Talk to Paul

NXTSTAGE FinTech Pilot Competition 2026 participant

For Kansas community banks and the partners who serve them

Bring AI into your bank without putting your next exam at risk.

Show your examiner every decision, on the core you already run.

Your core keeps running. Every change is approved, and every decision is on the record.

Keep what works, build what’s next, retire on your terms. Prove it all with evidence built in, not bolted on.

AI drafts. Your people approve. A bad change comes back exactly, data included, or it is refused with the reason.

A pilot to improve operations and offer new digital services on the core you already run.

See the pilot ask Pilots we want ↓

Early stage. We are looking for our first community-bank design partners.

Pilots we want

Four bank problems we want to pilot with you.

Your core and your vendors keep running. Tallyseal adds a layer of rules, approvals and record over them, so the work your examiners review can be shown from one record.

Built for the exams you already sit: FDIC, Federal Reserve, OCC and the Kansas Office of the State Bank Commissioner (OSBC).

BSA/AML is partly built; the other three are being designed with partners.

BSA/AML alert and SAR decisions (lead)

AI summarises an alert and drafts a disposition, an analyst proposes, a second named person approves, and the alert’s history becomes evidence for the examiner. Show the examiner why each SAR was filed, or wasn’t. We’re building a rule library where each alert decision keeps its rule, reviewer and reason on the record, ready before the examiner asks. Supports FinCEN SAR and CTR duties and follows the FFIEC BSA/AML Examination Manual.

Partial

Call Report assurance

Call Report (FFIEC 031, 041, 051) figures traced back to source, with adjustments approved before filing.

Designing with partners

Section 1071 lineage

Small-business lending fields traced to their source documents ahead of the 2028 compliance date (dates may move).

Designing with partners

Change records over legacy cores

Routine core updates that today rely on manual re-keying, checked against business rules and recorded so they can be reversed.

Designing with partners

What’s ready to pilot today

We're building BSA/AML, and its first parts already work; the other three are being designed with partners. What is in place today:

Bring in AI while the core runs. Application errors roll back changes for humans and AIs alike.

Today

Approval is yours to set: four-eyes sign-off on the work you choose, in your order, and the maker never signs as checker.

Today

Your record, not a sample: each field change records old value, new value, who and why, and exports as a signed evidence bundle for auditors and regulators.

Today

Pilot ask

We are looking for one or two Kansas or US community banks, or the integrators who serve them.

No bank match yet. We would like to design the first pilot with you.

What we need from you

  • One problem from the list above, such as an alert backlog or a Call Report tie-out.
  • A sponsor, such as a BSA officer, CIO or operations lead, and time with the people who do the work.
  • Sample or test data, and your view of what your examiners would want to see.

What you get

  • A pilot shaped around your core and vendors, with nothing ripped out.
  • Your people approve each change; the record shows who did what, and when.
  • Direct work with the founder, and a say in what we build next.

Pilot scope and terms agreed with each bank.

Kansas community banks

The compliance pressures we hear about, and where Tallyseal fits.

Lean teams, legacy cores and examiners who want to see how decisions were made. Each item below is something we are designing with partners.

AI oversight and model risk

Show examiners and your board how AI-assisted work was checked: a person approves, and the record shows which rule applied. Supports SR 26-2 model risk guidance.

Designing with partners

BSA/AML monitoring and SAR workflows

Clear alert backlogs without overloading staff: AI drafts, an analyst proposes, a second person approves. Supports FinCEN SAR duties.

Designing with partners

Manual workarounds over legacy cores

Replace copy-paste routines and ad-hoc scripts with changes checked against your business rules, recorded and reversible.

Designing with partners

Vendor and third-party risk

Keep your core vendor and add a clear record of what each tool changed, ready for third-party risk reviews. Supports the interagency third-party guidance.

Designing with partners

Section 1071 and CRA data

Lending and community data traced to source, so reporting holds up as rules and thresholds shift.

Designing with partners

Instant payments fraud controls

Decisions on FedNow and RTP fraud alerts approved by a person and shown from one record.

Designing with partners

Examples, not a full list. Bring the problem your team spends the most time on.

The problem

What we hear from bank teams.

Our core is decades old and hard to reach.

Batch jobs and vendor screens, with few people left who know why it works that way.

Examiners flag our workarounds.

Copy-paste routines and custom scripts fill the gaps, and they show up as operational risk.

We want AI, but not the risk.

A lean compliance team cannot stand behind an AI black box at exam time.

How it works

Connect. Build and modernise. Run. Prove. With AI at each step.

Not an IDE: an AI development environment. Design it with your experts, build it with AI, and your people approve each change before it goes live.

01Connect

Use what you already have.

Connect by web calls, files, queues, databases or existing code: your team or SI builds each connection on Tallyseal; each result returns through one door.

  • Over the web: rules and processes call the web services your existing apps already have, and each call is on the record.
  • Files and old formats: read spreadsheets, files and legacy record layouts as they are, and turn them into a working app.
  • Code and databases: Your team or SI can easily build calls to routines in your systems and connections to your other enterprise databases on Tallyseal.

