Toast IQ can answer plain-language questions from the sales and labour data held in your Toast account: items sold, sales against earlier days, labour hours and spend by day or shift, voids and discounts. It cannot see costs outside Toast, explain why a number moved, or be trusted unchecked; Toast itself says AI can make mistakes.
Toast's product page says the assistant is included with Toast POS at no additional cost. IQ Team and IQ Grow are separate offers, and that page shows no price for either. What follows is about the included assistant and its agents, and about the gap between a clean-looking answer and a number you can act on.
What Toast says IQ can do with sales and labour data
The assistant sits at the top left of the navigation menu in Toast Web and at the bottom of the Toast Now app. You need either the 4.1 Sales Reports or the 4.3 Labor Reports permission to use it, and what it can show or change also follows your Toast Web permissions. Mid-Market and Enterprise accounts have to go through their Customer Success Manager to get access. Toast says it sends the relevant data to a third-party language-model provider through a dedicated enterprise-only instance, that third parties do not use it for general training, and that your data is never mixed with other customers'.
| Task | Examples Toast gives | What sits behind it |
|---|---|---|
| Sales questions | How many large drinks were sold this week; what changed in sales this week; which add-ons sell best with a given item | Sales data in your Toast account |
| Labour questions | Which days have the highest labour spend in the past month; correcting an employee's shift end after an auto clock-out | Time-clock data; shift edits need permission and a confirmation |
| Voids and comps | Toast's FAQ lists discounts and voids among the areas you can ask about, without a sample prompt | Void and discount records |
| Menu and kitchen | Suggest menu modifications to simplify operations; view or edit prep time, course or station | Your menu set-up; changes need confirmation |
| Marketing | Draft an email or text campaign | Drafts only; you review and send them |
Two statements matter more than the examples. Toast says IQ makes no changes without explicit confirmation, and that it "cannot be fully disabled or hidden account-wide". The first is reassuring; the second means permissions are the only dial you have, a point I come back to below.
Eight owner prompts, and what to check in each answer
These are prompts I would try in the first week. Some echo wording Toast shows and some are my own. The right-hand column is the part Toast's pages leave to you.
| Prompt | A usable answer includes | Check against |
|---|---|---|
| "What were net sales yesterday, and how do they compare with the last four same weekdays?" | One figure per day and the definition of net | Your sales report for the same business day |
| "Break down last week's sales by daypart and payment type." | Dayparts as you have set them up in Toast | The totals: do the parts add up to the whole? |
| "Which five items gained the most units in 28 days, and which lost the most? Say whether modifiers are counted." | Unit counts and how sizes were treated | The product mix report |
| "Which days last month had the highest labour cost as a percentage of net sales? State the denominator and what counts as labour cost." | A table of days, the formula in words | Your payroll summary |
| "List every auto clock-out this week and the hours it added." | Employee, shift, extra hours | Timecards, before payroll closes |
| "Which shifts had sales per labour hour below $45?" | Shifts and the sales and hours behind each ratio | Two shifts rebuilt by hand |
| "Which servers had the most voids and discounts this week, with times?" | A list with times and any manager comments | The tickets themselves |
| "Create a prep list for Saturday from the last four Saturdays." | Quantities by item and station | Bookings, events and weather you know about |
The $45 in the sixth prompt is a placeholder; use your own target. Prompts that ask the assistant to state its definition do the most work, because they expose the next section's problems before you act on a number.
Why "how are we doing on labour?" gets a weak answer
A vague prompt gets a vague answer, and the assistant will choose a period, a measure and a definition for you. A usable prompt names four things: the period, the measure, the definition and the shape of the reply.
Before
How are we doing on labour?
After
For the four weeks ending last Sunday, show labour cost as a percentage of net sales, one row per week,
with columns week, net sales, labour cost and percentage. Say which sales figure you used and whether
labour includes overtime premium and payroll taxes. Then name the two days with the highest percentage
and the hours that drove them.An illustrative reply: week one, $28,900 net sales and $9,010 labour, 31.2%; week two, $29,500 and $9,184, 31.1%; week three, $27,300 and $9,040, 33.1%; week four, $30,200 and $9,150, 30.3%. It names Tuesday of week three (sales $1,650, labour $677, 41.0%) and Monday of the same week (sales $1,420, labour $540, 38.0%), and notes that the Tuesday labour includes a six-hour training shift.
