Case study

From operational data to shared intelligence

DecafSupport AI turns an operational datasource that only specialists could use into a governed layer any authorized team can explore, query in plain language, and use for decisions.

Built inside Decaf by Decaf Labs. We tested the pattern on our own support operation first.

The problem

The knowledge existed. Almost nobody could reach it.

Support conversations, tickets and claims held the clearest picture of what customers were hitting. Reaching that picture meant knowing the source system, writing queries, or asking the one person who could. Most questions simply went unasked.

What we built

Ask the operation a question, get an answer with receipts

An intelligence layer that connects to the datasource, indexes and structures it under shared definitions, applies team permissions, and answers plain-language questions with the evidence linked.

decafsupport · internal
Ops · Ana
Which claim issues drove the most contacts this week?
DecafSupport AI

Three issues account for most of this week's claim contacts:

Cashout delays41
KYC retries27
Link expired16

Cashout delays are concentrated in two corridors and rose midweek.

12 supporting conversations8 related claimsView evidence →
"Cashout started Monday, still says processing. Client asking twice a day."conversation · anonymized
"Second KYC attempt rejected again, same document."claim note · anonymized

Representative interface · synthetic sample data

explorer · claims
This weekCashoutsKYCAll corridorsAssigned
IssueContactsStatusTrend
Cashout delay4118 open▲ rising
KYC retry loop276 open▼ falling
Payment link expired163 open▼ falling
Balance mismatch92 open— flat

Interface recreated with sample data

How it fits together

One governed path from source to answers

SOURCE OUTPUTS Support data tickets · chats · claims one specialist system Index Definitions Permissions AI LAYER answers with evidence Answers Dashboards Reports ASK IT QUESTIONS "Why did contacts spike on Tuesday?"

One source, different questions

The same layer answers every team

Support

What issues are generating repeat contacts?

Product

Which workflows create the most confusion?

Operations

Where are cases becoming stuck?

Leadership

What changed this week, and why?

The build

Four phases, run on ourselves

build.logfour phases
  • Phase 1Connect the source.
  • Phase 2Establish shared definitions and permissions.
  • Phase 3Put evidence-backed answers in the team's hands.
  • Phase 4Generate dashboards and recurring reports.

The result

Questions stopped waiting for a specialist

DecafSupport AI is one specialized agent in Decaf's internal AI operating layer. The same architecture can turn support, sales, finance, compliance, or operational data into governed intelligence that every authorized team can use.

Your knowledge is already there

Your company already has valuable knowledge trapped in specialist systems. We can make it visible, queryable, and useful across the organization.