Every source-to-pay process, an intake, a sourcing event, a supplier record, a contract, a payment, rests on a small set of facts: a tax ID, a term end date, a price, a match between a PO and an invoice. An agent can only act on a process once those facts check out. Most of the time nobody has looked. This board reads a source-to-pay program's own systems and marks, process by process, exactly which facts are verified, which two systems disagree on, which were never found, and which are old enough to doubt.
It writes nothing and automates nothing. It answers one question before anything is asked to act: is this ready.
The problem framing is public: data quality as the standing blocker on agentic ROI in procurement, and the September 2026 New York session on the capacity gap agentic AI is meant to close, both already on the public record. Deloitte's own 2026 survey of enterprises already piloting agentic AI found 5% highly prepared and 15% at scaled, cross-functional adoption, an enterprise-wide finding, not a procurement-specific one, cited here only as outside confirmation of the gap. Every process, supplier, contract and figure below is sample data generated for this page. No real company, supplier or system appears here.
Every process instance runs the same check: a short list of facts it needs before an agent can safely act, verified against the systems that hold them.
Requester, budget owner, category, business case.
Bid deadline, evaluation basis, awarded supplier match.
Legal name, tax ID, banking, risk status.
Term end, renewal window, governing version, signatory.
PO reference, three-way match, terms, tax treatment.
Filter by process, readiness, or the kind of block. Select any row to see, fact by fact, what checked out and what did not.
| Reference | Process | Subject | Facts checked | Readiness | Last read |
|---|
This page runs on sample data. The production board reads from the systems a source-to-pay program already runs, the ERP, the supplier master, the contract repository, the sourcing platform, the AP system, through access issued and revocable. It verifies. It does not decide, and it never acts.
The board reads. It never writes to the ERP, the supplier master, or any system of record, and it never triggers an agent or a workflow.
Access is limited to the processes in scope. For a pilot, that means one process, sourcing or supplier onboarding, and nothing outside it.
Revoke the access key and the board goes dark. Nothing to uninstall, and no data left behind that was not already held.
Small enough to run alongside an existing rollout without touching it. Large enough to say, in writing, how many of the processes that program is about to automate were actually ready.
Connect one process through read-only access. The board fills from what already exists, and every fact names the system it was read from.
A weekly read. What is ready, what is blocked, and whether the same fact keeps failing the same way.
Keep it and add a process, or switch it off. Either answer is a clean one, and the readiness rate on day 90 against day 1 is the number that matters before scaling an agent further.
Great Breads Studio is a design and technology studio in Harlem, New York. The studio designs and engineers systems for founders, brands, and public-sector clients, and builds its own software products alongside brand, web, and editorial work.
The studio built this board. It is shown here with sample data so it can be read without a login. The board is not a source-to-pay platform and does not compete with one: it is a check that runs in front of whatever platform is already in use, and it makes no claim about which system is at fault when a fact does not check out.