higher amounts of service units upfront (or even bundles of services) for a lower price-per-unit charge.

Pay-for-use

In a pay-for-use model, vendors need service meters, accessible to both parties (e.g., an electricity meter on a home) to measure consumption. Then, over some period, usually a month, the vendor counts the amount of service used.

And in terms of billing, the conversation might go like this:

“Let’s help you guess at how much you’ll need. For the first 12 months, we’ll bill based on those assumptions.”

“At the end of 12 months, we’ll look at actual use and if you’ve used more, we’ll true up… “

“Oh, and at that point, we’ll give you the choice to work the difference into a new monthly fee.”

If applied to CRM software, the focus shifts from the entity receiving the service (the user or contact) to the amount of service they consume. For the model to work, the service units must be understandable and closely related to outcomes. As an example, all parties will be able to easily understand a metric such as emails delivered. However, one like IOPS, which is tied to infrastructure consumption, will be too technical and obscure and should be avoided.

Conclusion

A pay-for-use pricing approach can work for CRM software and should be good for both buyer and seller. It’s fair and efficient if there’s a way to estimate upfront consumption, meter it, and if the unit prices are understandable. In terms of fairness, there’s certainly plenty of providers and competition in this market, with Trust Radius listing 261 companies as of this writing[ii]. And CRM vendors are in better shape to provide burstable services to handle surges in demand, as the cloud infrastructure they use becomes more elastic.

Indeed, from the onset, it seems enticing and comfortable for both sides. For the seller, it rewards them for finding more ways to expand the use of their software. For buyers, it removes intra-contract restrictions such as the number of users or channels. Yet at the same time, it does put the onus on the buyer to watch their use so:

It stays within their assumed draw-down levels
They aren’t surprised and can afford the true-up costs

Consequently, if CRM vendors adopt this model, buyers will need the discipline to use the services carefully, as they’re ultimately responsible for every service unit that flows. As Tom Bodett of NPR used to say for Motel 6, “We’ll leave the light on for you.” Although that has a heartwarming ring, remember, with a pay-for-use CRM pricing model, it’s the buyer that will pay for the service powering that welcoming light, so use your CRM electricity wisely.

[i] Teradata.com, https://www.teradata.com/Cloud/Pricing/Consumption-Pricing, 2021

[ii] TrustRadius.com, https://www.trustradius.com/crm#products, 2021

April 11, 2025April 11, 2025

Will AI in digital marketing lead to marketer obsolescence?

I just returned from attending several spring digital marketing conferences – Adobe Summit and Martech West. In both the art of the possible was on full display and got me thinking about whether fully-automated AI-driven digital marketing could ever be a thing, and what realistic automation goals look like.

Spurred on by conference highs and three years of ramped-up writing on AI in digital marketing, it seemed the right time to step back and ask a few questions:

Is fully automated digital marketing even possible?
Will marketers wake up one day and find themselves obsolete?
What should marketers be doing today to prepare for tomorrow?

Is fully automated digital marketing even possible?

Digital marketing continues to be on the leading edge of AI advances and high-tech innovations. Surveys repeatedly indicate AI professionals aim their efforts at infusing intelligence into digital marketing. In fact, for nearly 10 years running, Rexer Analytics’ Data Science survey lists these digital marketing pursuits in the top 10 analytics goals of data scientists [i]:

Improving understanding of customers
Retaining customers
Improving customer experiences
Selling products/services to existing customers
Market research
Acquiring customers
Improving direct marketing programs

And every day we’re inundated with news of more advances in marketing automation such as:

Customer behavior data that is automatically processed, like whether someone used a mobile app in the last 30 days
Lights-out, always-on marketing programs that are triggered automatically and run themselves, often fueled by the pre-computed customer behavior data
Advances in process automation and examples of programmatic marketing jobs that previously involved human intervention, such as wave/drip campaigns and paid media buys
Helper tools out the wazoo, like website and mobile app builders, grammar checkers, content taggers, SEO using AI [ii], and many more

Couple all that with a host of technological factors spurring on AI and automation like:

Digital devices, tags, and pixels gathering data at breakneck speeds
Falling data processing and cloud storage costs
Free open source tools and freemium pricing models that put more software in the hands of more marketing users
Adaptive algorithms that learn which offers customers respond best to without supervision or manual testing
Natural language generation handling some previously thought to be untouchable human tasks, such as automated writing (although today’s NLG performs just simple writing tasks)

From all this, some might conclude the end of human-powered marketing is close at hand. But others, unconvinced by these tenuous signs, might simply respond, “Poppycock!”

