AI Ordering vs Delivery Apps: What a Takeaway Actually Pays

Per order, usually yes. Delivery apps take 15% to 30% of each delivery order's subtotal, while an order through your own AI phone line or website costs about 3% plus 30 cents in card fees. But you also pay for the AI tool, roughly $220 to $350 a month for a phone agent that takes orders, and for couriers, so volume decides it.

One warning before the numbers: the two options don't do the same job, so a straight commission comparison flatters AI ordering. Below you'll find every charge side by side, a worked month for a busy takeaway, and a break-even worksheet to run on your own orders before you sign anything.

Follow me on Instagram@sagnikteaches

What each option is really selling you

A delivery app bundles three things into one percentage: customers who have never heard of you, a courier to reach them, and the ordering and payment system. "AI ordering" usually means an AI phone agent or chat assistant that takes orders into your own till or ordering page. It replaces only the third item, and it only works on customers who already know your number or website.

Connect on LinkedInSagnik Bhattacharya

That's why the honest question isn't "which is cheaper?" but "which of my orders am I overpaying for?" A regular who has ordered the same kebab every Friday for two years doesn't need an app to discover you. A student who moved into the area last week probably does.

Subscribe on YouTube@codingliquids

The charges, line by line

These are published list rates as of September 2026. Rates vary by country and some restaurants have negotiated custom deals, so check the figure in your own merchant portal before you run the numbers.

You can check the current app rates on DoorDash's merchant pricing page and a typical AI phone agent's plans on Maple's pricing page. Notice that the phone agent's cheapest plan doesn't take orders at all. Read plan names carefully: "AI phone answering" and "AI phone ordering" are different products at different prices.

A worked month for a 1,000-order takeaway

Say a pizza-and-grill takeaway (an illustration, not a real business) handles 1,000 orders a month: 600 app deliveries, 150 app collections and 250 phone orders that staff take by hand. The average basket is $30 and it's on a 25% app plan.

• Delivery commission today: 600 × $30 × 25% = $4,500
• Collection commission today: 150 × $30 × 6% = $270
• Total paid to the app: $4,770 a month

The owner adds an AI phone agent at $350 a month (monthly billing) and pushes regulars to ring or order online. After two months, 120 of the app deliveries and 75 of the app collections have moved to direct. A direct $30 order costs $1.29 in card fees (3.3% plus 30 cents).

Now the delivery line decides everything:

Two things jump out. Moving collection orders saves almost nothing, because app pickup commission is already low. And moving delivery orders only pays if someone other than you covers most of the courier cost.

The phone agent's other value is capacity, not commission. Those 250 phone orders carry no commission now, but each one ties up a member of staff for a few minutes, and at 7pm on a Friday some calls ring out. If staff spend about three minutes per call, that's around 12.5 hours a month. Count your unanswered calls from the phone provider's call log for two weeks before you buy; that number, multiplied by the share of callers who would have ordered, is often a bigger figure than any commission saving. What AI phone answering costs a restaurant and how it pays back covers that side in detail.

The break-even worksheet

Copy this into a spreadsheet or onto the back of an order pad. Use the average basket from your app dashboard and your real courier cost.

Average basket (B):                          $_____
App commission rate on delivery (c):         _____%
Card fee on a direct order (p% of B + fixed): $_____
Your delivery cost per direct drop,
  after any delivery charge to the customer (d): $_____
Monthly cost of the AI ordering tool (F):    $_____

Saving per moved delivery order:  s = (c x B) - card fee - d
Orders you must move each month:  F / s   (round up)
If s is zero or negative, moving delivery orders loses money.

Filled in for a different illustrative business, a noodle bar on a 30% app plan with its own part-time driver:

Average basket (B):                          $24
App commission rate on delivery (c):         30%
Card fee on a direct order:                  3.3% x $24 + $0.30 = $1.09
Your delivery cost per direct drop:          driver $15 an hour,
                                             4 drops an hour = $3.75,
                                             less $2.50 delivery charge
                                             = $1.25
Monthly cost of the AI ordering tool (F):    $220 (annual billing)

Saving per moved delivery order:  s = $7.20 - $1.09 - $1.25 = $4.86
Orders you must move each month:  $220 / $4.86 = 45.3, so 46

Forty-six orders a month is fewer than two a night, which this noodle bar cleared in its first fortnight. The driver line is the one to be honest about: on a wet Friday the same driver manages two drops an hour, not four, so the cost per drop doubles to $7.50 before the delivery charge, the saving falls to about $1.11, and those nights are better left on the app.

