Photo illustration: Tiffany Herring / Axios; Photo: Portgage
Adam Felesky, CEO of fintech-focused VC firm Portage, says his AI investment strategy starts with highly regulated financial services businesses that have high switching costs.
- This interview has been edited for brevity and clarity.
Axios Pro: What makes an AI-enabled fintech defensible?
- Felesky: "We look for platforms that are the source of truth or a ledger, deeply connected into the core of an institution or an insurance business. That could be a core banking system or a policy-administration platform that is part of quoting and retaining all the risk of the policies issued.
- That has more of a moat. Financial services is a little more protected from generic AI risk because of regulation and stickiness."
What does AI change for those businesses?
- "We think the long-term operating margins of those solutions or platforms will be superior to anything we have seen previously.
- We believe a neobank will be the most efficient bank that we have ever seen. The ROEs, operating margins and capital-efficiency ratios that emerge as these things scale could make them the most profitable institutions in their category."
Where are you less convinced?
- "Some of the first AI solutions that accumulated a lot of revenue quickly were in wealth, proposal software and note-taking for advisors.
- "We see the enablement to the advisor. We think it is an amazing tool and super useful, but we do not see the long-term moat because those products are not tied into the core source of truth."
How do you assess more established companies that were not built AI-native?
- "On the direct-to-consumer side, scaled platforms that are now embracing AI are seeing the best unit economics we have ever seen.
- That is not every company. There are many that do not have the DNA to continue to keep up with the pace of change, and that is going to be problematic."
How do you separate durable AI revenue from just a proof of concept?
- "Some of this revenue is very POC-oriented, and we have not gone through a renewal. We would prefer to wait to see how that plays out and pay a higher price for those that show it is a sticky product, where people are not only renewing but NRR is increasing."
In an AI world, how important is it for product deployment to be fully repeatable, rather than being led by forward-deployed engineers?
- "The ability to move from a core system should no longer be measured in two years, which is what it used to be in the past. Agentic migration tools help you get through the core rip-and-replace in a much more efficient manner and at way less cost to the end customer.
- I honestly do not think agentic migration tools are a proprietary wedge or moat. I think everyone can build those really easily."
What pricing model will financial-services buyers accept for AI?
- "I do not think we know the answer yet. Procurement and CFOs want to have a budget and want to know that they are not surpassing that.
- Usage-based pricing could work if it is very clearly tied to a positive outcome for the end client. If they are making more revenue and there is positive net margin, they should be OK paying more because they are earning more."
