FI Works' Keith Henkel on turning bank data into action

FI Works founder Keith Henkel says community banks and credit unions have the data, but lack a way to turn a warning sign into a call to the right person.

Keith Henkel, founder and chief executive of FI Works, in a portrait photograph

FI Works, which builds a relationship intelligence and engagement platform for community banks and credit unions, has launched Signals, a feature that uses machine learning to spot households likely to add a product, take up digital services or leave the institution.

Keith Henkel, the company's founder and chief executive, has spent more than three decades building technology for financial institutions. He said the tools banks already buy tend to stop at the list. “Most analytics tools give a community bank or credit union a list but no next step,” he said. Institutions that bought them told FI Works the results felt like a “so what”, because they pointed to a handful of opportunities or to ones already lost.

Why transaction data is not enough

Part of the problem, in Henkel's view, is what those tools look at. Transaction data shows what happens inside a single account and little else. If a customer starts paying a mortgage to another lender, that matters, but it leaves out who else lives in the household, which products the institution already provides, what the relationship earns once funding and costs are allocated, and which banker manages it.

“By the time that payment shows up in the data, the customer or member has usually already decided to use a competitor,” he said.

Transaction-only approaches also narrow the field. They rely on primary checking activity, so a relationship that holds only a loan, a certificate or a business account may produce nothing, and they depend on the text in each transaction description. Signals instead combines FI Works' existing models of what a household is likely to need next with product ownership, household links, digital engagement and transaction insight, so a team sees opportunities that no single source would show.

From signal to task in one place

Because Signals runs inside the same platform as the company's marketing automation and CRM, Henkel said an opportunity never has to leave the system to become work. A team can target its best opportunities with one click, or ask a plain question such as which customers offer the best chance of deposit growth. From there a marketer can build an audience and launch a campaign, a manager can create an opportunity, or a branch employee can receive a task that already says whom to contact and why.

“Usually, there are manual steps between insight and action, like exporting a list, cleaning it, and loading it into an email tool and then a CRM,” he said. “Those steps go away.”

A household with a strong likelihood of taking a home equity loan might receive relevant outreach and a follow-up call from a banker, while an attrition signal can prompt a conversation about what is worrying the customer. Reporting then ties results back to new accounts and balances.

Reading what is missing

The earliest warnings, Henkel said, tend to come from what is absent rather than from activity elsewhere: an account opened without online banking ever being set up, a household that still holds one product a year after joining, a certificate nearing maturity with no contact from the institution, or a complaint left unresolved.

Context decides what a signal means. A stopped direct deposit could point to a real problem for a long-standing household with a mortgage, a home equity line and three deposit accounts, but only reflect a new job for someone who opened a single account a few weeks ago. “Telling the two apart takes product, service, channel, and household information together, with each relationship compared against its own past behaviour,” he said. The platform generates new signals every day as that behaviour changes.

Each opportunity carries two numbers: a score for how likely a household is to act, and a predicted balance and profitability for what the relationship could be worth. Henkel argued that likelihood alone sends effort to the wrong places, since two equally likely households can be worth very different amounts. Profitability is calculated with funds transfer pricing and allocated costs, and every figure can be traced to its source. “A team that does not have a data scientist can use the ranking with confidence because nobody has to accept a black-box number,” he said.

Where the platforms go next

What has surprised him most is how consistent the reaction has been from institutions that had bought signal tools before. “Institutions are not short on data or on ideas; they are short on a way to turn what they know into a conversation with the right person at the right time,” he said.

He expects three shifts. AI will matter most when it works on an institution's own clean, well-governed data: a large language model can write a decent email, he said, but it cannot say which households are likely to need a home equity line in the coming months without that foundation. Marketing will move from calendar campaigns to always-on programmes for onboarding, cross-selling, maturities and attrition. And CRM will do more of the work for the user, so a frontline employee opens a record that already says whom to call and why.

“Community financial institutions are running leaner than ever and they need a platform that does the whole sequence for them,” he said.