The lending back office, staffed by AI employees.
Vera reads every credit file. Asha completes every docket. Kai works every alert. Ira closes every break. Named AI employees join your bench, work inside your maker-checker, and put a draft in front of a named approver before anything counts. All of it audited, exportable, and inside your approval rules.
Where the opportunity sits.
We spent our time inside lending operations before we built anything. The pattern repeats: the intelligence went into the front of the business, and the hours are still in the middle and the back.
The file that gets assembled by chasing
A loan file is not produced. It is assembled, one follow-up at a time, across customers, DSAs and branches. Your LOS can tell you what is missing. A person still has to chase it, and after disbursal the deferral register ages quietly in a drawer. That chase is a role, and a role can be staffed.
The review function that samples
Second line review typically samples a few percent of files, because reviewer hours are finite. The sample is defensible. It is also why the portfolio question stays open between audits. Read every file against the policy and the question closes every week, with evidence attached.
The queues your detection systems create
Every detection system you bought creates a queue it does not work. Alerts land faster than analysts can disposition them, so triage becomes the job and judgment becomes the backlog. The evidence, the draft disposition and the trail can all be ready before your analyst opens the case.
The books rebuilt by hand every cycle
Matching engines clear most transactions and leave the tail to people. Co-lending splits, partner settlements and regulatory returns get rebuilt by hand every cycle, from the same sources, to the same template. Work that repeats to a template is work an AI employee can carry.
Core. Control. Business Support.
We organise the work of a financial institution into three sides and build AI employees for each. Control is where we are deepest today. Roles marked on demand are scoped and built for your institution.
The revenue engine. You likely have AI here already; our employees slot into the document work around it.
Chases missing documents from customers, DSAs and branches, completeness-checks the docket, clears pre- and post-disbursal deferrals.
Meet AshaReads title reports and legal documents against your checklist, flags gaps for counsel to rule on.
Drafts notices and legal correspondence in approved language; counsel signs every send.
The checking side: risk, review, audit. Built first, because this is where approval-first AI matters most.
Reads every sanctioned file against your credit policy. Your second line samples; Vera reads all of it.
Meet VeraWorks the alert queue oldest-first, drafts closures with evidence, escalates real cases with a file.
Meet KaiAssembles control-test evidence and drafts working papers for your auditors to sign.
The back office that keeps the books honest.
Closes the exception queue your matching engine leaves behind, item by item, evidence attached.
Meet IraKeeps partner files, agreements and schedules reconciled across co-lending arrangements.
Compiles recurring returns from the source systems for review and sign-off.
Or start with one employee across all three.
Tell us the single most painful job on each side of your house. We scope one AI employee that takes the P0 from Core, Control and Business Support together, so one hire proves the model across the institution.
The work, on screen.
Not renders. This is the shape of the product: queues your managers can see, findings with the evidence beside them, and an approval between every draft and the record.
Every file read. Every finding cited.
Vera works the sanction queue file by file against your own credit policy, not a generic checklist. When she finds a breach, the finding arrives with the page it came from, and it stays a draft until your reviewer signs it. Approve, edit or reject; each decision is logged under a name. Your second line stops arguing about the sample and starts talking about the portfolio.
Your SOP becomes the way she works.
Hand over the credit sanction SOP and it compiles into a living process map your risk team can inspect. Every rule links to the page it came from, so when a check fires you can see exactly which clause fired it. Change the policy and the map changes with it, on the record, with a named approval. This is how the employee stays yours.
The docket completes itself. Politely.
Asha tracks every file against the product checklist from login to disbursal and beyond. Whatever is missing gets a drafted follow-up, in your templates, waiting for a named person to approve the send. Deferrals get chased to closure and escalated before a committed date breaches. PDD stops being the place where files go quiet.
From first file to full deployment.
The custom-grade engagement, for institutions that move carefully and want proof first. It mirrors the Winsen Labs engagement model: 3 to 4 months, a forward-deployed team working alongside your people.
Understand, in week one
A working session with your credit, ops and risk teams. Nothing to sell in it. We map the queue, the maker-checker rules, and the historical files your team has already reviewed.
Deploy narrow, weeks 2 to 4
One team, one workflow, on your stack, inside your approval rules. The first runs are a paid pilot on files your team already checked: three hundred sanctioned files, a quarter of reconciliations, a month of alerts. Measured against your own ground truth. Exports are enough to start; nothing touches production.
Measure, weeks 4 to 12
Approved-work volume, time returned, error rates, on shared dashboards both sides read. What your process caught, what it missed, cost and turnaround per file.
Expand, when proven
Only when the measurements make the case. More desks, more workflows, same approval rules. We take on a small number of engagements at a time.
Single-tenant, your regionEvery output human-approvedFully audit-trailedNo expansion without measured results
The questions your board will ask.
Straight answers, before the deck reaches the risk committee.
Our data science team already builds AI. Why not build this in-house?+
What does the regulator see?+
Does any data leave our environment?+
How does the pilot work?+
What does it cost?+
What happens when an employee gets something wrong?+
Put an AI employee on your worst queue.
Bring one SOP and three hundred historical files. We will show you the process map, the review queue, and the findings your sampling never saw.
