There is no single best AI for a small food business. A general assistant such as ChatGPT or Claude (about $20 a month) suits costing, menu copy and supplier emails; the AI in your till suits sales questions; a phone service suits missed orders. Pick by job, starting with your biggest time drain.
Results for this keyword are mostly industry-scale trend pieces about robots and forecasting at chain level. A cafe, a caterer and a takeaway with one till and four staff face different questions: what does this dish really cost, who answers the phone during the lunch rush and where do the allergen records live. These six picks are ordered by how quickly they usually pay back.
Six picks at a glance, and who each one is for
| Pick | Best for | Typical monthly cost | Main caution |
|---|---|---|---|
| 1. A general assistant for costing | All three | $20 (free plan to test) | It does the sums you give it, not the prices you forgot |
| 2. The same assistant for menu copy and notices | Cafes, takeaways | Same $20 | Never let it describe ingredients you did not state |
| 3. Allergen records formatted by AI, checked by a person | All three | Same $20 | Not a source of truth |
| 4. Supplier order drafts from sales data | Cafes, takeaways | Same $20 | Check quantities against stock on hand |
| 5. AI inside the till (Toast IQ, Square AI) | Those already on that till | Often included | Beta features can change |
| 6. AI phone answering | Busy takeaways, caterers taking enquiries | Varies by call volume | Count missed calls first |
Notice that four of six use the same $20 assistant. That is the point of the "no single best tool" answer: the assistant is a general-purpose tool and the other two are narrow ones that win on their own turf.
Pick 1: Costing a dish, and keeping it costed
This is the first job to hand to an assistant, because food cost drifts and nobody has time to redo the sums. The assistant is good at arithmetic when you give it the inputs. The main worked example: a thirty-seat cafe sells a flatbread sandwich at $9.50. Ingredients per portion total $2.85, so food cost is 30% of the price. Then the cheese supplier raises prices by 12%. The cheese portion was $0.90 and is now $1.01, so the dish costs $2.96, and food cost becomes 31.2%. The gross margin per sandwich falls from $6.65 to $6.54.
That is a small drop on one dish, but a menu of twenty dishes each shifting by a similar amount adds up. Paste your recipe sheet and last week's invoice prices into the assistant and use a prompt like this.
Here are my recipes (ingredient, quantity per portion, unit) and the latest supplier prices
(ingredient, price per kg or each). For every dish, give: cost per portion, menu price, food cost %.
Sort by food cost % highest first. Flag dishes over 33%. Show the working for the top three.
If an ingredient has no price, write MISSING rather than estimating.What you get back is a ranked table. What you fix: any row marked MISSING, and unit mistakes, such as a price per kilogram used as a price per gram. Check one dish by hand each time. If you want a deeper treatment of menu choices, AI menu engineering: which dishes to promote, reprice or drop picks up where this leaves off.
Pick 2: Menu copy, notices and the Friday special
Menu descriptions and chalkboard text are low-risk, repeatable writing. Give the assistant the ingredients and a style note, and it produces three versions to choose from. The rule is simple: it may only use ingredients you list. An assistant told to "make it sound delicious" will add "locally sourced" and "house-made" unprompted, which you cannot always support.
Write three short menu descriptions (under 20 words each) for this dish.
Facts only: roast cauliflower, tahini, pomegranate, flatbread, vegan.
No claims about sourcing, freshness or house-made. Plain, warm voice.A reasonable result: "Roast cauliflower on warm flatbread with tahini and pomegranate. Vegan." What you fix: nothing here, and that is the point. For longer voice-matching work, the menu-description tutorial in the further reads goes further.
Pick 3: Allergen records, formatted by AI and checked by a person
For a food business this is the most safety-sensitive record you hold. An assistant is a good formatter but a poor authority. The workflow: collect the label for each ingredient, write the recipe, ask the assistant to build a table with products down the side and allergens across the top, and then have a person confirm each cell against the label and date the check. Whenever a supplier or recipe changes, re-check that row.
A realistic mistake: a takeaway asked a chatbot whether a sauce was free of sesame and got a confident "yes", based on the typical recipe. The supplier's version contained sesame oil. The fix is a rule, not a better prompt: customers are answered from the checked table, never from a chatbot's general knowledge. Should an AI chatbot answer allergen questions for your restaurant? goes into the limits in detail.