The reply is useful and still needs work from you. Whether training belongs in labour cost is your decision, and it moved a whole day's percentage here. Check that "last Sunday" ended the week where your payroll week ends. Then ask a follow-up for the hours behind the Monday figure before you cut anyone's shifts, because a quiet Monday with the minimum crew will always look expensive as a percentage.
Labour percentage: three numbers for one week
An illustrative week at a 55-seat restaurant shows the spread. Gross sales are $31,400. Discounts, comps and refunds total $1,900, so net sales are $29,500. Wages come to $9,440 in the time-clock data. Three auto clock-outs, where staff never clocked out and the system closed the shift, added 4.5, 6 and 5.5 hours at an average $16 an hour, which is $256 of pay nobody worked.
| Version | Labour | Divided by | Percentage |
|---|---|---|---|
| As first shown, against gross sales | $9,440 | $31,400 | 30.1% |
| As first shown, against net sales | $9,440 | $29,500 | 32.0% |
| Clock-outs corrected, against net sales | $9,184 | $29,500 | 31.1% |
| Corrected, plus payroll costs at an assumed 12% | $10,286 | $29,500 | 34.9% |
The same week reads anywhere from 30.1% to 34.9%, and no version is a mistake. They answer different questions. Toast's own labour-cost guide gives the formula as total labour cost divided by total revenue, while other guides argue over gross versus net sales, so ask the assistant which one it used and keep it the same from week to week. Guides also describe labour cost as wages plus benefits and payroll taxes. A POS can only add what it holds, which is usually pay rates and hours, so its percentage will sit below the one your accountant produces. The 12% here is an assumption; use your payroll figure.
The reason to be fussy is money. One percentage point on $29,500 of weekly net sales is $295, or $15,340 over a year, so a rota cut made to cure what is really a measurement problem can cost more in lost service than it saves.
Fix the clock-outs first. Toast has a built-in agent for exactly this, described below, and the chat can correct a shift end time if your permissions allow. A corrected week moves the figure by 0.9 points before any debate about denominators. For a way to test rota changes against the percentage once it is trustworthy, see Restaurant Labour Cost Percentage: Test Rota Changes With AI First.
Where the answer stops: what Toast IQ cannot see
The assistant draws on data inside Toast. Anything that never reaches Toast does not exist for it, and Toast's pages do not list those gaps. These are the ones to test in your own account.
- Costs. Food purchases, supplier invoices, rent and utilities sit in other systems. Sales and labour alone cannot say which shift or dish made money. If you want prime cost, which is food cost plus labour, you need the invoice side. A cost spreadsheet is covered in Recipe Costing Spreadsheet for Restaurants: Build It With ChatGPT.
- Sales taken elsewhere. Orders on a delivery marketplace tablet that does not post into Toast, deposits taken in a booking system and cash handled off-system are invisible. Ask what feeds your totals before you compare a month with last year.
- Rotas kept in another tool. If your schedule lives in a separate rota app, the assistant sees hours worked, not hours planned, so it cannot tell you the rota was wrong.
- Causes. It can show that Tuesdays fell 14%. It cannot know a road closure, a competitor's promotion or a staff change. Toast's feed does surface local events, but treat any reason it offers as a hypothesis.
- How fresh the data is. Toast's pages do not state it. Ask about the last hour, then compare with the live sales screen.
- Quality of your set-up. Wrong job roles make labour by role wrong. Items categorised inconsistently make item rankings wrong. The assistant reads what you configured.
Sales counts depend on how the menu was built
Toast's own example is "how many large drinks did we sell this week". Picture a café-bar where an iced latte has a Large modifier, but the mocha and the cold brew were set up years ago as separate items called "Large iced mocha" and "Large cold brew". The assistant counts the modifier and answers 260. The product mix report, which can show modifiers as their own level beneath each item, shows another 152 sold as separate items, so the true figure is 412.
Nobody made an arithmetic error. The question meant "large drinks" and the data held "the Large modifier". The fix is on your side: standardise how sizes are built, and until you do, ask for "all items whose name or modifier includes Large", then compare with the product mix report. That report lives under Reports, Menus, Product mix, exports to CSV or Excel and compares periods, but it does not email itself on a schedule.