What it meansNothing has to be rewritten to start.

02Build and modernise

Ship new products, and shrink the old estate as you go.

Wrap, re-use and extend the apps you already run. Your team or SI checks each new part against the old and switches the old off step by step; we’re building the final switch-off check.

  • Build brand new applications with AI.
  • Generated, not hand-written: Tallyseal generates the tables, screens and the API description for changes from your approved definitions, and regenerates them when a rule changes.
  • Re-use: your experts settle the rules buried in existing apps once, and each answer stays linked to the source lines it covers.
  • Import your models: bring in your existing decision tables and process models.
  • Extend: AI drafts and your people approve, and a bad change to your data comes back exactly.

What it meansNew products without waiting on the integration queue, and the old estate shrinks with no big bang, so you stop paying twice.

03Run

Change the business without breaking it.

Bring in AI while the core runs. Application errors roll back changes for humans and AIs alike.

  • Real databases: your data sits in an enterprise database, a standard relational database, with tables built from your definitions.
  • Standalone code: we’re building ahead-of-time compiled code you can inspect, which runs in your environment without calling us to process a transaction.
  • Explainable decisions. Trace any value to who changed it, when and why.
  • Fits your enterprise tooling: diff and test rules in your pipeline, export the record to plain files, and keep AI tools in one reviewed catalogue, with the API generated from the same definitions.

What it meansAI-speed delivery on a backend that’s deterministic by design.

04Prove

Say yes to AI and stand behind it.

Show the board and the regulator who or what changed what and why, from your own record, with four-eyes sign-off where you set it.

  • Your record, not a sample: each field change records old value, new value, who and why, and exports as a signed evidence bundle for auditors and regulators.
  • The rule in force: each outcome shows the rule in force at the time and who approved it.
  • Checkable without us: evidence follows CRAWCUS, an open spec we publish (CC-BY-4.0), and an open-source verifier lets anyone check it.

What it meansAttest to every change, not a sample, and answer auditors from the record, not a spreadsheet.

Enterprise control, built in

People in the loop

Approval is yours to set: four-eyes sign-off on the work you choose, in your order, and the maker never signs as checker.

Roles and permissions

Roles, permissions and scope are clear records your team can review.

AI as a named actor

AI is named as the actor on each change it makes, and fenced out of the fields you choose.

Enterprise sign-in, one door

Sign in through your own identity provider. People, systems and AI agents use the same door and land on the same record.

No back doors

Fixes, configuration and the data you load go through the same door onto the same record.

How a day runs

People approve. A bad change comes back exactly, data included, or it is refused with the reason.

  1. AI proposesAI reads the case and drafts a summary and a proposed outcome for your analyst. We’re building this step.
  2. A named person approvesYour analyst proposes, and a second named person approves. The person who made a change cannot approve it.
  3. Your rules are checkedWe’re building this so committed behaviour comes from readable rules, not the model, before anything is written back to the core.
  4. One record, with evidenceWho proposed, who approved, which rule applied and what changed, on one record you can show the examiner. A bad change comes back exactly, data included, or it is refused with the reason.

US community-bank examples

Built around what your examiners would see on the record.

On the core you already run. The lead pilot is BSA/AML and SAR decisions.

BSA/AML and SAR

The lead pilot, above. Show the examiner why each SAR was filed, or wasn’t. We’re building this.

We’re building

On the core you already run

Use the apps you already run in many ways, with each change going through a single door onto one record.

Today

Evidence for auditors and regulators

Your record, not a sample: exports as a signed evidence bundle for auditors and regulators.

Today

Our first build is Basel regulatory reporting in Europe; DORA support is being built.

On the roadmap

On the roadmap for US community banks

Each one is just a pack on the same platform: the rulebook loaded as data, with its checks, decisions and evidence, not a custom build.

BSA/AML SAR decisions, with CTR and the Kansas exam record

Show why each SAR was filed, or wasn’t: the decision, reviewer and reason already sit on the record, as does the Kansas OSBC exam record. We’re building Currency Transaction Reports and the rest of this pack.

Partial

Call Report assurance (FFIEC 051)

On our roadmap: a Call Report pack for FFIEC 051, so each figure traces to its source and each adjustment is approved by a named person.

Roadmap

OFAC sanctions screening record

On our roadmap: an OFAC sanctions screening pack, so a blocked or rejected party is kept on the record with the reason, ready for your examiner.

Roadmap

Third-party and core-provider risk

On our roadmap: a third-party and core-provider risk pack, so each relationship keeps its reviews, issues and due diligence on the record through the life of the contract.

Roadmap

HMDA loan register

On our roadmap: a HMDA Loan Application Register pack, so each mortgage application line is kept with its source, ready for the annual filing.

Roadmap

Later: CRA, fair lending and Reg B, Reg E / Z / DD / CC, flood, FedNow, and Section 1071.

Contact

Talk to the founder.

Paul WanderFounder, Tallyseal
hello [at] tallyseal.com

Download the deck (PDF, 10 slides)