The state of financial AI distribution today

Last updated: 24 August 2026

There are two kinds of finance app on ChatGPT, and only one of them matters for distribution.

The first kind gives you data. Market feeds, dashboards, accounting queries, credit analytics, FinOps tooling. Useful, but you cannot buy anything through them.

The second kind sells. You can get a quote, start an application, compare real offers, or reach the person who completes the sale. This is distribution, and it is the part of the app store that changes how financial products reach customers.

We pulled the public registry of finance apps live on ChatGPT, sorted every one into those two buckets, and then did something we have not seen anyone else do: we read the full list of tools each distribution app publishes, so we could say what these apps can actually do rather than how many there are.

147 of them are distribution apps. 83 are insurance.

Insurance is over half the field

Insurance is more than mortgage, lending, banking, tax and investing put together. The full split by business model:

Note that insurance comparison marketplaces sit in the comparison row by business model but count as insurance by product. Counting insurance by business model alone would undercount it by 17.

The category is notably international and spans motor, home, pet, travel, life, bicycle, funeral, taxi, personal-guarantee and commercial cover. That long tail of specific products is the telling signal. This is not two big players testing a channel. It is a broad set of insurers, brokers and comparison sites all arriving at the same conclusion at roughly the same time.

Three apps to 147 in nine months

The field has grown steadily rather than in a rush. Three apps were live in December, 21 by the end of February, 48 by March, 88 by May, and 147 today. The busiest months were March (+27) and July (+28), but nothing in the series reads as a step change.

That is worth stating plainly, because an earlier version of this piece claimed the opposite. We had dated each app by when the public registry first saw it, and on that basis April showed 5 arrivals and May 34 — an apparent inflection. It was not one. The registry expanded its own crawl coverage in May, and the jump measured the observer, not the market.

Every app here is now dated by the earlier of two signals: the date the provider published the listing, and the date the registry first saw it. Neither is clean on its own. The provider date moves whenever a listing is updated, which is why 41 of the 147 carry a provider date later than the day they were first observed live: Aviva’s provider date is 3 September against a first sighting of 2 April. The registry date lags a quiet launch. Taking the earlier of the two gives a defensible “live no later than” bound, and under it the curve is a ramp, not a hockey stick.

What has not changed is the mix. Insurance is 83 of the 147 and has led cumulatively all year, but it has never been the whole story: mortgage, banking, lending and tax have arrived alongside it throughout.

Most of the “insurers” are not insurers

Zoom into the 83 insurance apps and split them by who actually holds the risk.

Only 17 of the 83 underwrite what they sell, and 14 of those 17 are established insurers. Meanwhile, of the 66 apps that do not carry the risk, 50 are challenger brands.

So the value chain has split cleanly along the channel. Legacy money is turning up as carriers. New money is turning up as distribution. Two different bets on the same surface, and you can see which side each company thinks it can win.

One caveat: a company’s licence status is not always in an app’s description, so treat the carrier count as a floor. The shape holds either way.

Almost nobody prepares the next step

Here is the finding that matters most if you are thinking about building one.

Buying is not available inside ChatGPT. No app in the set publishes a tool that takes payment or binds cover, and that is a platform fact rather than a failing of these companies. So the useful question is not whether an app can complete a sale. It is how well it prepares the step that follows.

By that measure, 69 of the 83 insurance apps return a price and stop. Not all of that is a gap: taking an application inside a chat means collecting personal data in a channel most compliance teams have not signed off on, so stopping at the price is a defensible decision as often as it is an unfinished one. What the tools do show is how few have built the step after it, whatever the reason.

Fourteen do more. “More” means one of three specific things, and it is worth being concrete about each.

A prefilled purchase link (8). The quote comes back with a way to finish buying it.

