CADA · Cognitive Augmented Discovery Agent
CADA takes the transactions your monitoring flags and the triage cases your team opens, assembles the evidence, tests competing explanations including the lawful one, drafts the reason for suspicion and recommends the next action. Your officers decide.
Complements your monitoring and case management · Built for Nigeria · Human decision on every case
How it works
The work your analysts do by hand today, done the same way every time, with the reasoning written down.
Every alert needs evidence from several systems, the innocent explanation ruled out, the right offence named, and a reason for suspicion that survives review. Done by hand, case after case.
Sent to CADA from your case manager
CADA takes the alert your monitoring system raised, or a triage case, straight from your case manager. Nothing is replaced. The alert is where the investigation starts, not where it ends.
The payment path, counterparties and finding are shown together. CADA separates the observed receipts from the onward movement suggested by the pattern indicator, then marks the link that still needs checking in the bank's records.
CADA tests whether the observed money movement supports the pattern, whether a fraud or corruption lead has independent evidence, and whether a lawful payment purpose better explains the activity. These are investigative leads, not findings of guilt.
The concern is reduced after verification; an officer must decide the next step.
Before recommending anything, CADA checks the linked payments against the finding, separates what was observed from what was inferred, and shows why the initial concern weakened.
E1E4E7
The draft connects the payments, finding, limitations and lawful alternative. The bank's officer checks its own records and decides whether the grounds warrant filing an STR.
Illustrative choice only. No report is sent from this website.
Explore the sandboxAgree or override, with the reasoning and evidence on file either way. A decision not to report is as documented as a decision to report.
Two observed receipts enter the customer account; the onward link still needs verification.
Why now
The CBN's Baseline Standards for Automated AML Solutions (10 March 2026) give banks 18 months to run automated AML, from monitoring and screening to investigation workflows and regulatory reporting, and require AI models to be independently tested every year.
Most banks already own monitoring, screening and case management. What they rarely have is the reasoning in between: whether the alert identified the right concern, what was missed, and what to do next. That is where CADA works.
Modules
CADA Core works inside the case management system you already run. Add the modules that matter to your institution.
For every flagged case: the evidence assembled, competing explanations including the lawful one, CADA's own critique, a recommended next action and a full audit trail, delivered into your case manager through an API or webhook.
A first draft from CADA's reasoning, for your analyst to edit and your MLRO to sign.
Graph investigation of subjects and counterparties, and an assistant grounded in the case evidence.
For the CCO: filings by state and LGA, by predicate offence tagged, by branch and filer, and where the tag differs from what the evidence supports.
Ranks your monitoring queue using your own past filing outcomes, so your team reaches the high-priority cases first.
KYC files, statements and documents a customer sends later, read into the case so CADA can test whether grounds for suspicion remain.
Your internal typologies and procedures configured into how CADA reasons.
CADA inside your own environment, on your hardware if you have it.
The evidence for the yearly independent testing of AI models the CBN standards require.
Pricing is shared with your team directly, with the modules you choose. Request pricing.
Trust
CADA at Bank A is not CADA at Bank B.
CADA's deterministic engine finds, tests and ranks. The language model drafts the wording and runs a bounded critique; it cannot create a finding.
The escalation rule is fixed in the software and readable by your MLRO, never the opinion of a model.
The 27 predicate offences in Nigerian law, mapped to the FATF's 21 categories. Every case states which offences were assessed and which were not.
CADA cannot reach an outside network to run its model. Any change is deliberate and recorded on every case.
Signals are ordinal and marked uncalibrated when they are. No black-box score presented as a probability.
Every change produces a new version stamp checked before the system will run.
Deployment
Connected to your alert source and case management system through an API or webhook. No new platform for your analysts to learn.
For teams that want graph investigation and the Witness copilot alongside CADA.
Managed in Nigeria by default. Private cloud or fully on-premises when your policy requires it.
Already have an AI model or subscription? It connects after passing our qualification run on a test set.
On-premises, your hardware replaces ours, with no change to your licence.
We alert you as volume approaches your band and keep processing. An investigation is never stopped.
See CADA at work
Watch the product film here, then explore a protected investigation case.
Playing the film loads YouTube content. Privacy information
Sandbox
Five anonymised cases, from the alert to a drafted reason for suspicion and your own decision. Invited reviewers receive an access code; we run the demo with you.
Explore sandbox accessOur story
CADA began in 2023 as FIAS, our first-generation analysis system, and was rebuilt as CADA in 2026 around one idea: an alert is a lead, not a conclusion. We know the problem because we have worked it, and we would rather improve CADA with your team than hand you a platform and leave.
Founder
Next step
Book a walkthrough of the sandbox, request pricing for the modules you want, or start a 90-day evaluation on your own cases.
Prefer email? cada@basmu.xyz