FIAB HighTech 2026 · San José

Compliance not as a brake on the business, but as the gate that enables it.

A team of AI agents that reads the KYC, verifies every person in the file and delivers two reports from the same data: one for Compliance and one for the commercial team. The decision still belongs to a person.

  • We run it on your own KYC form, using a test file.
  • Under a confidentiality and data processing agreement.
  • No agent writes to the master record. An officer decides.
Commercial Team · Snap Compliance
Agentic AI in Capital Markets category

Request a personalized demo

With your team, not on a stage. We will write within 48 hours to schedule the session.

Do you have your own KYC form for the test? *
What would you like to see?

Do not submit real customer data in this form. The proof of concept is arranged under a confidentiality and data processing agreement (Law 8968, Costa Rica, or its local equivalent).

250–450
hours to transcribe 850 KYC forms by hand
70
files held up by one field, not by risk
60,000
records verified in 20 minutes, in production
01 The problem

At every table there are two people. One says «sell». The other says «wait». And both are right.

For the commercial team the process has no visibility: the client has signed, the capital is ready, and no one can say when approval will come or where the file stopped. On the other side the uncertainty is the same: the compliance officer does not know whether the form was read in full or whether every related person was verified.

A commercial executive and a compliance officer sitting across from each other at a meeting table
Analyst transcribing ID documents from paper forms at night
250–450
hours

Transcribing a portfolio of 850 KYC forms by hand. One analyst, two or three months, copying ID numbers.

Client watching the clock in front of a stack of held-up files
70
files held up

In a single implementation, over one block — the consent — that was missing from the load. Not over risk. Over a field.

Screenshot of the screening engine showing a batch of processed records, with the subject's data blurred
60,000
records in 20 minutes

What the engine already does in production, at low cost. The question is who reads the form and who decides.

That is the gap we close.

02 The team

Three new members of the compliance team, dedicated to working KYCs.

The three agents of the due diligence team

Orchestrator Orchestrates.

Calls each member, waits and retries, watches the timings, routes errors and records every step in the log.

Reader Reads by vision.

Reads any KYC form with no prior template. It extracts the full graph of people — representatives, shareholders, ultimate beneficial owners — and flags what it could not read instead of inventing it.

Reviewer Consolidates.

Turns each person's checks into a single verdict for the file and drafts the reports. The file inherits the most severe signal; it never averages downward.

Anthropic models on Amazon Bedrock AgentCore. The agents never write to the master record.

The agent proposes. The person decides.

No agent closes an alert. No agent onboards a client. The only step that writes to the master record is executed by a compliance officer, by name.

03 The walkthrough

A fictitious file, from start to finish.

Comercializadora Andina del Pacífico, S.A. A Costa Rican brokerage firm, KYC-PJ-02 v4.2 form, nine related persons, two of them PEPs. Fictitious data for illustrative purposes.

Three KYC forms from different institutions lined up on a dark background
Step 1

The system has never seen this form.

No template, no field mapping. Nobody told it where the ID number is.

Step 2

That is why traditional extraction fails.

With line-by-line text reading, ID numbers land one row down and the PEP flag lands on the wrong person. In a KYC that is not a formatting error: it is a verification failure.

The form read as lines of text: ID numbers shift one row down
Text extraction
The same form read by vision: every field stays with its person
Vision reading
Traceability page of the risk report showing the detail of lists checked
Step 3

Every person is verified, not just the account holder.

UN, OFAC, European Union, United Kingdom, PEP lists, adverse media, court records and commercial signals in the press. What did not match is recorded too: before the regulator it counts just the same.

Step 4

One PEP pulls the whole file with it.

A sitting deputy minister holding 15 % through a BVI vehicle, and his sister a member of congress. Verdict: ESCALATE, high risk, committee approval before onboarding (FATF R.12). Plus two discrepancies in the form itself, reported, not corrected.

Report verdict page: high risk and committee approval
Verdict
Risk factors page showing the shareholding and the PEP status
Risk factors
Cover of the risk analysis report for Compliance
Report 1 · Risk
Cover of the suitability and opportunity report for the commercial team
Report 2 · Suitability
Step 5

Same file, same run, two reports.

Version A with commercial consent: a risk report and a suitability and opportunity report. Version B without consent: the risk report only. The commercial report exists solely if the risk report came back clean and the client consented. That order is the thesis.

Both reports belong to the sample file: fictitious people and data, for illustrative purposes.

Commercial signals in the press, from the suitability report
Step 6

Even the good news goes through Compliance.

Three positive signals in the press help prepare the commercial conversation. A fourth — a Colombian state programme touching the deputy minister — was escalated as a possible conflict of interest and is not used as a commercial argument until the unit rules on it.

Due Diligence module inbox showing the requirement list and completeness percentage, anonymized data

The commercial team can already see that their client is in, what state the file is in and what is missing. Nothing has been written to the master record yet.

04 Scale

One architecture. Five teams.

Something arrivesIt is verifiedSomeone decidesIt is documented
A file
Due Diligence
A transaction
AML Monitoring
An alert
Alert Handling
An incident
Operational Risk
A date
Board Reporting

The same team. What changes is what they read.

Five systems already in operation, each with its own users, inside the Snap Compliance Compliance OS.

Request a personalized demo for your organization.

We run it on your own KYC form, using a test file and under a data protection agreement.

Go to the form