In most fast-food franchises the franchisor decides: drive-thru ordering is usually a brand-standard technology, so the system, vendor, timing and often the fee are set by head office. Your franchise agreement and operations manual show whether you can join a pilot, opt out or pick an approved vendor, so read those before speaking to any supplier.
Reports of recent rollouts describe a pattern: a pilot in company-owned restaurants, an invitation to selected franchisees, and then, if it works, a standard. Dairy Queen's franchisor, for example, widened its voice-ordering pilot from company stores to selected franchisees in April 2026, and trade press described it as a pilot, not a chain-wide rollout; the published coverage doesn't give the terms those franchisees were offered. Pilots can also stop. McDonald's ended its automated order-taking test with IBM, which ran in about 100 restaurants, in 2024 without giving a reason, after social-media videos showed the system getting orders wrong. The pilot stage is when a franchisee has the most say.
Three ways brands handle drive-thru technology
Which model your brand uses decides how much room you have. The three below are patterns, not quotations from any one agreement.
| Model | Who decides the system | Where you have room | What it feels like |
|---|---|---|---|
| Mandated standard | The franchisor | Install timing, training, local set-up, the terms of any pilot you're offered | A required system, a fee set by head office, a rollout schedule you're told |
| Approved-vendor menu | The franchisor approves several; you choose one | Which vendor, subject to point-of-sale integration and brand standards | A certified list; you sign with the vendor and may negotiate price |
| Open, with approval | You propose; the franchisor approves | Wide in theory, slow in practice | A supplier-approval process with security, brand and integration reviews |
Voice ordering has to connect to the brand's point-of-sale, menu data and payments, so the first two models are the likelier ones for a drive-thru lane. If a vendor's announcement names franchise groups rather than a brand, read it carefully: it rarely says who approved what inside each brand, which is exactly the question to ask yours. For the wider picture of what a franchise agreement allows around AI, what your franchise agreement allows before you add AI tools goes through the common clauses.
Where your paperwork says who decides
This is a reading guide, not legal advice. Have a franchise lawyer review the actual wording, especially before you sign anything connected to a pilot. These are the places to look.
- The technology or computer-systems section. What you must buy, from whom, and whether the franchisor can require new systems or upgrades during the term.
- Approved-supplier clauses. Whether any vendor needs prior approval, and who can refuse it.
- The operations manual and how it changes. The manual is usually part of the agreement and the franchisor can update it, so a new standard can arrive as a manual update rather than a new contract.
- Fees. Technology fees exist in some agreements. Look for a fixed fee, a per-transaction fee, or a pass-through of vendor charges, and any cap or notice period.
- Capital spending, renewal and transfer conditions. Whether equipment upgrades are a condition of renewing or selling.
- Customer and transaction data. Who owns it, who may use it and whether you get access. This varies by agreement.
- Advisory council or association rights. If your brand has one, it may be a formal route for franchisees to be consulted on technology changes.
- Exit and transfer terms. What happens to systems and data when the agreement ends or the store is sold.
If the agreement says you follow the manual and the manual says use the approved ordering system, the decision was made before any vendor called you. Where the documents are silent, the franchisor's written position is worth far more than a phone call, so get it in writing.
Why a pilot is your negotiating moment
Before a standard is set, a franchisor needs franchisees willing to test, and willing franchisees can ask for terms. After the standard is set, the same questions look like objections. A pilot offer should state, at minimum:
- How long it runs, at how many stores and in which lanes.
- What success means: order accuracy, speed of service and the share of orders handed to staff, each defined and measured by someone you can check.
- Who pays for hardware, installation, the subscription and any per-order fee, during the pilot and afterwards.
- Who can see the data and for how long it is kept.
- Your right to pause or leave without penalty, and what happens to the equipment.
- How the results will be reported back to franchisees.
An early test can end, as McDonald's IBM test did, and the terms should say who carries the cost if it does. If your brand runs a franchisee association or advisory council, use it: a question raised by many franchisees gets a clearer answer than one raised by you.