With years of history to draw on, during which digital marketing was born and came of age, perhaps the real answer lies somewhere in between. In pouring over this progress and assessing the full landscape of today’s automated processing and AI in digital marketing, it’s clear humans have played and will continue to play a crucial role.

In the last 25 years, we’ve seen human ingenuity cause marketing to go digital, get more scientific, and its content to get richer, more compelling, and hyper-personalized. So, without a doubt, AI in digital marketing is a real thing. It has improved reach and targeting, quickened the pace of experimentation, automated direct impression and response tracking, and compressed marketing program cycles.

To accomplish all that, marketers have employed a plethora of applications, from SEO tools to email tools to campaign management systems to creative suites to mobile messaging platforms. And make no mistake – automation has paid dividends and enabled more personalized marketing at scale. Companies that use marketing automation in the development and execution of their tactics have improved conversion rates, grown revenues, and become more efficient. In my career, I’ve seen piles of firms lift response and conversion rates up to 3x, and in some cases as high as 12x, doing so without exponentially increasing team sizes. And all along, technology-minded humans have remained instrumental components of both developing and executing those systems.

Will marketers wake up one day and find themselves obsolete?

Ironically, during this period of hyper-automation, the number of digital marketing jobs has grown, not shrunk. And the outlook is rosy, not bleak. In fact, the US Bureau of Labor statistics projects another 24k jobs will be added by 2026 in the US alone [iii]. In 1995, there was no such thing as an SEO analyst, a social media marketer, or a mobile marketing manager. Though some old jobs have disappeared, and some are at risk, there’s no indication that people gainfully employed in digital marketing today have anything to worry about.

If you’re a grizzled old marketing vet like me, you’ve already seen automation encroach on various tasks. For instance, writing email subject lines, deciding the best time to send that email, or even deciding how much to bid for digital ads – can all be automated. Yet simultaneously you’ve watched as new roles have emerged. If you’re a little younger, you’ll see even more amazing things in the future.

As certain aspects of marketing production become machine-executed, such as when to trigger a re-marketing treatment, new marketing opportunities emerge that require human intellect, such as determining if augmented reality provides marketing lift.

It’s a given AI will improve to where more complex tasks – and whole jobs – can be performed by machines, such as completely replacing telemarketers as suggested in this landmark Future of Employment study[iv]. This phenomenon is not new. It’s known as collective learning and has been affecting human advances since people invented language, writing, and tools. It’s one of the main reasons for societal and technological evolution as it fosters sharing and accumulating knowledge by allowing us to pass down knowledge and build on it via spoken and written communications. And it’s what’s enabled us to refine and evolve our machines.

But collective learning is not just about machines advancing. It’s the combination of humans evolving together – essentially becoming one system. As such, it’s not rational to assume AI in digital marketing will cause all human jobs to succumb to computerization. Expect instead machines to take over simpler roles just as they’ve taken over much of our manual labor in farming. They’ll ride shotgun with us, learning from us, correcting us, and mastering our simple and repetitive tasks – while we move on to create, invent, and fine-tune new, more complex ones. And along the way, we’ll be assisted by machines as we invent those new brand-new roles.

In my recent article on AI and technology trends that are affecting marketing (5 predictions for CRM’s AI applications in 2019), I discuss the trend toward effective combinations of crowds of human and machine intelligence. Organizations that build open systems that take advantage of this power and embrace it will outperform those that don’t.

To better understand why digital marketing won’t be fully autonomous anytime soon, it’s helpful to inspect the full journey of a marketing initiative, from idea inception to program execution. Marketers work in two main factories:

Creative factories where ideas are born, iterated, refined, vetted, simulated, and approved for use
Operational factories where select ideas are executed, monitored, tested, tracked, and retired

Creative Marketing Laboratories

Narrow (single purpose) AI systems have enjoyed some interesting achievements recently, such as painting masterpieces and writing books (this one, a riveting read on lithium-ion batteries). In marketing, AI has written image captions, tagged content, and even helped with content generation. But it’s far – way far – from putting all these pieces together and being the de-facto creator of new ideas. Humans own this space and will for some time. Responsibilities in jobs such as program strategists, copywriters, SEO experts, product marketers, and graphical/interactive designers have progressively increased. Consequently, these people do more and do it faster with automation’s assistance.

Operational Marketing Factories

Operation managers are inherently driven to streamline, rationalize, and automate processes. And they use industrial engineering and machines to do it. But just as an old-model car has limits on how much it can be modified, production marketing systems have the same inherent limitations. And there is no evidence that the cloud, low code, or agile is changing that. After about ten years, old systems need to be replaced or they reach a point of diminishing returns.

