case study8 min read

Pune QSR pilot: 30-day forecast accuracy at 91% dish-level

How a Baner quick-service outlet replaced gut-feel prep with dish-level forecasts — 91% accuracy, 68% less over-prep, and salary-cycle demand flagged a week ahead.

By Forkcast Editorial · HORECA research team

This is a fictional-but-realistic composite of a Baner, Pune quick-service outlet — 28 covers, 72% aggregator revenue, three years of gut-feel prep. Thirty days into the Forkcast pilot, dish-level forecast accuracy hit 91% and over-prep dropped 68%. Numbers below match what we see on comparable QSR pilots.

The outlet

Rolls and bowls format, Baner High Street, Petpooja POS since 2022. Weekday lunch is office delivery; weekends are walk-in families. No dine-in liquor. Aggregator share runs 68-76% depending on the week. Historically, prep was set by the owner's WhatsApp gut check at 10am — adequate on Tuesdays, wasteful on salary-cycle Fridays.

Onboarding: 9 minutes

  1. Petpooja API — read-only credentials; no menu changes.
  2. Recipe mapping — top 12 SKUs by revenue; auto-suggest matched 10; two manual portion tweaks on protein bowls.
  3. Fixed costs — rent, three salary lines, utilities entered once.
  4. First sync — 60 days of order history overnight; forecast live next morning.

Forecast accuracy at day 30

MetricBeforeAfterDelta
Dish-level forecast accuracy (MAPE)N/A (no forecast)91%
Daily cover-count accuracy±22% vs actual94%+72 pts
Over-prep waste (₹/day)₹1,800-2,400₹580-820−68%
Prep planning time35-40 min8-10 min−74%
Salary-cycle Friday missDiscovered at 1pmFlagged Tuesday brief+3 days lead

Top 12 SKUs account for 81% of revenue. Forecast accuracy on those SKUs was 91%; long-tail items (13-22) ran 78% — acceptable because prep batches are shared across bowl formats.

What the forecast caught that intuition missed

  • Salary-cycle Friday spike — delivery orders up 28% on the last Friday of the month; previously discovered when the chicken batch ran out at 12:30pm.
  • Monday post-weekend dip — walk-in covers down 18% vs Sunday; kitchen was over-prepping Monday lunch by habit.
  • Rain-day delivery shift — aggregator share jumps 12-15 pts on heavy-rain days; dine-in prep could be cut without affecting revenue.
  • IPL match evenings — partial signal; forecast undershot by 18% on one Saturday. Flagged for manual override next season.

Dish-level accuracy breakdown (top 5 SKUs)

SKU30-day unitsForecast MAPENotes
Chicken tikka roll1,8427.1%Stable weekday lunch driver
Paneer bowl1,2048.4%Salary-cycle spike modelled
Egg roll9869.8%Weekend-heavy; slight under-forecast
Peri-peri fries81211.2%Add-on; batch-shared with bowls
Mango lassi6456.5%Weather signal helped
We stopped guessing chicken quantity on Friday mornings. The brief tells us Tuesday what Friday needs — we just follow it.
Owner, Baner QSR (composite pilot)

What the owner actually does

The 7pm brief lands with tomorrow's SKU-level prep quantities, break-even check, and any calendar flags. The owner adjusts two prep batches and occasionally shifts one cook from 11am to 10am on flagged Fridays. No daily dashboard — same push-not-pull pattern as the Hindkesari pilot.

QSR pilots differ from casual dining: aggregator share means refund drag and ad-spend volatility aren't in the forecast yet. The 91% number is on fulfilled orders — plan a 3-5% buffer on top for aggregator cancellations.

See the product tour

Try the free break-even calculator →

More for you

Pune QSR pilot: 30-day forecast accuracy at 91% dish-level | Forkcast