Pick 4: Supplier orders from yesterday's sales
Takeaways and cafes reorder the same twenty items. An assistant can turn a sales export and a par-level list into an order draft. Say a takeaway sells about 60 portions of chicken on a Friday and uses 180 g raw per portion: that is 10.8 kg, so with a 15% buffer you order about 12.4 kg, rounded to 13 kg. Give the assistant par levels and current stock, ask for a draft order by supplier, and ask it to list assumptions. Check against the freezer before sending.
Format the output as an email to each supplier with the delivery day and your account number placeholder. The time saved is the daily ten minutes of counting and writing, not the thinking. For the forecasting side, forecasting covers and cutting food waste with AI shows how to include weather and events.
Pick 5: The AI inside your till, and when it beats a general assistant
If you already use Toast or Square, try their built-in AI before anything else, because it sits on your sales data. Toast's own page describes Toast IQ as included with Toast POS at no additional cost, able to answer questions about sales, labour, menu performance and guest trends, and to take actions such as updating items or marking something out of stock with your confirmation. It says it does not use your business data for general model training. Toast also sells separate packages named Toast IQ Grow and Toast IQ Team, so check what your own plan includes before assuming everything is free. Square's page describes Square AI as a beta that lets you ask questions about your business data in plain words, and it can bring in local weather and events.
Use these for questions like "which days have the highest labour cost" or "what sold least last month". For writing, costing and supplier emails, a general assistant still does the work, because the till tool is built around reports and actions in the POS. If you are weighing a till change partly for AI, AI food insights: Toast IQ, Square AI or a spreadsheet compares the options. If you have neither till, a spreadsheet export plus an assistant gets you most of the way.
Pick 6: Phone answering, once you have counted the misses
During a lunch rush the phone rings and nobody picks up. AI phone services answer, take an order or a booking enquiry and pass it on. Pricing varies. As one data point, Smith.ai's AI receptionist lists 25 calls free and then $150 for 75 calls; other providers charge per minute or per call. Before paying, run a two-week count: how many calls did you miss, what share were orders, and what was the average order worth?
Illustrative sum: 12 missed order calls a week at an average of $22 is $264 a week, or about $1,100 a month. Even if a service recovers a third of those, that is a few hundred dollars. If instead you miss two calls a week, the fee probably exceeds the gain. For takeaways, setting up an AI phone line for takeaway orders walks through the build, and the phone-answering cost tutorial in the further reads has the sums.
Five sales questions worth asking, and the action each leads to
Whether you ask the till's AI or paste an export into an assistant, specific questions beat broad ones. Here are five that repay the effort, each with the action that follows.
- "Which five items had the lowest gross profit per portion last month?" Action: reprice, shrink the portion or drop the dish.
- "Which weekdays have labour above 35% of sales?" Action: trim the rota on those days or add a promotion to lift sales.
- "What were our top ten sellers by hour between 11 and 2?" Action: prepare those first for the lunch rush.
- "How many voids and discounts did each shift have?" Action: a coaching chat, not an accusation, since voids have innocent causes.
- "Which items are bought together most often?" Action: a combo with a small saving.
Be careful with conclusions on thin data. A dish that sold twice on a rainy Tuesday has not "failed". Ask the assistant to state how many sales each finding rests on, and ignore anything below about twenty. If the till's AI reports a number you doubt, check it against the standard report once before acting on it.
A phone-answering script that sets limits
If you try an AI phone service, the setup matters more than the product. You write the script of what it may say, and an assistant is a fast way to draft it. Give it your opening hours, address, menu highlights and the rules for what it must pass to a human. A filled-in set of rules (illustrative):
It may: give opening hours, say we take takeaway orders until 9pm,
take a name, number and order for items on the standard menu, and quote
a ready time of 20 minutes (30 on Fridays after 6pm).
It must pass to a person (take name and number, say we will call within 15 minutes):
any allergy or dietary question, any order over 10 items, any complaint, any catering enquiry.