Here's how that plays out for a $30 basket and a $350-a-month tool:

The uncomfortable row is the top one. On a 15% plan, the app's commission on a $30 delivery ($4.50) is less than a typical courier fee, so if you subsidise delivery you're better off leaving those orders where they are. Cheap app plans shrink the case for direct delivery; they don't touch the case for answering your own phone.

Where the apps still earn their cut

• First-time customers. If a large share of your app orders come from people who have never ordered before (some merchant dashboards show new against returning customers), the commission is partly a marketing cost. You'd spend money reaching them some other way. Put a number on it. If 30% of the pizza-and-grill takeaway's 600 app deliveries were first orders, that is 180 new customers a month, and the commission on those orders is 180 × $7.50 = $1,350, or $7.50 per new customer. Compare that with what else you would spend: a leaflet drop that costs $120 and brings eight first orders (illustrative figures) works out at $15 a customer. On those numbers the app is the cheaper way to find people, and the expensive part is paying the same 25% on their tenth order.
• Delivery range and couriers on busy nights. An app can put ten drivers on your Friday rush. Your own driver can't. For a takeaway with no drivers, the courier alone justifies a good slice of the commission.
• Quiet nights. On a slow Tuesday the app may be the only thing bringing in orders. Judge commission against the margin on orders you wouldn't otherwise get, not against zero.
• Customers who will only ever use the app. Some people simply won't ring a takeaway or type in card details on a new site. Chasing them off the app just loses them.

Which orders to move first

1. Phone orders you already get. No commission is involved, so the gain is staff time and fewer missed calls at peak. This is where an AI phone agent is easiest to justify.
2. Large group and family orders. Commission is a percentage, so an $80 order on a 25% plan costs you $20. One regular office order a week moved direct can cover a month of a mid-priced tool.
3. Short-radius deliveries you can batch. If your own driver can drop three orders on one run, your cost per drop falls well below an on-demand courier's.
4. Regular app customers who live nearby. A card in the bag with your direct number and a small loyalty reward is the usual prompt. Check your merchant agreement on bag inserts first. If it's allowed, keep the card to one idea. An illustrative version: "Order direct next time and your fifth order's on us. Ring [number] any time; our ordering line answers straight away, even on Friday nights. Or order at [web address]. Same menu, same prices, collection ready in 20 minutes." It names the reward, the channel and the reason (no engaged tone at peak), and "same prices" matters if your app menu is marked up, because a customer who finds the direct price higher will not try twice.
5. Collection orders last. The saving per order is tiny, as the worked month shows. Move them because customers prefer it, not for the fee.

If your regulars prefer messaging to phoning, taking takeaway orders on WhatsApp with AI is another direct channel with a different cost profile. Check that profile for your set-up: from 1 October 2026, ordering bots running on the WhatsApp Business Platform pay for service replies beyond the first 1,000 a month per business number, at rates that vary by the customer's market. The free WhatsApp Business app is a separate product, but Meta's own AI replies inside it (Meta Business Agent) have been charged per token since 1 August 2026, which Meta puts at roughly 4 to 5 cents a message.

Setting up AI ordering so it doesn't cost you orders

A phone agent that mishears "no onions" is worse than a missed call, because the customer gets the wrong food and blames you. Before it goes live:

• Map the menu exactly as the till has it. Every size, every half-and-half option, every extra topping and its price. Most ordering errors come from modifiers the agent wasn't told about.
• Run 30 test calls. Get staff and friends to ring with background noise, accents, changed minds mid-order and addresses on new streets. Check every ticket that prints.
• Set the allergy rule in writing. The agent should never answer an allergen question itself; it transfers to a person. Whether an AI should answer allergen questions at all explains why.
• Decide the payment route. Card over the phone through the agent, pay on collection, or a payment link by text. Each changes your card fees and your no-show risk.
• Plan the failure mode. What happens if the till integration drops mid-rush? The answer should be "calls transfer to the shop phone", not "orders vanish".

An illustrative test call shows how a missing modifier turns into a wrong ticket:

Caller:  Can I get a large pizza, half pepperoni, half veggie,
         no onions on the veggie side?
Agent:   Sure. That's a large pepperoni and a large veggie with
         no onions. Anything else?