• Nomad Insurance captures subscriber details and returns a finalisation URL to pay
• Sundays returns the premiums plus a link to finish the purchase
• Fidamy opens a hosted checkout session for the chosen plan
• PANDA Tierversicherung creates a link to buy the selected package
• Suri decodes your plate, prices it, then creates a link to complete the order
• UkraineBorder returns a prefilled checkout link
• Simplis returns a pre-filled subscription link for the RC Pro cover
• Allianz Direct España returns a direct contracting URL with the quote

A submitted application (1). One app, and it is the only one.

• Insurify records data-sharing and application consent, submits the application to carriers, and returns their actual prices rather than an estimate

A handoff to a named adviser (5). No purchase, but a person who can finish it.

• InsurConnect hands a consented lead to a named licensed agent from its Texas directory
• Allianz forwards validated form data as a lead to an Allianz agent
• MAIF requests a callback from a MAIF adviser
• JeRenegocie records a consented callback request after the estimate
• Trucordia creates a handoff link to an adviser for personalised options

The other 69 return a number, and the conversation ends there.

To be precise, because it matters: many of these quotes are real. Nine apps submit a genuine request to insurers and return live carrier prices rather than an estimate. Seguro Match submits to multiple insurers through Codeoscopic. GoCompare returns real provider results. The gap is not the quote. It is everything after it.

Precise on price, silent on the product

The tool lists reveal a lopsided investment.

On pricing, 17 of the 83 insurance apps look the customer’s risk up in real data rather than accepting what was typed: a licence plate resolved to a vehicle, a postcode or cadastral reference resolved to a property, a company number resolved in Companies House, an address geocoded and scored for natural risk. That is genuine underwriting infrastructure exposed through a chat interface.

On the product itself, almost nothing. Only 10 publish a tool that can surface the actual policy wording. 34 can describe cover at a summary level. 39 expose no tool about cover at all.

So an app is far more likely to price you accurately than to tell you what you would be buying. That is worth sitting with, because the price is the easy part to automate and the cover is the part a customer actually needs explained.

One nuance in fairness to the apps: for some, the silence is deliberate. Neptune Flood’s tool instructions explicitly forbid it from interpreting policy language or deciding whether a specific loss would be covered, and require it to offer a licensed agent instead. That is a compliance guardrail, not neglect, and some of the 39 are likely the same.

It is not a US story

The single largest market is the US at 53 of the 147 apps, and nothing else is close. But add Europe up and it comes to 66 across 15 countries.

In insurance specifically it is 46 in Europe against 25 in the US. 28 countries have at least one distribution app, and four more apps operate across borders. A market map that starts and ends in the US is looking at roughly a third of the field.

One limit worth naming: the registry records a country per app, not a region within it, so a state-level view of the US is not available from this data. Of the US apps, almost none name the states they serve.

What the landscape actually tells you

Put the picture together.

Financial distribution on ChatGPT is real, it is already mostly insurance, most of those “insurers” do not carry their own risk, Europe is a bigger block than the US, it inflected in May and has not slowed, and it is frictionless at the top (84% require no login) but almost entirely undeveloped after the price.

The market has crossed the first threshold. Being live is no longer the differentiator, because 147 apps are live and more arrive every week.

The second threshold is depth, and the field is wide open there. Everything past the price is barely built: how well the handoff works, whether the app holds a real conversation, whether it stays compliant while it sells. Nor is anyone measuring which conversations turn into customers.

That gap between “we shipped a quote” and “we run the funnel” is where the next year of this market gets decided.

This is the problem WaniWani works on: the distribution app itself, plus the analytics to measure the funnel from discovery to lead, plus compliance monitoring as the rules catch up. We have been testing these apps one by one and scoring them on product depth, compliance and conversation quality, and we will publish those scores shortly.

One deliberate scope limit: this is ChatGPT only. The same registry covers Claude (208 finance apps) and Codex (19). Claude has a single app in an insurance category and it is claims operations, not distribution; Codex has none. Insurance distribution inside AI assistants is, for now, a ChatGPT story.