Short answer
Of 25 AI vendors selling to US banks and credit unions, two publish pricing a buyer could budget against: Microsoft 365 Copilot at $18 to $30 per user per month plus a required base licence, and UiPath with a $25 per month entry tier that excludes document extraction at scale. Everything else is quote-only.
Pricing transparency is one of the five criteria this site ranks on, and it is the one nearly every vendor fails. That is worth stating as a finding rather than a complaint, because it has a practical consequence: at a community institution, the cost of finding out what these products cost is measured in staff hours, and that cost lands before you know whether the product is affordable.
What is actually published
Across the vendors covered on this site, published pricing breaks down into three groups, and the largest by a distance is the third.
The two published numbers are also worth reading carefully. The Microsoft add-on price excludes a qualifying base licence that is required on top, so the all-in per-user figure is higher than the sticker. The UiPath entry tier explicitly does not include document classification and extraction at scale, which is the capability a lender would be buying it for.
| Group | What you get | Examples |
|---|---|---|
| Real published pricing | A per-seat or per-tier number you can budget against | Microsoft 365 Copilot, UiPath entry tier |
| Published model, no numbers | A stated pricing structure with no dollar figures | Gradient Labs outcomes-based pricing |
| Quote only | Nothing published; every number requires a sales cycle | Everything else covered here |
The pricing models behind the quotes
Even without numbers, the shape of a vendor's pricing tells you how the cost will behave over time, and that matters more than the first-year figure.
Asset-based pricing is designed to grow with your balance sheet. One lending platform disclosed moving from seat-based to asset-based pricing in fiscal 2025, which means the subscription is intended to scale with portfolio growth and AI consumption. That is defensible from the vendor's side and it means your third-year cost is not your first-year cost.
Volume-based pricing dominates document processing, and it is where community institutions most often get an unpleasant number. Products built for lenders processing tens of thousands of files a month carry minimum commitments that do not scale down gracefully to a few hundred.
Outcomes-based pricing, where you pay for resolutions the software actually delivered, appears once in this market. It shifts deployment risk toward the vendor, which is unusual enough to be worth asking every conversational AI vendor whether they will consider it.
The costs that are not on the quote
The subscription is rarely the whole number, and the additions are predictable enough to budget for.
Integration is the largest. A product delivered through your core provider's marketplace carries its integration as part of the arrangement; a custom API integration is a project you pay for once and maintain forever. Implementation and content curation come next, and for conversational AI the content work on your side usually sets the timeline rather than vendor engineering.
Then there is governance. Documenting, validating and reviewing a model is recurring internal cost that never appears on a vendor quote and never goes away.
- Integration build, and maintenance across core and LOS upgrades
- Implementation services, often quoted separately from the licence
- Content curation and knowledge base preparation
- Model documentation, validation and annual review
- Required underlying licences, which is where the Microsoft price hides its real total
How to get a number faster
Ask for a written not-to-exceed figure in the first or second conversation, framed on your asset size, user count and volume. Vendors who intend to sell to institutions your size can produce a range quickly. Vendors who cannot are usually telling you that you are below their normal deal size, which is useful to learn in week two.
Two other tactics work. Ask what a comparable named institution pays, in a range rather than a figure, since most vendors will describe a band even when they will not quote one. And ask for the floor: the minimum annual commitment. In volume-priced categories the floor, not the unit price, is what decides whether the product is affordable at your scale.
Frequently asked questions
Why do so few AI banking vendors publish pricing?
Most price on institution size, volume or configuration, so a single published number would be wrong for nearly every buyer. That reasoning is genuine and it still leaves a community institution unable to budget without spending staff hours, which is why this site treats pricing transparency as a ranking criterion.
What is the cheapest way for a bank to start with AI?
General productivity software inside a tenant you already run, which is the only category here with a published per-seat price and no new third-party risk review. It does nothing banking-specific, which is the trade-off.
Should we expect implementation fees on top?
In most categories, yes. Origination platforms, financial crime systems and document processing deployments typically quote implementation separately, and conversational AI often carries content preparation work that lands on your staff instead of the invoice.
How should we compare quotes that are structured differently?
Convert everything to a three-year total including integration, implementation and the internal hours you expect to spend on governance and content. Per-seat, per-asset and per-volume quotes are not comparable in year one and can invert by year three.