The agents: hourly alerts and daily digests
Agents are recurring jobs rather than chats. Each checks your Toast data on a schedule and sends a short update by push notification or email. Digest agents always send; alert agents usually check hourly and send only when a condition is met. Toast says they are in early access with selected customers, two per user per location are included at present, and push notifications reach only the person who set an agent up. They do not change anything in Toast. The ones that matter for sales and labour, with Toast's default schedules:
| Agent | Type and default time | What it sends |
|---|---|---|
| Clock-out catcher | Digest, daily 8 a.m. | Yesterday's auto clock-outs, before payroll closes |
| Morning coffee briefing | Digest, daily 8 a.m. | Yesterday's results, year-on-year comparison, one action for today |
Labor running hot | Alert, hourly | Mid-shift labour above target |
| Overtime radar | Digest, daily 8 a.m. | Staff on track for weekly overtime |
| Missed break check | Digest, daily 8 a.m. | Missed meal breaks, before payroll closes |
| Sales pace check | Alert, daily 2 p.m. | Pace against a normal day |
| End-of-day closeout | Digest, daily 10 p.m. | Sales, labour, discounts, top items |
| Loss prevention alert | Alert, hourly | Unusual voids, discounts or refunds |
An alert is a number without context. At 3 p.m. the labour-running-hot alert might report sales of $1,900 and labour of $760 so far, which is 40% against a 30% target. Sending two people home would leave the dinner service thin, because the prep crew arrives at 3:30 and a day like this ends at about $4,800 of sales on $1,440 of labour, exactly 30%. You know the shape of your day; the agent does not. Use alerts to prompt a look, not to act.
The overtime digest is the opposite case, where the number is usable straight away. Suppose a server has 33.5 hours by Thursday and two six-hour shifts still rostered. That is 45.5 hours. If overtime starts at 40 hours and costs time and a half (assumptions; use your own rules), at a $15 base the 5.5 overtime hours cost $41.25 extra. Moving one shift costs nothing. A digest like this can repay the time spent setting it up within a week or two.
Voids and discounts: a pattern is not a reason
The loss-prevention alert and the voids prompt both produce lists, and lists invite conclusions. Suppose the assistant reports that one server had 14 voids across five shifts and the next highest had six. That is a pattern worth a conversation. It is not evidence of anything. The assistant cannot tell a void caused by a guest changing their mind from one caused by a mis-keyed modifier, a training error or something worse, and Toast's pages do not claim it can.
Pull the tickets and any manager comments for the 14, and see who was working the bar, who was newest and whether the voids cluster at one time of night. Then talk to the person with the list in front of you and let them explain. If the explanation does not hold up, that is a staff matter for you and your adviser, not a number to show the whole team. Ask for the void reasons and any comments in the prompt, because whatever explanation staff or a manager recorded at the time will be there.
A five-minute check before you act on any figure
- Ask for the definition. "Which sales figure did you use, and what does labour cost include?" Write the answer on the page where you track it.
- Match one number to a report. Pick the biggest figure in the answer and find it in the matching Toast report for the same business day.
- Rebuild two rows by hand. Two shifts or two items from timecards or tickets. If both match, the rest probably does too.
- Check the date range. Late-night sales can land on the previous business day in some systems, and a "last week" can start on a different weekday than yours.
- Note the mismatch and report it. Toast lets you send feedback on an answer from inside the assistant. A list of recurring mismatches also shows you which parts of your set-up need tidying.
A ten-minute Monday log you can copy
Keep one page, on paper or in a shared sheet, with five columns: the question, what the assistant said, what the report said, whether they matched, and a note. An illustrative first month looks like this.
| Question | Assistant said | Report said | Match | Note |
|---|---|---|---|---|
| Net sales, last Saturday | $4,312 | $4,312 | Yes | Both net of comps |
| Large drinks, last week | 260 | 412 | No | Separate "Large" items; standardise sizes |
| Labour percentage, last week | 32.0% | 31.1% | No | Three auto clock-outs worth $256 |
| Top five items | Same order | Same order | Yes | Modifiers excluded in both |
When a question type has matched three weeks running, ease off to a fortnightly check for that type and keep the weekly one for the types that failed. After a month the log doubles as a short list of set-up fixes: here, the size modifiers and the clock-out habit.
Read Toast's own posts about IQ with the same caution. Usage figures and customer results that a vendor publishes about its own product are self-reported, so test any claim that matters to you against your own log before you rely on it.
