Recently I read a fascinating article on how crime investigators are using machine learning to find patterns and uncover insights that otherwise wouldn’t be detectable. Investigations ranged from criminal to cybersecurity, competitive counterintelligence, and corporate litigation. In one example, a firm showcased how – if this technology had been available during the Enron era – they might have proactively detected the scandal, potentially averting the crisis. Their demo traversed 500,000 Enron emails, and as it did it learned how to flag suspicious ones.
It’s not news that pattern detection software works for fraud detection. For years, banks have employed systems that scan structured credit card transaction data, and flag unusual activity. Yet what occurred to me is this same approach and its specific techniques are extremely well suited to the art and science of developing customer intelligence against unstructured big data. Customer experience pros could then unlock new mysteries, and take appropriate actions leading to great customer success outcomes.
The first thing criminal investigators do is they gather all the facts they can, from any available source – Emails, phone records, texts, web activity. The adage being “Leave no stone unturned.” It’s never clear at the start of any investigation which clues might matter, and may link to others – so all are important regardless of their form. As the investigation unfolds, machine learning techniques, such as neural networks that use self-organizing cluster maps (known as SOMs) can help find patterns, and eventually help the investigators form a hypothesis. Available evidence is used to test whether the facts fit the theory.
Shifting the frame of reference, gaining customer intelligence and using it to solve for marketing and customer experience challenges can benefit from this same methodology and technology:
- Consumers leave clues about their preferences and behavior in many places; sometimes in unstructured forums, like social media, product reviews, and blogs.
- It’s virtually impossible to sift through this data without the aid of technology and automation.
- Machine learning can be used to find patterns in customer activity, such as what product they are most interested in buying, or that their sentiment is trending toward total dissatisfaction.
- Once patterns are detected, predictions can be made and actions triggered in efforts to anticipate needs or alleviate matters.
As a consumer, my natural reaction might be to say, “That’s creepy and spooky”. Ironically though, most firms simply want to use this to improve your experience with their brands since they know it’s critical to their health. Repeatedly, surveys show above price and product, people leave because of dissatisfaction with the way they are treated.
But the level of dissatisfaction is qualitative and differs by customer. One customer who experiences a single network issue may become enraged, while another may be more tolerant. Knowing this and the value of each customer helps the company treat each situation with a custom tailored response.
That all sounds like common sense and easy, right? Try doing it on millions of customers, with billions of bytes of unstructured data in their direct conversations and behaviors, and their indirect musings on social media, in blogs, and elsewhere. Moreover, try to learn when each customer reaches various stages of interest or displeasure, and overtime improve your ability to predict these and take timely action.
Since the dawn of time, we learned that to survive we needed help from machines. Use this newest breed of machines along with time tested investigation techniques to crack the enigma of your customers, gauge their state of mind, and delight them with personalized experiences.
Note: These views are my own, and not that of my employer