And radical change involves teams of people organized in large efforts to transition to new technology and redesigned processes. No machine accomplishes this without humans. Enterprises still employ hundreds if not thousands of humans to design and test their new marketing engines. Once in production, people feed the machines, monitor them, and fix them when they break. There’s plenty of work to go around. Expect digitization and human-machine process reengineering to accelerate, possibly compressing the marketing system replacement cycle from ten years to five years.

As this happens, expect marketing tasks like this to become more automated:

Customer data processing and summarization
Some feature engineering for model data prep
Pattern detection and event-based triggers automatically kicking off insight gleaning tasks and/or marketing treatments. Examples:
- Detecting customer intent, such as intent to purchase based on multiple signals
- Predicting and detecting key life events, such as a move, and providing proactive valuable offers
Certain marketing programs that run completely lights out that with automatic monitoring and tuning
Send-time optimization for all types of customer interactions
Automatic impression and response capture

What should marketers be doing today?

One lesson is don’t fear automation and AI in digital marketing, embrace it. Marketers and CX professionals must automate to remain competitive and ensure their survival. They must reduce the cost to acquire and deepen customer relationships, or standby while their competitors do it.

Ultimately, the savviest marketers build systems that entice customers to engage in less expensive and fully automated channels. By doing so, buying cycles are compressed, the cost to serve comes down, and customer satisfaction rises. But marketers must carefully monitor to see that satisfaction, revenue, and market share are improving. They must constantly ask, “Are these automated or semi-automated interactions delivering hyper-personalized, frictionless, and valuable experiences?”

So regardless of how you rate your marketing capabilities today, there’s plenty of room for improvement. Presently, most marketers still take about three months to create and execute new programs from inception to execution. And many interactions, whether via branch, store, call center, or email, are still inefficient and labor-intensive.

Marketers must strive for efficiency and excellence in creative and operational areas. Here are some things to consider in both:

The Modern Creative Laboratory

Consider these checklist items and benchmarks:

Form small teams of five to seven using modern collaboration tools like Slack
Incent teams to do research before brainstorming sessions. Don’t reinvent the wheel
Hold brainstorming sessions in public and private. Limit sessions to a few hours in public and a few days in private
Don’t mute wild ideas; encourage them. Don’t just end up with a better buggy whip
Limit testing wilder ideas to 10% of budget, time, effort
Ensure the process, ideas, and outputs are transparent and easily accessible by others

The Modern Marketing Operations Factory

Consider these checklist items and benchmarks:

Employ small teams of five to seven using modern agile and marketing operations tools that force item prioritization, templated based case management, workflow, task creation, and work sprints (for repeatable processes)
Integrate content management systems and processes with marketing automation. Fuse output to final step revision and deployment management systems and processes
Separate business-as-usual (BAU) changes from capability changes. For example, making minor adjustments to an existing promotion would be treated as BAU, and these kinds of changes packaged into revisions that are deployed daily. Capability changes that require more work, testing, and approvals can be deployed weekly or monthly
Automate and streamline as much work as possible, looking for ways to turn it into BAU work, via templates, re-use of assets, and rationalizing unnecessary steps and approvals
Look for as many places as possible to reduce or eliminate key entry
Once programs are in production, set up ways to automatically throw alerts when things go wrong or need to be adjusted. For example, if response rates are low trigger a notification that includes key statistics and diagnostic tips
Use an always-on marketing approach to most customer interactions. In this way, there becomes just one main program: The Next-Best-Action program. That system is simply fed new offers that come and go, but the program itself is evergreen – just on 24x7x365 -always arbitrating for the next-best-action when a customer comes in a channel, and the next-best-touch for outbound outreach.

Conclusion

Although “never say never” applies to nearly anything, it’s unlikely your marketing job will be taken by AI. But another human that knows how to use AI in digital marketing more effectively than you, well that’s another story.

And as modern marketers design the transition and march toward using technology for increased automation and AI in digital marketing, they must ensure consumers get what they want – relevant, rewarding, and timely value exchanges. If marketers relentlessly automate without seeking continuous feedback from customers on the resulting experiences, they’ll fail. Great brands will be those that can think creatively, design effectively, and execute flawlessly to deliver seamless experiences woven together by machines and humans. Using this approach, marketers and their marketing machines will stay gainfully employed.

[i] Rexer Analytics, http://www.rexeranalytics.com/data-science-survey.html, 2017

[ii] Websitepromoter, https://websitepromoter.co.uk/how-to-adapt-your-seo-for-ai, 2019

[iii] Bureau of Labor Statistics, https://www.bls.gov/ooh/management/advertising-promotions-and-marketing-managers.htm, 2019

[iv] Oxford Martin School, https://www.oxfordmartin.ox.ac.uk/publications/view/1314, 2013