It must never: state ingredients, promise a refund, or confirm a booking for more than 8 people.Test it by ringing it yourself with ten awkward calls: a nut allergy question, a large order, a mumbled name, a request for a half-price deal. Fix the rules each time it slips, and keep a printed log of the first fifty real calls for the first week. If the log shows two or more errors in fifty calls, the service is not ready for your counter yet.
Writing the staff notices and daily brief
The assistant earns a second use as the writer of internal notes. Paste your day's bookings, specials and deliveries, and ask for a one-page brief: what is on special and its allergens (from your table), which suppliers deliver and when, and any large bookings. A before and after (illustrative):
Before: "Fri: 12 booked 1pm, soup = leek+potato (cream), fish delivery late, Sam off"
After:
Friday brief
- 1pm: table of 12. Pre-order sheet to the kitchen by 11:30.
- Soup: leek and potato (contains milk). Check the allergen table before serving.
- Fish delivery is expected late: prepare the vegetarian special as the first push.
- Sam is off. Cover the counter with a second person from 12 to 2.What you fix: the assistant should not state allergens it has not been given, which is why the brief points to the table. A paragraph like this takes a minute to read aloud at the start of a shift.
Cost summaries for three sizes of business
Illustrative monthly costs, using list prices and the picks above.
| Business | Picks used | Monthly cost |
|---|---|---|
| One-person coffee cart | Free assistant for menu copy and costing; no till AI | $0 |
| Thirty-seat cafe on a till with AI | Paid assistant ($20); till AI included | About $20 |
| Busy takeaway with phone rush | Paid assistant ($20) plus a phone service (varies, for example $150 for a 75-call block) | About $170 or more in busy months |
The last row shows why the missed-call count matters: the phone service can cost many times the assistant. A takeaway that loses over $1,000 a month in missed orders may justify it; a cafe that misses two calls a week will not.
Caterers: quotes, menus and the proposal pack
A caterer's week is dominated by quotes. The assistant drafts a proposal from a short brief. Example: 40 guests, a lunch buffet, two vegetarian guests and one nut allergy. Ask for three menu options with per-head costs from your cost sheet, a cover paragraph and a payment terms section. Costing: if your food cost per head is $11 and you target a 35% food cost, the price per head is about $31.40 ($11 divided by 0.35), so the event comes to $1,257 for 40 guests before service and delivery. The assistant does the division and the formatting; you add labour, travel and your margin policy. How caterers use AI to quote events and plan menus has the full workflow.
The allergy line in that brief is exactly where to slow down. List the allergy in the proposal, but have the kitchen confirm each dish against the checked allergen table before the proposal goes out.
A supplier email drafted from the order sheet
Once the quantities are settled, the assistant can write the message. Paste the order lines and the delivery day, and ask for a short email per supplier. A before and after (illustrative):
Before: "chicken 13kg, yoghurt 6 tubs, flatbread 80, deliver thurs am pls, usual drop"
After: "Hi [first name], our order for Thursday morning delivery: chicken thigh 13 kg, natural yoghurt 6 tubs, flatbread 80. Same drop point as usual, before 9:30 if possible. Please confirm by reply. Thanks, [your name]"Keep the prices out of the draft unless you want the supplier to see them, and never let the assistant "negotiate". If a price has gone up, you decide whether to switch, which is a judgement the assistant cannot make for you. A cafe that sends four such emails a week saves perhaps 40 minutes, enough to cover the monthly plan twice over in time terms.
How the picks change by business type
The same six picks apply, but the order changes.
- Cafe: costing and menu copy first, then till AI for the morning and lunch peaks, then supplier orders. Phone answering rarely matters.
- Caterer: quotes and proposals first, since they are the revenue, then costing, then allergen records, which matter most because every event has special diets.
- Takeaway: phone and order accuracy first, then supplier orders, then costing. Delivery-app fees often matter more than any tool, so read the delivery-app tutorial in the further reads before spending on tools.
For cafes in particular, the cafe tools tutorial in the further reads covers a longer list of tasks, and the AI receptionist comparison there names the phone products in more detail.
Food-business AI that is not worth buying yet
- Fully automatic order-taking by chatbot or voice, unless you have tested it on every item and modifier. A wrong order costs more than a missed call.
