- Key insight: Goldman Sachs led a $110 million Series C in Taktile, betting AI is ready to make banks' and insurers' high-stakes, regulated decisions.
- Supporting data: Financial institutions spend an average of $72.9 million a year on KYC and AML operations, according to Fenergo, which found 82% already use AI in that work.
- Forward look: The open question is whether 2026 is the year banks trust AI with decisions that, when wrong, cost millions and summon regulators.
Overview bullets generated by AI with editorial review.
Goldman Sachs recently led a $110 million investment in a startup that wants banks and insurers to let artificial intelligence make their highest-stakes decisions.
The money went to Taktile. The company's software takes decisions such as approving loans, triaging money-laundering alerts, paying insurance claims and onboarding customers and delegates them to AI agents (programs that carry out multistep tasks).
Taktile
The startup, founded in 2020 by machine-learning engineers Maik Taro Wehmeyer and Maximilian Eber, has now raised $184 million.
Taktile did not immediately respond to a request for comment.
The company has not disclosed its latest valuation, annual recurring revenue or total number of customers. More than 200 people worked at Taktile as of late May, according to Tracxn, a firm that tracks private companies.
Growth Equity at Goldman Sachs Alternatives (part of the firm's asset-management business) is backing an idea banks have been slow to accept: that they can trust AI to make regulated decisions.
If the wager works, it could influence how banks staff their underwriting, fraud and compliance desks. If it fails, it risks Taktile customers drawing scrutiny from regulators (whether state or federal) over fair lending and model risk.
Off-the-shelf AI tools work for simple tasks but fall short of "operating mission-critical financial decisions where errors can cost millions," said Wehmeyer, Taktile's CEO, in the announcement about the latest funding round.
That harder job is the one he says Taktile is built to handle.
AI agents for making regulated decisions
Taktile sells what it calls its "Agentic Decision Platform," a modular system that combines AI agents, hard-coded rules, business context and human oversight.
The pitch is that the people who own a decision, such as a head of credit or a fraud officer, can themselves build and control an automated workflow to make those decisions, no coding required.
The company says it already powers millions of decisions a day for customers including business-banking startup Mercury, U.K. digital bank Monzo, wholesale marketplace Faire and spend-management firm Pleo.
Its $54 million
These named customers skew toward fintechs and digital-first banks rather than the large, traditional institutions (such as Goldman Sachs itself) Taktile is also pitching.
Allianz, a conventional insurer, is the exception. Neither Taktile nor Goldman has said publicly whether Goldman uses the software itself.
Taktile claims customers reach 95% automation in business-to-business underwriting and see 75% fewer false alarms from anti-money-laundering systems.
One large insurer it did not name projects more than $90 million in claims-processing savings, according to Taktile, which did not say over what period.
American Banker could not independently verify these results.
Bankers should be skeptical of vendor figures like those, said Trace Fooshee, a strategic advisor on the fraud and anti-money-laundering practice at research firm Datos Insights.
The numbers are not necessarily false, but "there is an agenda at work," he said. A vendor can claim to automate 95% of a task without specifying which cases that covers or where the automation stops.
The stakes for banks
The work Taktile automates is both expensive and heavily policed.
Financial institutions spend an average of $72.9 million a year on know-your-customer and anti-money-laundering operations, according to a 2025
Fenergo found that 82% of institutions now use advanced AI tools somewhere in that work, up from 42% a year earlier.
But handing credit, fraud and compliance calls to software invites
Those concerns are why, for now, AI is "still better suited to assisting human operators than replacing them" on end-to-end work like loan origination, claims adjudication and alert triage, Fooshee said.
Banks will keep a human in the loop, he said, until the technology can satisfy "the regulators' need for model transparency and 100% consistency of outcomes."
Until then, compliance officers will hesitate to let AI execute decisions that draw regulatory scrutiny, he said.
Part of the problem is that large language models are "exceptionally more difficult to explain" than the simpler machine-learning models banks already use, according to Fooshee. That complicates the job of demonstrating a model's transparency and consistency to bank examiners.
No bank is yet automating a regulated process from end to end, according to Fooshee; they are experimenting with discrete pieces of it.
In one common setup, AI drafts a recommendation on a fraud or money-laundering alert, but a human still makes the call, and only after having "checked the AI's homework," he said.
Goldman's investors framed Taktile's advantage as understanding "how regulated financial institutions actually operate," Christian Resch, a partner at Goldman Sachs Alternatives, said in the announcement.
