Talk to us

Multi-site restaurant operator · AI practice

Operations teams now ask their sales data questions, and an AI analyst answers from the live POS.

  • AI data analysis
  • AI anomaly detection
  • Oracle Simphony
  • MCP
  • Claude
The analyst answering a week-on-week sales question with a comparison table
  • 11 daysfrom first commit to full release
  • 48commits across two services
  • DailyAI findings across every store
  • 16POS data tools exposed to the AI
Before

Point-of-sale data could only be queried with AI from one person's desktop. There was no shared access, no sign-in and no audit trail, and nobody was looking for the problems they didn't know to ask about.

What we built
Ask
Plain-English questions answered from live data, with every tool call shown as it runs.
Brief
Six fleet-wide metrics against a baseline, each one click from a drill-down question.
Pulses
Daily AI findings across every store, ranked by severity, with follow-up questions.
Model tiers
Fast, Standard and Deep answers, with a confirmation before switching to a costlier model.
Brief station with six fleet-wide metric cards
Brief: fleet metrics straight from the POS, no AI in the numbers.
Pulses page listing daily AI findings by severity
Pulses: findings the team didn't know to look for.

Where AI works in the solution

analysing data on request

Ask a question, get an answer from live data

Staff ask in plain English, such as how one store did week on week or where discounts are running high. The AI pulls the figures from the POS through 16 data tools, compares periods and stores, and explains the result. It never estimates: if the data isn't available, it says so.

spotting anomalies every day

Finds what nobody thought to ask

Every day the AI checks each store against its 28-day baseline and writes up to eight findings, such as sales drops, unusual discounts or voids. Each comes with a severity, the stores affected and follow-up questions, and every finding is validated before the team sees it.

Integration

One secured server is the only thing that touches the POS

  1. existing systemOracle Simphony BIsales, items, voids, tenders, staff
  3. we builtMCP data server16 tools, audit log, rate limits
  5. we builtAnalyst workspaceAsk, Brief, Pulses, sign-in
  7. AI modelAnthropic Claudeanswers and daily findings
SystemDirectionMethodWhat flows
Oracle Simphony BIinREST · OAuth PKCESales, checks, items, discounts, voids, tenders, staff and timing per location
Anthropic ClaudeoutSDKChat turns, tool results, daily findings
IdentitybothSDKSign-in, and user identity for audit attribution
next case studyOne source of truth, with an AI agent leadership can question
Read it →