Turn a pile of customer reviews into action — in one pass, in under 30 seconds.
FreshTake is a workflow tool for restaurant managers. Paste or upload a batch of reviews, and a Microsoft-Copilot-driven engineered prompt returns a structured table — sentiment, aspect tags, priority, a suggested action, and a ready-to-edit first reply — for every review, in one pass. Managers stay in the loop: every reply is reviewed, edited if needed, and approved by a human before it's sent. High-priority issues route straight into an action queue so nothing slips through.
North Star metric
≥ 90%
of reviews get a personalised reply sent within 24 hours — up from a baseline under 30%.
≥ 90%
categorisation accuracy vs. expert review
< 1 min
staff time per response (was 5–10 min)
< 30 sec
to process a 10-review batch
0%
hallucinated tags on N/A reviews
Who uses FreshTake
Four personas drawn from the PRD's "who" — each sees a tailored slice of the product.
Maria Reyes
General Manager · Riverside
Primary user. Fast daily triage of new batches; approves and personalises AI-drafted replies; owns the high-priority queue.
Sees: Dashboard, Batch results, Review detail, Action queue
Jordan Tate
Front-of-House Supervisor · Riverside
Drafts and sends day-to-day replies; edits AI drafts for tone before approval; logs follow-up actions with the team.
Sees: New batch, Review detail, Action queue
Dana Okafor
Regional Operations Manager · 12 locations
Calibrates priority judgement across locations, reviews aspect-level insight, and watches month-over-month sentiment trends.
Sees: Trend report, Action queue, Dashboard (multi-location)
Sankar Kumar Palaniappan
Prompt Engineer / Admin (PRD §10 team)
Owns the prompt library and version history, runs the weekly 20% accuracy audit, and tunes brand-voice settings.
Sees: Prompt library, Settings
The five flows (PRD §3.3)
This is the prototype's skeleton — click through any flow start to finish.
Flow 1 · Story 1
Review batch processing
Paste or upload a batch → submit to the engineered prompt → processing state → structured 5-column table → export to CSV/Excel.
Flow 2 · Story 2
Priority-based action routing
Filter by Priority = High → assign the suggested action to a department → log it with a timestamp → track Open / In Progress / Resolved.
Flow 3 · FR9
First reply drafting & sending
Open a review → read the AI-drafted reply → edit/personalise → human approves and sends. There is no automated send — ever.
Flow 4 · P1 · FR10
Prompt iteration & improvement
Weekly review of miscategorised cases → versioned prompt change log → accuracy tracked against the 20% manual-audit baseline.
Flow 5 · P2 / V2 · §3.6
Trend monitoring
Monthly aggregation of % positive / negative / neutral, most-tagged aspects, and an alert when negative reviews cross 30% of the month's volume.
All screens
Every screen, in storyboard order. Each one links back here and to its neighbours.
Sign in
Entry point for managers and admins.
Dashboard
GM home: North Star metric, weekly snapshot, high-priority queue, response-rate trend. Includes a first-run empty state.
New batch
Paste or upload reviews from Google / Yelp / TripAdvisor / direct form; batch-size validation (≥10).
Processing
"Copilot AI is processing your batch…" with a live <30s progress indicator.
Batch results
The 5-column structured table — sentiment, aspect tags, priority, suggested action, 1st reply — sortable, filterable, exportable.
Review detail & reply editor
Full review, rationale, "why this priority?", editable AI draft reply, regenerate, approve & send (FR9).
Action queue
High-priority items, department assignment, Open / In Progress / Resolved pipeline, ≥80%-actioned-in-48h target.
Trend report
Monthly sentiment mix, most-tagged aspects, a >30%-negative alert, rating trend toward +0.3 stars.
Prompt library & accuracy audit
Versioned prompt change log, weekly 20%-sample accuracy audit, "report a miscategorisation" flow.
Settings
Locations, brand-voice tone controls, team & roles, notifications, the locked "human approval required" toggle.