From loan application to red flag report,governed out of the box, with borrower data never leaving your tenant.
Co-Analyst monitors your loan book against a register of red flag parameters mapped to the fraud types your institution actually sees, assesses applications against your credit policy, and drafts the reports your risk team acts on, all inside your own Azure or AWS tenant with a lineage record on every step. Your analysts keep the decision, with the evidence attached, and nothing is exported or uploaded at any point.
governed by design · no borrower data leaves your tenant
76% of US organisations faced attempted or actual payments fraud in 2025. Only 17% use AI against it..
AFP 2026 Payments Fraud and Control Survey, January 2026
Fraud now costs US businesses 9.8% of revenue, up 46% year on year..
TransUnion H2 2025 Global Fraud Report, 2025
ASSESS / DECIDE
Applications assessed against your policy, not the loan officer’s memory of it
Credit policy is applied at origination by people working under volume, and exceptions to it leave no trace.
- The problem
- Every application is checked against credit policy and affordability rules by someone with a queue behind them. The rules are documented, but the check is a reading, so consistency varies by officer and by branch, and an application that should have triggered an exception is approved without anyone recording that the exception existed. That approval becomes visible only at default, when the cost is fixed.
- What ProSyft does
- Co-Analyst reads the application and its supporting documents against your own credit policy and affordability rules, flagging every departure from policy with the clause and the document that triggered it, and assembles a decision record for the credit officer to approve.
- Outcome
- Exceptions to policy surface before approval rather than at default.
The problem: the red flags are raised, but nobody acts on them in time.
In most lending institutions fraud is detected after the money has moved. Collections, disbursements and account activity are reviewed by a separate team on a separate cadence from the branch staff who handle the borrower, so an undeclared collection or a pseudo borrower is found weeks later in a reconciliation rather than in the week it happened. Across hundreds of branches, that gap is where losses accumulate.
Even where monitoring exists, the reporting cycle runs at manual assembly speed. A flag is raised in one system, the context sits in another, and the report that reaches a decision maker is compiled by hand, so the interval between a red flag and a named person acting on it is measured in days. A monitoring programme that detects fraud but does not shorten that interval has not reduced the exposure.
Does your data stay with you?
The usual assumption is that monitoring a loan book for fraud means a vendor holds your borrower and transaction data.
Co-Analyst’s answer: it runs inside your own tenant, so no borrower, branch or transaction data leaves your environment at any point.
Can your solution show you its reasoning?
The usual assumption is that a fraud flag from an AI system cannot be explained to an auditor or a regulator.
Co-Analyst’s answer: every flag carries the parameter that raised it, the fraud type it maps to and the transactions that triggered it, so the reasoning is the record rather than a reconstruction.
Who is accountable when it is wrong?
The usual assumption is that automated monitoring produces a list of flags with no one standing behind any of them.
Co-Analyst’s answer: every flag is routed by severity to a named person in the risk team, and anything that fails verification is escalated rather than passed through.
Can you control the cost?
The usual assumption is that running AI across every branch and every account means a bill that grows with the portfolio.
Co-Analyst’s answer: each step sees only the evidence it needs, so every run is bounded and monitoring cost stops tracking portfolio growth.
How it Works
EVIE™, the orchestration engine inside Co-Analyst, deploys inside your own tenant and runs credit and fraud workflows end to end. Your team reviews governed outputs with the evidence attached, and anything that fails verification goes to a named reviewer.
Connect to your existing systems
Native connection to your loan origination system, core banking system, collections data and Databricks or SQL, with no export and no upload.
Select and register the evidence
EVIE™ runs red flag parameters across branch and account data and checks applications against your credit policy, recording which evidence each step used and why.
Generate verified, hallucination-contained outputs
Credit decision records, red flag reports and weekly briefs in Word, Excel or PDF, with every claim checked line by line against its source.
Review inside your perimeter
Your Azure tenant or on-premise, human-in-the-loop approval at every stage, a named reviewer for any exception, and audit logs retained locally.
Benefits
Sovereign
Borrower, branch and transaction data never leaves your tenant.
Verifiable
Every flag is checked against the transactions that raised it, so there are no unevidenced alerts and no fabricated patterns.
Accountable
Every red flag is routed to a named person by severity, with the rationale recorded at the moment it is raised.
Bounded
Each step sees only the evidence it needs, so the cost of every run is bounded and monitoring cost stops tracking portfolio growth.
Speed
Red flag reporting that took a day runs in minutes, with the first workflow live in weeks.
Scale
Portfolio and branch growth stop driving risk headcount.
Four ways to automate credit and fraud risk. Three cost you something you can't get back.
General-purpose AI assistants
Days to deploy, minimal governance, vendor-side data custody, unmanaged hallucination risk. Compliance stops these at the pilot.
Consultants and custom agents
Eight to sixteen weeks, high cost, variable governance, shared data custody. Typically shelf-ware within six months.
Building it yourself
Six to twelve months to first workflow, highest total cost, permanent maintenance obligation.
ProSyft Co-Analyst™
Deployed inside your own tenant with the first workflow live in weeks. Every claim traces to source, every output is verified line by line, a named reviewer stays accountable, and data custody never leaves you. The governance arrives with the software rather than being added afterwards.
The control of a self-build, the speed of off-the-shelf, governed out of the box, inside your own tenant.
AML and fraud detection, Philippines
RAFI Microfinance Inc., a regulated microfinance institution serving 2.6M clients across 323 branches and 443,315 active loan accounts, has run Co-Analyst inside its own infrastructure since July 2025. The red-flag reporting cycle on its core fraud monitoring workflow moved from 24.5 hours to around 5 minutes, a 99% time reduction, delivering £112K in high-confidence validated savings in Phase 1 of the fraud red-flag workstream. Total annual value across all workstreams is estimated at £239K to £404K, conservative to optimistic.
The architecture that monitors fraud across 443,315 active accounts is the same architecture any lender needs across its loan book.
- 01
Juniper Research, 27 October 2025: fraud detection and prevention spending. Tier 1, analyst firm.View source
Let’s map your highest cost compliance workflow and show you what governed AI looks like running inside your own environment, on your data.
- 01No data leaves your tenant.
- 02Every claim traces to source.
- 03A named reviewer stays accountable.
- 04Live in weeks.