Evidence first
Every decision begins from identified source records and their freshness, not an unsupported model answer.
Decision engine
Context changes the action.
Brainfish AI is designed to separate facts, interpretation, and authority. Data functions produce the numbers. AI interprets unstructured requests. Policy decides what may happen next.
Discuss your use caseRule precedence
Company-wide policy, team procedure, customer terms, and case-specific exceptions are evaluated in a visible order.
Confidence-aware paths
Ambiguous entities, low-confidence matches, missing data, and conflicting rules produce a review task—not an invented answer.
Dry-run mode
Rehearse decisions against sample or historical cases without writing to a connected operational system.
Explainable outcomes
Show which facts and rule caused an action, a stop, or a request for approval.
Change control
Version rules and test changed behavior before it becomes part of an active workflow.
Illustrative sample · Northstar Components
A valid order that still should not post.
The customer has enough credit and every line maps to the catalogue. A contract price creates a discount outside the sales team's authority.
Order value€18,460Incoming PO
Credit available€24,100ERP + receivables
Price variance−9.5%Contract vs catalogue
WRITTEN RULE
Discounts above 7% require finance approval, even when customer credit is sufficient.
OUTCOME
A complete draft is prepared; posting is held for a finance reviewer.
Start with one decision
Bring the workflow your team knows by heart.
We’ll map the systems, facts, rules, and approval point before discussing a pilot.