Market scan guide · AI

How to run an AI market scan for a financial institution

An AI scan is unlike the others: the field changes monthly, every vendor claims it, and the regulator's questions are about governance before capability. The scan has to begin with the use case and the data it will touch, and hold every vendor, AI-native or not, to the same questions on where the model runs, what it learns from and how its answers are explained.

Figures from the Vederca registry as of 2026-10-24: 464 AI vendors, 188 products, 31 countries.

The AI landscape in numbers

AI software for financial institutions: the Vederca market map
The AI market map: every vendor's own mark, by named customers then name, drawn from the registry. Download the map (PNG, or as SVG).

The registry lists 464 AI vendors with 188 products, headquartered in 31 countries. 345 name at least one financial institution as a customer. 53 of the products mention an AI capability in their own text (28%).

AI vendors by headquarters country
United States: 244United States244Australia: 69Australia69United Kingdom: 49United Kingdom49India: 16India16Canada: 14Canada14Germany: 11Germany11France: 10France10Singapore: 8Singapore8Switzerland: 7Switzerland7Israel: 4Israel4
AI vendors by headquarters country: the figures as a table
ItemValue
United States244
Australia69
United Kingdom49
India16
Canada14
Germany11
France10
Singapore8
Switzerland7
Israel4
Capabilities most often mentioned by AI products
AI agents and assistants: 29, 0 confirmedAI agents and assistants29Grounding: 22, 0 confirmedGrounding22Generative AI: 18, 0 confirmedGenerative AI18Governance: 13, 0 confirmedGovernance13AI search and summarisation: 12, 0 confirmedAI search and summarisation12Fine tuning and custom models: 7, 0 confirmedFine tuning and custom models7Privacy: 6, 0 confirmedPrivacy6Conversational AI: 5, 0 confirmedConversational AI5Predictive models and scoring: 5, 0 confirmedPredictive models and scoring5Document intelligence: 2, 0 confirmedDocument intelligence2Human loop: 2, 0 confirmedHuman loop2
Capabilities most often mentioned by AI products: the figures as a table
ItemValue
AI agents and assistants 0 confirmed29
Grounding 0 confirmed22
Generative AI 0 confirmed18
Governance 0 confirmed13
AI search and summarisation 0 confirmed12
Fine tuning and custom models 0 confirmed7
Privacy 0 confirmed6
Conversational AI 0 confirmed5
Predictive models and scoring 0 confirmed5
Document intelligence 0 confirmed2
Human loop 0 confirmed2
AI vendors founded, by year
2017: 191920172018: 121220182019: 161620192020: 8820202021: 7720212022: 4420222023: 5520232024: 3320242025: 020252026: 02026
AI vendors founded, by year: the figures as a table
YearCount
201719
201812
201916
20208
20217
20224
20235
20243
20250
20260

The largest by named customers

VendorHQ countryNamed customers
AgiloftUnited States20
AKUVOUnited States20
AtlanSingapore20
BUSINESSNEXTIndia20
Clearwater AnalyticsUnited States20
ComarchPoland20
OneSpanUnited States20
PigmentFrance20

The newest entrants

VendorHQ countryFounded
AlcovaAustralia2024
KastleUnited States2024
MarlooAustralia2024
Bretton AIUnited States2023
Mistral AIFrance2023
Norm AiUnited States2023

The full list, with every vendor's flagship product and named customers: AI software for financial services.

How to evaluate AI vendors

  1. Start from the use case, not the vendorFraud scoring, document reading, a customer assistant, a staff copilot, a credit model: each has its own vendors and its own risk. One scan per use case, or one scan with sections.
  2. Separate mention from confirmationA page that says AI is a mention; a capability the vendor confirms in writing, or demonstrates on your data, is evidence. The Vederca registry shows the two apart.
  3. Ask where the model runs and what it learns fromHosting region, whether customer data trains anything, the foundation models used, and the contractual position on each.
  4. Demand explainability for decisions about peopleCredit, insurance and fraud decisions need an explanation the customer and the regulator can read; ask for one from the demonstration.
  5. Read the governance packModel validation, monitoring, bias testing, human review points and incident handling, mapped to your model risk policy.

The method for the scan as a whole, longlist to shortlist, is on the market scan page; the report template gives the headings.

The requirements that separate them

Must-haves in most scans of this category: the register's gates. Each needs evidence named beside it.

  • Confirmed capability on the use case, demonstrated on the institution's own data
  • No training on customer data without consent, stated in the contract
  • Hosting and processing in the institution's country, or disclosed
  • Explanations for decisions about customers
  • Model governance evidence: validation, monitoring, bias testing, human review
  • Citations and audit trails for generated answers

Go further