Questions to put in writing before you accept a pilot or a mandate
Group your questions so the answers are easy to compare, and send them as one email so the replies sit together.
- Money. Who pays for equipment, installation and support? Is there a monthly or per-order fee, and can it change? What credit do I get for orders the system gets wrong?
- Accuracy. What order-accuracy and speed targets apply, how are they measured and who audits them? What happens if the targets aren't met?
- Control. Can I switch it off for a shift or a lane? How quickly can I change menu items, prices and limited-time offers? Who sets the upsell prompts?
- People. Who monitors orders as they are taken? What training do my crew get, and what is the takeover procedure?
- Data. What is recorded, who owns it and who can see my store's records?
- Exit. What happens to equipment, data and fees if the pilot ends or the brand changes vendor?
Subject: AI drive-thru pilot - questions before we agree
Hi [first name],
Thanks for the invitation to take part in the voice-ordering pilot at
[store]. Before we confirm, could you answer these in writing?
1. Who pays for equipment, installation and support, and is there any
monthly or per-order fee during or after the pilot?
2. What order-accuracy, speed and hand-over targets apply, who measures
them, and can we see store-level results weekly?
3. Can we pause or leave the pilot without penalty, and what happens to
the equipment if we do?
4. What audio, transcripts and order data are kept, who owns them, and
what access will we have to our own store's records?
5. Who is responsible for telling customers an AI is taking their order?
Thanks, [your name]Keep the replies. If the same question gets different answers from head office and from the vendor, send both back to the franchisor and ask which one governs.
Measure your lane before the system arrives
The most useful thing to do after the first announcement is also the cheapest: record your lane's current numbers for two weeks. Without a baseline, every later claim about improvement is the vendor's word against your impression. Four figures are enough. Here is an illustrative baseline for the lane in the sum further down.
| Measure | Two-week baseline | How to get it |
|---|---|---|
| Orders through the lane | 5,600 | The point-of-sale report for the lane |
| Remakes caused by order errors | 224 (4.0%) | A tally sheet at the pass: tick every remake with a one-word reason |
| Average time at the speaker | 41 seconds | Your timer system, or a stopwatch sample of 100 orders |
| Order-related complaints | 11 | Comment cards, app reviews and calls |
The tally sheet is the figure that needs a person. Keep the reasons to a handful: wrong item, missing modification, missing item, wrong size, other. That short list also shows later where the AI differs from your crew. A system that cuts "wrong item" but raises "missing modification" has a pattern worth reporting to the franchisor.
Who owns the voice, the order and the customer
A voice system creates more data than a menu board ever did: audio of customers in cars, a transcript of each order, the order itself, sometimes a link to a loyalty account. Customer data ownership varies by agreement, so your first question is what you can see about your own store. You need access for three jobs: checking accuracy, settling a complaint about what a customer said, and coaching your crew.
- Recordings. Are they kept, for how long, and who can listen? Some places have specific rules on recording calls and voices, so ask who is responsible for notices.
- Voiceprints. Some places treat a voiceprint as biometric data. Ask directly whether the system creates one.
- Access. Can you export order logs, audio clips and accuracy reports for your own store, and how quickly?
- Deletion and exit. What is deleted when the pilot ends or you sell the store?
- Customer notice. Customers should know an AI is taking the order. For customers in the EU, the chatbot transparency duty in Article 50 of the EU AI Act has applied since 2 August 2026. Ask who writes the sign or the spoken line, and keep it short: one sentence at the speaker post is enough.
