The platform

One engine under every Analyst.

All four Brisc Analysts run on one shared engine: same secure intake, same golden schema, same deterministic matching core, same review queue. Learn it once and every Analyst behaves the same way.

The Brisc engine
What arrives
broker-statement-june.eml june-bordereau.xlsx remittance-scan.pdf
The engine same five stages under every Analyst
Ingest · secure inbox, upload, scheduled pull3 files in
Normalise · one golden schema214 rows standardised
Match · rules decide, AI argues the borderline213 matched · 1 flagged
Review · exceptions queue with evidencequeue: 1 item
Post · matched records to your systemsevidence attached
Where it lands · yours, not ours
your PAS ✓ posted your GL ✓ posted data lake ✓ synced
four Analysts · one engine every value evidenced
Intake

Send the file as it arrives.

No template to fill in, no format mandate, no re-keying before the work can start. Brokers and banks make your data messy; that isn’t yours to fix.

A secure inbox of your own

Forward the broker email, statement attached, and the platform picks it up from there.

Batch upload

Drop a month or a backlog in one go: PDF, Excel, Word, scans, email bodies.

Scheduled pulls

Recurring feeds collect themselves, so monthly files stop needing a person to fetch them.

Normalisation

Absorb the format. Keep the schema.

Every file, whatever its shape, lands in one golden schema: payer resolved, currency and dates aligned, commission and tax positions made explicit.

One golden schema

Forty broker layouts become one structure your team and your reports can rely on.

Mapping that persists

Teach the platform a layout once and it holds, month after month.

Format-drift detection

When a broker changes a layout, the platform notices and asks only about what changed.

Configuration

Configured to your book. Edited by your admins.

The engine is shared; the configuration is yours. Everything below is versioned, per tenant, and editable by your own admins, no change request to us.

Field catalogs

The fields each Analyst extracts and validates, defined per document type and per program.

Versioned prompts

The instructions the AI reads from, versioned per tenant, so a change is deliberate and reversible.

Broker tables

Your brokers, payers and aliases as living reference data the whole engine draws on.

Rule libraries

The matching and validation rules that decide the arithmetic, readable and maintained in your tenant.

Matching & checks

Match on rules your team can read.

The numbers are never a guess. Matching runs on deterministic rules, tolerances and confidence tiers your team can inspect; AI argues the borderline cases, never the arithmetic.

The full trust story, lineage included: accuracy and audit.

Deterministic core. Rules and tolerances decide the arithmetic. The same inputs produce the same answer, every run.
Confidence tiers. Routine volume clears; uncertain rows carry their confidence and queue for a person.
AI where reading is hard. Extraction from messy documents and the argument for a borderline case. Never the final call.
Checks at ingestion. Totals balanced and rates checked the moment a file lands. What the platform isn’t sure about, it queues; it doesn’t quietly correct.
What you see

Work the queue, not the workbook.

An exception-first dashboard: routine volume clears itself, uncertain rows wait for judgment, and every decision lands with its reasoning attached.

Reasoning per value. Every extracted value carries a written explanation and a click-through to the exact source document behind it.
Workflow states. Draft, reviewed, approved, export. Nothing leaves the platform while it is still an opinion.
Access control. Role-based access by team, geography and submitter, so people see the work that is theirs to see.
Field-level audit trail. What the AI proposed, what a human chose, who approved it and when, on every record, exportable in full.
What the AI learns

Turn every correction into permanent knowledge.

Not vague “learning from feedback”: three specific loops, each one inspectable.

Loop 1

Deterministic rules

A rejected match becomes a rule you can read; the same mistake doesn’t happen twice.

Loop 2

Prompt tuning

Rejected cases sharpen the model on your binders and your brokers, not a generic corpus.

Loop 3

Learned aliases

Confirmed payer and broker aliases become permanent reference data that survives staff turnover.

80% 92–95%

The honest ramp. 80% of records match on day one; 92 to 95% by about 90 days, as the rule base absorbs what the model surfaces.

Where output goes

Push into your systems, not over them.

Brisc doesn’t replace your PAS, your GL, or your data lake. It posts finished, evidenced work back into them and leaves the records where they belong.

API push

Matched cash, validated rows and structured submissions post into your PAS, GL or data lake, evidence attached.

CSV and JSON export

Day-one output with the evidence columns included; no integration project required.

Your own template

Where a system has no API, the platform fills the Excel template you already use.

Every channel and target, in detail: integrations. The engine on one job: bank reconciliation, bordereaux, submissions, claims.

Tenancy & trust

Run in a tenant that is yours alone.

Brisc is SaaS on Microsoft Azure, one dedicated tenant per customer. Not a shared database with your rows filtered out: your own database, storage and key vault, with no shared fallback to land in.

Isolation, SSO and model posture in depth: security and tenancy.

Dedicated tenant. Own database, own storage, own key vault per customer. No shared fallback.
Entra SSO. Your identity provider, your access policies, your allowlist.
Model posture. Your data never trains shared models.
Compliance. SOC 2 Type II · GDPR and CCPA compliant · trust.brisc.ai
Getting live

Go live in stages, each one earned.

Deployment is staged on purpose. Autonomy isn’t switched on; it’s earned, and you set the pace.

01

Day one: uploads in, evidenced output out

Your real files as they arrive, an evidenced CSV back, no IT project first.

02

Parallel-run a live month

The platform runs beside your current process while your team reviews every match.

03

Auto-push when the numbers have earned it

Finished work posts into your PAS and GL with evidence attached; exceptions still route to your team.

Walked in order, most teams are live in 2 to 6 weeks. Pricing follows the Analyst you start with: see pricing.

Go deeper

The platform, in depth

Common questions

How the platform works, answered

Is the matching based on rules or AI, and what does it learn from our corrections?

Matching runs on rules, tolerances and confidence tiers; AI reads the documents and argues the borderline cases, never the arithmetic. Your corrections become three things you can inspect: new deterministic rules, sharper prompts tuned on your data, and a permanent store of confirmed payer aliases. 80% on day one, 92 to 95% by about 90 days.

Does Brisc replace our PAS?

No. Your PAS stays the system of record. The Analysts do the work and post the results back, matched cash, validated bordereaux, structured submissions, with the evidence trail attached. Brisc is not a second ledger you have to keep in sync.

Where does our data live?

In a dedicated Microsoft Azure tenant that is yours alone: its own database, its own storage, its own key vault, with no shared fallback. Access federates through your Entra SSO. Your data never trains shared models. SOC 2 Type II, GDPR and CCPA compliant. Documentation at trust.brisc.ai.

Bring your own bordereau. Leave with it matched.

30 minutes, no slide deck, your real files: a bordereau, a statement, a submission. Your business is specific; the walkthrough should be too.

Book the walkthrough