- Demand-forecasting software priced for chains. One site's data is small. A spreadsheet and a weekly assistant check come close.
- AI-generated photos of dishes you do not serve. Customers complain and platforms may object. Use real photos.
- Replacing staff scheduling judgement. An assistant can draft a rota from availability, but a manager must check legal break rules and fairness.
A decision rule for choosing among the six
- List the five jobs that cost you the most hours a week. Write the hours next to each.
- Mark the jobs that are writing or arithmetic (costing, copy, supplier emails). Those are the assistant's.
- Mark the jobs that need your sales data (best sellers, labour cost). Try the till's AI if you have one, otherwise a spreadsheet export.
- Mark the jobs that need a voice (phone orders). Count the misses before buying.
- Start with the assistant on the free plan for two weeks, and add one paid tool at a time.
Total cost for a typical small cafe: $20 for the assistant, nothing extra for the till's AI if included, and a phone service only if the count justifies it. For a monthly view across a whole restaurant, see the budget tutorial in the further reads.
What to check after the first month
Look at three numbers. First, hours: how long did the costing, ordering and writing take before and after? Second, food cost: did your weekly food cost percentage move? Third, errors: did any wrong figure, allergen or order slip through? If errors appear, tighten the prompts and add a second pair of eyes before expanding. A tool that saves two hours but causes one wrong order a week is not a saving.
Keep the process boring and written down. A one-page note for each job (the prompt, the data you paste in, who checks the result) lets a new team member follow it. That note, rather than a clever tool, is what keeps AI useful when the busy weeks arrive.
Questions food owners ask about AI
Is there one best AI for a small food business?
No. A general assistant such as ChatGPT or Claude suits costing, menu copy and supplier emails. The AI inside your till suits sales questions. A phone-answering service suits missed calls. Each handles a different job, so choose by job and start with the one that costs you the most time.
Can AI be trusted with allergen information?
Not as a source. Allergen facts must come from supplier labels and your recipes, checked by a person and dated. An assistant can tidy those records into a table for the counter, but a customer's allergy question should be answered from the checked record, not from a chatbot's memory.
Do I need to change my POS to get AI?
Not necessarily. Toast IQ is included with Toast POS and Square AI is in beta for Square users, so those help only if you already use the till. If you do not, export your sales to a spreadsheet and ask a general assistant the same questions, which costs about $20 a month.
What does an AI phone answering service cost?
It varies by provider and call volume. Smith.ai's AI receptionist, for example, lists 25 calls free and then $150 for 75 calls. Before paying, count how many calls you actually miss in two weeks and what an average missed order is worth, then compare the monthly fee against that.
Further reads
- Best AI Tools for Cafés and Coffee Shops, Grouped by Task — Tools grouped by task for cafes and coffee shops.
- AI Phone Answering for Restaurants: What It Costs and Pays Back — What AI phone answering costs and pays back.
- How Much Does AI Cost a Small Restaurant Each Month? — A monthly cost view for a small restaurant.
- AI Ordering vs Delivery Apps: What a Takeaway Actually Pays — What a takeaway actually pays on delivery apps.
- AI Receptionist for Restaurants: Yelp Host, Slang, Popmenu, Maple — Comparing AI receptionists for restaurants.
- How to Write Menu Descriptions With AI That Still Sound Like You — Menu descriptions that still sound like you.
- Catering Quote Software: Which Tools Turn Enquiries Into Quotes? — Time an enquiry-to-quote run with and without AI, then pick between caterer-specific and general quote tools using dated list prices.
- Catering Software for Small Business: Free Options and AI Extras — A free quote, order and invoice stack for a small caterer, then what a free AI assistant adds to each step and where it stops.
- Food Truck POS: Offline Mode, Fees and Which AI Features Pay Off — Square, Toast Go and Clover compared for a truck: offline limits, card fees worked through on real sales, and the AI features worth paying for.
- Can a Restaurant Use AI Food Photography on Menus and Delivery Apps? — Where AI food photos are safe (lighting, backgrounds) and where they mislead (portions, ingredients), with a spec-card check and a real-photo fallback.
- 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; Square AI product page; OpenAI, Anthropic and Google pricing pages; Smith.ai pricing page. Checked October 2026.