What accuracy is worth against the fee, for one lane
Consider a lane that handles 12,000 orders a month at an average ticket of $11.50. Say 4% of orders go wrong today, which is 480, and each one costs about $6.50 in remake, wasted food and goodwill. The franchisor proposes a $450 monthly fee for the system. Assume extra sales carry a 65% margin. None of these numbers is real; replace them with your own. This is how three outcomes compare.
| Scenario | Error rate | Average ticket change | Effect of errors | Effect of ticket | Fee | Net a month |
|---|---|---|---|---|---|---|
| A. Works as pitched | 3.0% | +$0.25 | +$780 | +$1,950 | -$450 | +$2,280 |
| B. Flat | 4.0% | +$0.05 | $0 | +$390 | -$450 | -$60 |
| C. Worse | 5.5% | $0 | -$1,170 | $0 | -$450 | -$1,620 |
One percentage point of accuracy on this lane is 120 orders, or $780 a month, which is more than the fee. So the fee isn't the main number. The accuracy is, and it's the number you have least control over, which is why audit rights belong in the pilot terms. With no upsell gain at all, the fee is covered if errors fall by about 69 orders a month, or 0.58 of a percentage point. Ask for pilot-store results measured by someone other than the vendor, and treat a claimed ticket increase as a hope until your own lane shows it.
How a pilot should end: a decision grid
Agree the stop-or-continue rule before the pilot starts, when it's still a judgement and not a sunk cost. A simple grid does the job. This is one after four weeks on the lane above (illustrative).
| Measure | Baseline | Pilot, four weeks | Verdict |
|---|---|---|---|
| Remakes from order errors | 4.0% | 3.0% (372 of 12,400) | Better |
| Orders handled by the AI alone | Not applicable | 83.1% (10,300) | 16.9% taken over: acceptable if each takeover is quick |
| Average time at the speaker | 41 seconds | 36 seconds | Better |
| Order-related complaints | 22 per four weeks | 19 | Slightly better |
| Fee | None | $450 a month | 124 fewer errors at $6.50 is $806, so accuracy covers it |
The verdict here is to continue, and to ask for the fee to be waived until 3% has held for eight weeks. If the error rate had come in at 4.4%, the same grid would say pause and send the tally sheet to the franchisor. The grid works only because the baseline exists, so the order of events matters: baseline, then pilot, then grid.
What you still control when head office chooses
- The human fallback. A takeover routine your crew have practised, so a confused order moves to a person in seconds and the customer never repeats it three times.
- Hardware upkeep. Microphones, speakers and headsets. A lane that sounds poor produces poor transcripts whatever the software does.
- Local menu data. Items, prices and limited-time offers that vary by store. A wrongly priced shake in the menu data is the kind of fault you catch first.
- Install timing. Ask to avoid your peak season and a training week.
- Your own accuracy log. Whatever the vendor's dashboard says, a simple end-of-shift log is evidence you can use in any conversation.
- Your voice in the franchisee group. Individual complaints are easy to dismiss; patterns across stores aren't.
The log needs five columns and two minutes a shift. This is a filled-in version (illustrative):
| Date | Orders | Handled by AI alone | Taken over by staff | Remakes from order errors | Note |
|---|---|---|---|---|---|
| Fri | 520 | 431 (82.9%) | 89 | 17 (3.3%) | Speaker post static in the rain |
| Sat | 610 | 489 (80.2%) | 121 | 24 (3.9%) | Limited-time shake priced wrongly until 14:00 |
Two weeks of rows like those turn "it seems worse on busy nights" into something a franchisor must answer. For a fuller habit around hand-offs and errors, monitoring AI that talks to customers applies directly.
Hard cases a pilot has to include
A demo uses a clear voice and a short order. Your lane doesn't. The pilot should include the situations below, and your tally sheet should record them.
| Hard case | Why it breaks | What to check in the pilot |
|---|---|---|
| Modifiers: "no pickles, extra sauce, swap fries for salad" | Modifiers get dropped or attached to the wrong item | Remake reasons, and read 20 tickets back against the audio |
| Corrections: "actually, make that a large" | The system keeps the first item or adds both | The final ticket against what the customer said |
| Several voices at once | Children and passengers speak over the driver | Takeover rate at weekends and family-heavy times |
| Noise: engines, rain, wind, a speaker with static | Poor audio produces poor transcripts | Error rate by weather and by lane |
| Accents and quiet speakers | Speech recognition is uneven across speakers | Takeover and error rates over several weeks, split by time of day |
| Allergy and dietary statements | Never something to leave to a model alone | How the system flags it and who confirms |
| