AI Workbench + BusinessBANKER

AI Adjudication for Commercial Lending

We capture how your best credit analysts decide, turn it into AI agents, and have independent models check each other before a person on your team makes the call. It runs on infrastructure you control, on top of the loan and core systems you already have.

See How It Works Schedule a Briefing
SOC 2 Type 2
Clients using our solutions

The Method

Four Steps From Expert Judgment to a Decision

No single model is trusted on its own. Every recommendation passes through the same four steps, and the final decision always belongs to your people.

01

Expert observation

We capture your senior analysts while they work on real files and record how they reason: what they check, in what order, and what makes them stop. That becomes the agent's playbook.

02

Two models check each other

At least two agents running on models from different labs process the same file. Each must return output in a strict format (JSON) against the same schema, and the outputs are compared field by field. Any disagreement is flagged, never averaged away.

03

AI supervisor

A supervising agent reviews the whole file and the flagged disagreements, and raises anything that needs a human's attention. It is live today at Al Rahma in Senegal.

Al Rahma case study
04

The human decides

Your analyst sees the recommendation, the evidence behind it and every flag, then approves, changes or rejects it. That decision is recorded in BusinessBANKER.

How Cognitive Foundry Captures Expertise

Where the Analysis Happens

AI Workbench

The agents that extract, analyze and recommend, built from your own credit experts' judgment. You choose the model for each agent and can swap it without rewriting the workflow. Deployed in your cloud, on top of your existing loan and core systems; nothing is replaced.

Explore AI Workbench
2+ Models per file
JSON Strict output schema
Logged Every disagreement

Where the Decision Is Recorded

BusinessBANKER

AI Workbench prepares the analysis. BusinessBANKER is where your credit team reviews it, signs off and keeps the record: every version, every approval and every change, ready for audit. Where your current loan system can't carry out the resulting action, BusinessBANKER does.

See BusinessBANKER in the Field

What We Can Show You

Evidence, With Its Limits Stated

Early benchmark, small sample

In a draft study of 10 loan files, the system reached the reference decision 96% of the time, against 57% for a senior human adjudicator.1 Ten files is a small sample, so the margin of error is wide. That is why every engagement starts in shadow mode on your own files.

In production at Al Rahma

Loan applications that took hours to write up now take about 10 minutes on a tablet, and money reaches the borrower in under four hours.2

Read the case study

Your cloud, your models

No model lock-in and no vendor-hosted black box. The platform runs in your own AWS account, and you pick the model behind each agent.

Governance built in

Audit trail, confidence thresholds, and an automation-threshold specification written against OSFI E-23, delivered as a model-governance pack.

Security

Cognitive Group holds its own SOC 2 Type 2 report, for the period ending 31 December 2025.3 This is our audit, not our cloud provider's. SSO, MFA and role-based access control, plus a live Fiserv DNA core integration.

1. Cognitive Group, “Beyond the Illusion of Generality”, draft white paper, January 2026. 10 loan files scored against a reference decision, comparing one senior adjudicator (10+ years) with the multi-agent system. Not peer reviewed. Small sample, wide margin of error.

2. Al Rahma, Senegal, production deployment. Time from starting an application to submitting it on the field tablet, and from request to disbursal, as measured by Al Rahma. Figures as of September 2026.

3. Service Organization Control 2 (SOC 2®) Type 2 report issued to Cognitive Group, for the period ending 31 December 2025.

What we don't claim: best-in-class depth in every niche, such as commercial real-estate valuation, consumer scoring or document forensics. We work with partners there. Our claim is breadth, governance and control in one platform.

Research

The Same Method Outside Banking

Banking is our business. Our research group applies the same expert-capture and cross-checking method in aerospace and defence, where a wrong call is expensive and every decision has to be explained afterwards.

Aerospace safety: SymbIA

Human-automation teaming for single-pilot operations, where pilot and system disagree.

SymbIA

Cognitive cockpit: C-Pilot

Capturing pilot reasoning in real time and supporting it with AI companions under high workload.

C-Pilot

Defence: situational awareness

A mission-planning AI in development, and a live demo that classifies air traffic with several cross-checking agents.

Mission planning and demos
See All R&D

How You Buy

Start With a Shadow-Mode Pilot

Start with a fixed-fee shadow-mode pilot of 10 to 12 weeks, then make a go/no-go decision. If it's a go, AI Workbench runs in production on your existing workflow for an annual platform fee plus a fee per approved transaction, and BusinessBANKER adds the execution layer where your loan system falls short.

12 wksPilot duration
$75K CADFixed fee
≥40%Target reduction
0Systems replaced
1Decision at the end
Schedule a Briefing