AI demand forecasting for restaurants: what it predicts, what it costs and how to check it

AI demand forecasting tools predict your sales, guests or orders by day and by hour from your POS history, then turn that forecast into a schedule, a prep list or a supplier order. Lineup.ai prints $79 a location a month for forecasts alone; most other vendors quote. Before you rely on one, check its error against last week's numbers for four weeks.

Our AI scheduling guide covers scheduling apps and fair workweek laws, and the AI tools shortlist names one tool for each job. This page is about the forecast underneath them: what goes in, what comes out, who sells it, and how to tell whether it beats the guess you make today.

What does a forecasting tool predict?

The vendors below describe six outputs. Most tools do two or three of them.

Output	How a vendor describes it	What you use it for
Sales by day and hour	Restaurant365 shows forecasts by week, day, hour, and 30- or 15-minute blocks	Staffing each hour
Guests or checks	Crunchtime forecasts by revenue, guest count or number of checks, and splits dine-in from takeout	Covers per server, counter staff at lunch
Item counts	Lineup.ai lists item-level forecasting	How much of each dish to make
Prep amounts	Crunchtime suggests prep in 15-minute blocks; ClearCOGS sends a daily prep sheet down to the ingredient	The morning prep list
Labor	7shifts predicts staffing needs from its sales forecast; Toast shows projected labor cost for each shift as you schedule	The weekly schedule
Supplier orders	MarketMan's Smart Ordering uses predictive analytics to forecast demand; Restaurant365 lists AI-driven purchase orders based on demand	What to order and when

What data does a forecast learn from?

Every tool starts with your POS sales history. Crunchtime says its engine analyzes over 400 days of sales, and its FAQ lists recent sales, last year's sales for the same day, guest counts, checks, transaction counts, and dine-in or takeout as inputs. Lineup.ai says it adds weather, events, traffic and holidays, and builds a separate model for each location.

Three gaps are worth knowing before you connect anything.

• A sellout hides demand. If you ran out of short rib at 7:30 last Friday, the sales history shows demand stopping at 7:30. Log each 86 with the time so you can correct for it.
• The tool cannot see what you know. A private party, a road closure or a new menu will not appear in last year's sales. Enter known events before the week starts.
• A new restaurant has little to learn from. With only a few months of history, a year-ago comparison does not exist yet, so expect wider misses at first.

Which vendors sell forecasting, and what do they charge?

Prices below are what each vendor prints. Where a page shows no price, the table says so.

Vendor	What its own page says	Printed price	Start here if
Lineup.ai	Hourly, daily and weekly sales forecasts, labor forecasts and item-level forecasts; forecasts work without its scheduler	Lineup.ai: Forecasts Only $79 per location a month; Forecasts + Scheduling $149, on a monthly commitment (annual commitment is 10% off)	You want a forecast and plan to keep your current schedule app
7shifts	Forecasts sales from past sales and labor data to predict staffing needs	7shifts Premium, whose card lists an advanced hourly labor forecast: $134.99 per month per location, billed yearly	You already build schedules in 7shifts
Toast Scheduling	"More accurate and automated sales forecasting" and projected labor cost for each shift	No separate price; Toast's pricing page lists Toast Scheduling Pro in its Team Management Essentials suite	You run Toast
MarketMan	Smart Ordering forecasts demand to suggest orders	MarketMan Growth, which adds Smart Ordering: $299 a month; Starter at $249 does not include it	Ordering is a bigger problem than staffing
Crunchtime	Forecasts by revenue, guests or checks in 15-minute blocks, with suggested prep and ordering	No price on its pages; sold by demo	You run several stores and want prep and orders from one forecast
Restaurant365	Forecasts from week down to 15 minutes, tied to scheduling and payroll	Custom quote	You want accounting, inventory and labor in one system
ClearCOGS	Food prep, labor and menu item forecasting from POS data; lists a Toast integration	No price printed	Prep waste is your biggest cost
Square	Its October 2025 release describes Square AI answering questions about item sales, staffing and labor costs; it describes no forecast feature	Part of Square; no separate price in the release	You run Square and want answers from your own sales first

Sources: Lineup.ai's pricing page and homepage; 7shifts' pricing and scheduling pages; Toast's scheduling page and pricing page; MarketMan's pricing and purchasing pages; Crunchtime's forecasting page; Restaurant365's sales forecasting page; ClearCOGS; and Square's October 2025 release.