Permissions, privacy and the switch you do not have
Because the assistant cannot be hidden account-wide, tidy the permissions. Print the permission sets for managers, supervisors and any shared logins and note who holds 4.1 Sales Reports or 4.3 Labor Reports. Remove them where the person has no need for sales or labour numbers. Then test: create or borrow a login holding only the labour permission and ask a sales question, then the reverse, and see what comes back. Ask a question about an employee's pay, too, since Toast's pages do not say how pay data is shown. Do not assume; test it.
Staff names and pay rates are personal data, and Toast says relevant data goes to a third-party language-model provider. If you have a staff privacy notice, check that it covers this, and ask your adviser if unsure. The reply to a customer or staff member who asks is easier when you have read the AI terms on your account rather than relied on a summary.
When to export instead of asking
Use the assistant for questions answered by Toast's own data. Use an export when the answer needs data from elsewhere. Prime cost is the classic: net sales from Toast of $29,500, food purchases of $8,260 from supplier invoices and corrected labour of $9,184 give prime cost of $17,444, which is 59.1% of net sales. Purchases stand in for cost of goods when you do not count stock, so treat the percentage as approximate. A general assistant can compute this from a spreadsheet, provided the export carries no employee names.
The choice between a POS-built assistant, a spreadsheet and another vendor's tool is covered in AI Food Insights Tool: Toast IQ, Square AI or a Spreadsheet?. Whether IQ Team or IQ Grow is worth paying for, once you know what the included assistant already does, is the question in Toast IQ Team or Toast IQ Grow: Which Suits a Small Restaurant?. For turning the numbers into talking points for staff, Restaurant Pre-Shift Meeting Notes: Draft Them From POS Data With AI uses the same data.
Toast IQ questions owners ask
Does Toast IQ cost extra?
Toast's product page says the assistant is included with Toast POS at no additional cost, while IQ Team and IQ Grow are separate offers with no price on that page. The agents, which are in early access with selected customers, include two per user per location at present. Ask your Toast representative to confirm what your account includes, in writing.
Can I switch Toast IQ off for my staff?
Not fully. Toast says Toast IQ cannot be fully disabled or hidden account-wide. Access depends on the sales-reports and labour-reports permissions (numbered 4.1 and 4.3), so the control you have is removing those from permission sets that do not need them. Test with a login that holds only one of them before you decide who keeps access.
Will Toast IQ change my menu or rota by itself?
Toast says it makes no changes without your explicit confirmation, and what it can change is limited by the user's Toast Web permissions. Campaign drafts are not sent for you. The agents send updates and recommendations but do not change anything in Toast. Treat every proposed change as a draft you read before confirming.
Is my restaurant's data used to train AI models?
Toast says it sends the relevant data to a third-party language-model provider through a dedicated enterprise-only instance, that third parties do not use it for general training, and that it is never mixed with other customers' data. Those are Toast's statements, so read the AI terms on your own account and ask what staff data the assistant can see.
How accurate are Toast IQ's numbers?
Toast's own pages say AI tools can make mistakes and tell you to check important information. In practice the common errors are definitions rather than arithmetic: which sales figure was the denominator, whether a size is a modifier or a separate item, and auto clock-outs still sitting in the hours. Compare one figure with the matching Toast report each time.
Further reads
- AI Software for Restaurants: POS-Built AI or an Add-On Tool? — Whether POS-built AI beats an add-on tool for your restaurant.
- Restaurant Menu Price Increase: Test Each Change by Dish With AI — Use your sales data to test a menu price change dish by dish.
- How to Build a Restaurant Staff Rota With AI Demand Forecasts — Turn demand forecasts into a rota after you trust the numbers.
- Restaurant AI Tools: 12 Questions to Ask Before You Sign Up — Twelve questions to put to any restaurant AI vendor.
- How to Buy AI Software for a Shop or Restaurant: Pre-Purchase Checks — Pre-purchase checks before paying for restaurant AI software.
- Free AI Shift Scheduling for Restaurants: Where Free Ends — What free AI scheduling covers and where it stops.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Toast IQ product page and FAQ; Toast support articles (Toast IQ Overview, Toast IQ Intelligence Ecosystem, Toast IQ Agents Overview, Product Mix report overview); Toast's guide to labour-cost percentage. Figures in the worked examples are illustrative.