Ticket printed:
  1 x Large Pepperoni         $14.50
  1 x Large Veggie            $13.50
      - no onions
  Total                       $28.00

The customer wanted one pizza at $15 and was quoted two. The agent read back what it did, so a sharp caller would have caught it, but many don't listen to the read-back. The cause was that the till had a "half and half" option the menu mapping never included, so the agent had nothing to map the request to. After adding it, the retest printed "1 x Large Half and Half: Pepperoni / Veggie (no onions)". Every modifier your staff key in by hand on a Friday night belongs in the mapping, and it is worth reading ten printed tickets from the first night against the call recordings.

The full build, from choosing a number to handling busy-signal overflow, is in how to set up an AI phone line for takeaway orders. And if a vendor quotes a mix of monthly fees, per-order fees and per-minute charges, reading AI software pricing helps you turn it into one comparable monthly figure.

Eight weeks of numbers that settle the question

Pull these every Monday into one sheet:

• Direct orders as a share of all orders, split into delivery and collection
• Commission paid to each app, and your plan rate
• AI-taken orders that needed a remake or refund (your error rate)
• Calls the agent transferred to staff, and calls that still rang out
• Average basket on direct orders against app orders
• Courier or driver cost per direct delivery, after delivery charges

An illustrative week-eight row for the pizza-and-grill takeaway, with what each figure told the owner:

On commission alone, the moved orders now save about $280 a month (120 deliveries at $2.01 each with the real $4.20 courier cost, plus 75 collections at $0.51), still short of the $350 tool. The phone line is paying its way through capacity and bigger baskets rather than commission. The owner kept the AI line, stopped pushing distant delivery customers to go direct, and looked at dropping one app to a cheaper plan.

At week eight, put the real figures into the worksheet. If the orders you've moved are above break-even and the error rate is below roughly one in fifty, keep going and consider annual billing, which on Maple's published plans cuts the Pro price from $350 to $220 a month. If you're below break-even, look at which line failed. Usually it's delivery cost. The fix is often to keep deliveries on the app, keep the AI line for phone and collection, and drop to a cheaper app plan.

Takeaway owners also ask

Can I charge higher prices on the apps to cover the commission?

Many takeaways price app menus higher, but check the platform's terms first. On Uber Eats, for example, pickup orders are charged 7% where in-store pricing is validated and 10% where it isn't, so a marked-up app menu can raise your collection rate. Customers also compare prices, and a big gap trains regulars to ring you instead, which may be exactly what you want.

Should I leave the delivery apps altogether?

Rarely at first. The apps bring customers who have never heard of you, and those customers are the ones you later convert to direct ordering. A safer path is to move your regulars, collection orders and large group orders to direct channels, keep the apps for discovery, and look again after three months of data. Some takeaways then drop to a cheaper plan rather than leaving.

What happens when the AI phone agent can't understand an order?

A well set-up agent transfers the call to a person or takes a callback number rather than guessing. Before launch, test it with noisy lines, strong accents, half-and-half pizzas and long modifier lists, and set a firm rule that any allergy question goes straight to staff. Count transfers each week: a high transfer rate means your menu mapping needs work.

Do I own the customer details from app orders?

Generally not in the way you own your direct customers. Apps typically mask or limit the contact details they pass to restaurants, and your merchant agreement sets what you may do with them. Read that agreement before contacting app customers or putting marketing inserts in bags, and build your own list through direct orders, in-store sign-ups and your loyalty scheme instead.

Further reads

• How Much Does AI Cost a Small Restaurant Each Month? — The wider monthly AI bill for a small food business, beyond ordering.
• Restaurant AI Tools: 12 Questions to Ask Before You Sign Up — Questions to put to an ordering vendor before you commit.
• How to Measure an AI Receptionist's Return in the First 90 Days — How to judge a phone agent on real numbers after launch.
• Outcome-Based AI Pricing: Paying per Resolution, Task or Result — When per-order or per-call pricing beats a flat monthly fee.
• Where AI Actually Saves Time in a Small Restaurant — Other places a busy kitchen gets time back from AI.
• Per-Seat vs Usage-Based AI Pricing: Which Costs Less for You? — How to compare flat, per-seat and usage-based software bills.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: DoorDash merchant pricing and Online Ordering pages; DoorDash Drive On-Demand product page; Restaurant Dive's report on the Uber Eats marketplace fee changes (March 2026); Maple pricing page; Owner.com pricing page; Square fees help article. All checked September 2026.