Ops & Supply

How to forecast product demand as Restaurant and Hospitality Founders

Ops & SupplyFor Restaurant and Hospitality Founders2 apps11 steps~22 min to set up

You're guessing how much salmon to order on Thursday based on what sold last Saturday — but last Saturday was a private buyout and this Saturday is a rainstorm. Your POS (Square or Toast) has the sales history, your reservation system (OpenTable or Resy) has the cover counts, and your supplier invoices are in your email or a MarginEdge spreadsheet. None of it connects. You over-order on a slow week and write off $400 in produce, or you 86 your most-ordered entrée at 7pm on a Friday because you under-prepped. The forecast is whatever your chef de cuisine remembers, plus a gut check on the weather.

Build this on Starch → See the apps used

Ops & SupplyFor Restaurant and Hospitality Founders2 apps11 steps~22 min to set up

Outcome

What you'll set up

A demand forecast that pulls actual cover counts from your reservation system and daily sales from your POS, then tells you what to prep and order for the next 7 days — before you write the order guide.

An alert that fires when a forecasted high-volume night (private event, holiday weekend, local convention) means you need to adjust your standing weekly order by more than 20%.

A variance report that compares what you prepped, what you sold, and what you threw away — every week, automatically, so you can see where your food cost is actually bleeding.

The Starch recipe

Apps, data, and prompts

The combination of Starch apps, the data sources they pull from, and the prompts you use to drive them.

Apps used

Scenario Analysis Transaction Insights

Data sources & config

Starch connects directly to Square or Toast from the integration catalog — the agent queries your POS live when the forecast runs. Starch automates your OpenTable or Resy reservation data through your browser — no API needed — to pull upcoming cover counts into the same surface. If you use Plaid, Starch syncs your bank transaction data on a schedule so you can cross-check deposits against POS totals and spot discrepancies before your bookkeeper does.

Prompts to copy

Connect my Square sales data and pull the last 90 days of daily revenue by menu category — breakfast, lunch, dinner, and bar — broken down by day of week. Overlay my OpenTable reservation counts for the same period so I can see how covers track against revenue per shift.

Build me a weekly demand forecast that looks at the last 12 weeks of sales by menu category, flags upcoming weeks where I have large reservation blocks on the books, and outputs an order quantity recommendation for my top 10 perishable ingredients — proteins, dairy, and produce — with a buffer for walk-in cover variance of plus or minus 15%.

Show me a food cost variance report: what I ordered each week, what I sold (from Square), and the difference. Flag any week where variance exceeded 8% of projected food cost.

Run these in Starch → or paste them into your favorite agent

Walkthrough

Step-by-step

1 Connect Square or Toast from Starch's integration catalog. The agent queries your daily sales history live — itemized by menu category and shift — every time the forecast app runs.

2 Set up browser automation to pull your next 14 days of reservations from OpenTable or Resy. Starch reads the cover count, party size, and any private-event flags directly from your account — no API required.

3 If you use Plaid for your business bank account, Starch syncs your transaction data on a schedule so you can reconcile POS revenue against your actual deposit within the same dashboard.

4 Tell Starch: 'Look at the last 90 days of daily sales by category and cover count. Identify my average revenue-per-cover by shift, and flag the 10 days that were outliers — high or low — so I can tag what caused them (event, weather, 86'd item).'

5 Tag your outlier days with a reason — private event, holiday, bad weather, supply issue. Starch stores those tags so future forecasts can weight those days differently instead of dragging down your baseline.

6 Tell Starch: 'Build a 7-day demand forecast using the last 12 weeks of sales, weighted to exclude tagged outlier days, and adjusted for the cover count I have on the books for next week.' Review the output against your standing order guide.

7 Set up an alert: 'Any time next week's projected covers are more than 20% above my 4-week average for that day, send me a Slack message with the specific night and the ingredient categories most likely to be under-ordered.'

8 Each Friday, run the order variance report: what you ordered Monday through Thursday, what the POS shows you actually sold and prepped through, and the gap. Starch calculates implied waste as the residual.

9 Use the variance data to tune your order multiples. If you consistently over-order salmon by 2 portions per 10 covers, Starch can adjust the recommendation formula — you tell it: 'Reduce salmon forecast by 18% from baseline and see if variance improves over the next 4 weeks.'

10 Before your weekly supplier call, tell Starch: 'Summarize next week's forecast, highlight any ingredients where my order will be more than 25% above my trailing average, and flag items where I had spoilage last week.' Use that summary as your order brief.

11 At month-end, run a 30-day report comparing forecasted demand against actual sales and actual food cost from your POS and bank data. Share it with your chef and use it as the starting point for menu engineering decisions — which items are high-waste, which are consistently under-prepped.

