Oct 29, 2021

Extracting Customer Knowledge with SupportLogic

SupportLogic screenshot
SupportLogic sentiment scoring screen. SUPPORTLOGIC

I have long been interested in how companies can capture, structure, and apply the knowledge and insights embedded in customer support transactions. Watching the progress in this area has been a long, strange trip. In the mid-1990s, when knowledge management was all the rage, I was on the board of a company named Inference—one of the early companies in the 1980s “AI Spring” era—that attempted to structure support knowledge with a technology called “case-based reasoning.” One of the company’s executives, Phil Klahr, and I wrote an article about the topic in an article called “Managing Customer Support Knowledge.” But case-based reasoning proved to be challenging as a method to capture and dispense tech support knowledge, and the company was bought by eGain. 

More recently, I wrote about a Boston-based company called Talla, which also attempts to capture and structure customer support knowledge. My sense is that it works well, but it is solely focused on answering customer questions with accurate knowledge. I also wrote about a system developed internally at TD Ameritrade to capture the content and insights from customer support calls. It was impressive, but developing such a system internally would tax the capabilities of most companies.

So I was pleased to learn recently about a company called SupportLogic that is attempting not just to deflect support requests from customers, but to capture important knowledge about customers and their support issues. The company uses deep learning models that it built on top of its own technology, and along the way it filed 11 different patents. 

Krishna Raj Raja, the CEO of SupportLogic, worked in customer support at VMware for a few years. He observed there that the much-discussed “voice of the customer” is typically buried in support conversations. Ticket-oriented CRM systems, he noticed, don’t capture the essence of the support conversation. He also concluded that the great majority of tech companies were interested in deflecting support interactions to self-service channels rather than learning from them. The knowledge derived from support could help not only to improve customer satisfaction and retention, but could also be used to make products and services better. 

Raja founded SupportLogic to address these opportunities in 2016, but that was still a bit early for the natural language understanding (NLU) capabilities the company needed. He and his first engineer experimented with various technology providers’ tools, but none of them worked on the training dataset (a publicly-available tech bug database) they were using. They were trying to identify conversational sentiment attributes in support text like customer frustration, confusion, and requests for product features—what Raja calls “signal extraction”—and it wasn’t possible with existing technologies. But deep learning algorithms for NLU matured quickly, and SupportLogic built its own technology stack on top of a complex ensemble model. NLU technology is evolving rapidly, and SupportLogic is currently evaluating BERT, GPT-3, and other newer model types.

What SupportLogic wants to capture is the overall context of the text-based (emails, case notes, voice-to-text) support experience. It analyzes multiple customer interaction points at different stages of the support journey. There is a scoring approach for both the individual support case and the overall account. The scores are excellent predictors of customer attrition, lifetime value, and customer escalation of support cases. They can reduce escalations by up to 50% if the company using the SupportLogic system responds quickly to low scores. Because the machine learning models are regularly trained on each customer’s support texts, every customer effectively has their own version of SupportLogic’s software.

SupportLogic at

Imply is a user of SupportLogic. The company provides service, support, and value-added features for Apache Druid, an open-source real time analytics database. Kevin Hodgkins, the VP of Customer Success at Imply, discovered SupportLogic at Fivetran, his previous employer. The customer success team was trying to read through support tickets to learn why customers were escalating cases, and they couldn’t keep up with the volume of them without assistance from AI. Hodgkins discovered SupportLogic on the Internet, and used it at Fivetran for a year. It helped them reduce customer churn by 25%.

Now, at Imply, Hodgkins has less of a support volume issue compared to his previous job; his concern is the complexity of Imply’s support cases. They often involve complex technical discussions, and Imply is needed to monitor customer sentiment while the ticket is open. If a customer is having a problem with Druid, it’s typically part of their operating processes, and it needs to be resolved quickly. SupportLogic can monitor customer sentiment while the ticket is open, let everyone know how the customer is feeling, and address problems as soon as possible. Using SupportLogic helps Imply predict and avoid escalations easily. Hodgkins notes, “It’s much easier to manage negative sentiment, tension, and customer dissatisfaction while the case is active. Operational support intelligence lets us change how something is being handled today, which improves customer satisfaction. And that leads to renewals and upsells.”

One potential benefit of the support intelligence from SupportLogic is the ability to pass customer feedback along to product management and engineering. Imply does provide “voice of the customer” feedback to that audience, but ensures they have the right business context and understanding first. Specific customer issues are only passed to engineering when required; there is no escalation just because a customer requests it. Hodgkins said, “We have a well-defined process for moving customer issues to Engineering.”

Imply was only founded in 2015, but it is growing rapidly. Kevin Hodgkins believes the company is right at the point where SupportLogic could head off escalations and provide much greater context for support engineers in solving customer problems. He knows what’s coming, and he knows that SupportLogic will be just the ticket for it.