How much should you trust a vendor's accuracy figure?

Treat it as a question to ask. Crunchtime's page says its AI predicts demand "with up to 99% accuracy." Lineup.ai's homepage says it is "35% more accurate than traditional forecasting methods." Neither figure comes with the measure behind it, the level it was checked at (a whole day or a 15-minute block), or what it was compared against.

Ask each vendor three things: which error measure they use, whether it is checked by day or by hour, and whether they will run it on your own last 90 days. Lineup.ai's pricing FAQ offers to prove its forecast accuracy on a trial or refund the trial period, so ask how that proof is measured before you start.

How do you measure forecast error yourself?

You need one number you can work out on a spreadsheet each week. For each day, take the gap between the forecast and actual sales and ignore whether it was over or under. Add the gaps for the week and divide by the week's actual sales. Lower is better.

Then compare it with a free forecast: the same day last week. A paid tool should beat that. The figures below are made up for one week at one restaurant.

Day	Actual sales	AI forecast	Gap	Same day last week	Gap
Mon	Mon actual: $2,100	Mon AI: $2,000	Mon AI gap: $100	Mon last week: $1,900	Mon last-week gap: $200
Tue	Tue actual: $2,300	Tue AI: $2,400	Tue AI gap: $100	Tue last week: $2,500	Tue last-week gap: $200
Wed	Wed actual: $2,600	Wed AI: $2,500	Wed AI gap: $100	Wed last week: $2,300	Wed last-week gap: $300
Thu	Thu actual: $3,100	Thu AI: $3,300	Thu AI gap: $200	Thu last week: $2,900	Thu last-week gap: $200
Fri	Fri actual: $4,800	Fri AI: $4,500	Fri AI gap: $300	Fri last week: $5,300	Fri last-week gap: $500
Sat	Sat actual: $5,400	Sat AI: $5,600	Sat AI gap: $200	Sat last week: $4,900	Sat last-week gap: $500
Sun	Sun actual: $3,600	Sun AI: $3,400	Sun AI gap: $200	Sun last week: $3,900	Sun last-week gap: $300
Week	Week actual: $23,900	Week AI forecast: $23,700	AI gaps: $1,200	Week last-week total: $23,700	Last-week gaps: $2,200

In this made-up week the AI forecast misses by $1,200 ÷ $23,900, or 5.0% of sales. Last week's numbers miss by $2,200 ÷ $23,900, or 9.2%. The tool wins this week. Both forecasts add up to $23,700 for the week, so only the daily gaps separate them. Both ran $200 under the $23,900 actual, or 0.8% of sales. A forecast that runs low week after week leaves you short-staffed.

Check your busiest hour the same way, because a good day can hide a bad hour. If the tool forecast 42 covers for 6 to 7 p.m. on Friday and 55 came in, it was 13 covers short, or 24% of that hour. That miss is the one your line feels.

How do you run a four-week trial?

1. Before you start: export at least a year of POS sales by hour if you have it, and list next month's known events. Keep your own forecast method as it is.
2. Weeks 1 and 2: run the tool alongside your usual forecast and change nothing. Each Monday, record your forecast, the tool's forecast, last week's numbers and actual sales for every day.
3. Weeks 3 and 4: let the tool drive one decision, such as prep for your two biggest sellers or the Friday schedule. Log every 86, with its time, and the waste on those items.
4. Every week: work out the error for all three forecasts by day and for your busiest hour, and note whether each ran high or low.
5. At the end: keep the tool if it beats both your own forecast and last week's numbers on days and on the peak hour. Stay on monthly billing until then; our vendor question list covers data export and exit terms.

A forecast learns only from orders that reach your POS. Calls that ring out during the Friday peak leave no sale in the history, so the tool cannot see that demand. Maple Pro takes phone orders into supported POS systems at $350 a month billed monthly or $220 billed yearly; see Maple's plans and the call-to-kitchen trial.

Published by Maple. This AI-assisted guide combines vendors' own product, pricing and FAQ pages and Square's October 2025 release with an original output table, a made-up worked week and a trial plan. It does not report a test of any forecasting tool, does not verify any vendor's accuracy figure, and does not promise any saving.