See this running on Starch

Connect your tools, describe what you want, and the agent builds it. Closed beta is free.

Try it on Starch →

Worked example

Valentine's Week 2026 — 68-seat Italian restaurant, NYC

Sample numbers from a real run

Forecasted covers (Feb 10–16)	620
Actual covers (from OpenTable)	683
Projected protein order — salmon, short rib, chicken	2,100
Actual protein cost (supplier invoices)	2,340
Implied over-order variance	240
Waste flagged by variance report	190

Valentine's week is the hardest week to forecast — you have two slammed nights (Friday the 13th and Saturday the 14th), a slow Sunday, and a dead Monday. The owner ran the demand forecast Monday morning: Starch pulled 90 days of Square sales, overlaid OpenTable's confirmed reservations (683 covers across the week, versus a 12-week average of 480), and recommended a protein order 38% above the weekly baseline. The owner trimmed the short rib order slightly based on a menu change, placed the order, and ran the week. Actual protein cost came in at $2,340 against a $2,100 forecast — a $240 overage, mostly on chicken thighs over-ordered for a staff meal program. The Friday variance report caught it. The next week, the forecast for chicken was adjusted down by one case. Prior to Starch, the same owner was working from a handwritten order guide updated every Sunday night and had written off $400–600 in produce spoilage every week during slow periods because the forecast was just 'same as last week.'

Measurement

How you'll know it's working

Food cost percentage by week — target under 28% for a full-service restaurant

Prep waste as a percentage of ordered protein and produce — flagged when above 5%

Forecast accuracy: projected covers vs. actual covers per shift, tracked as a rolling 4-week average

86'd item frequency — how often you run out of a menu item before close, as a proxy for under-forecasting

Order variance vs. forecast — actual supplier invoice total vs. Starch's recommended order value

Comparison

What this replaces

The other ways teams handle this today, and how the Starch version compares.

MarginEdge

MarginEdge is excellent at tracking food cost against invoices, but it doesn't generate a forward-looking demand forecast from your reservation data — you still have to estimate your order quantities manually.

Toast built-in reports

Toast gives you historical sales data well, but it has no way to pull in your upcoming cover counts from OpenTable or Resy and produce a prep or order recommendation from the combined data.

Google Sheets or Excel

A spreadsheet can hold the formula, but you're updating it manually every week — it doesn't pull live from your POS or reservation system, so the forecast is only as current as your last copy-paste.

Craftable or Sysco's ordering tools

Sysco's order management tools are tied to their own inventory catalog and don't know your actual sales velocity from your POS or how many covers you have booked next Saturday night.

On Starch RECOMMENDED

One platform — scenario planning, transaction insights all running on connected data. Setup in plain English; numbers stay current via scheduled syncs and live agent queries.

Try it on Starch →

FAQ

Frequently asked questions

Does Starch connect directly to Toast, or only Square?

Both Square and Toast are reachable from Starch's integration catalog — the agent queries your POS data live when the forecast runs. If your POS isn't in the catalog, Starch can automate it through your browser as long as you can log in and navigate to a sales report.

My reservation system is Resy, not OpenTable. Does that work?

Yes. Starch automates Resy through your browser — no API needed. It logs in, reads your upcoming reservation counts, and pulls that data into the same forecast surface as your POS sales history. Same approach works for OpenTable, Tock, or any other web-based reservation system.

Is the Demand Planner app available right now?

The dedicated Demand Planner app is currently in development — you can request beta access to get notified when it launches. In the meantime, you can build a working demand forecast today using Starch's Scenario Analysis app (which connects to your sales data) combined with a custom app you describe in natural language. Most restaurant operators can get a functional version running within a session.

Will Starch store my historical sales data, or does it only look at things live?

If your POS is connected through Starch's integration catalog, the agent queries it live each time your forecast runs — it doesn't archive a copy. If you connect Plaid to sync your bank transactions, that data is stored in Starch on a schedule. For most demand forecasting purposes, live queries against 90 days of POS history are enough. If you need a long-term data warehouse, that's outside what Starch does today.

Is Starch SOC 2 certified? I'd be connecting my POS and bank account.

Not yet — Starch is not currently SOC 2 Type II certified. That's worth knowing before you connect financial data. It's on the roadmap, and the team is transparent about where things stand.

My chef writes the order guide, not me. How does this fit into that workflow?

The forecast output can be framed however fits your kitchen — Starch can generate a prep sheet summary or an ingredient order brief that your chef reviews and adjusts. You'd tell Starch: 'Every Thursday at 8am, run next week's demand forecast and Slack my chef a summary of projected covers by night and the top 10 perishable items I should order above baseline.' The chef still owns the call; Starch just replaces the blank-page starting point.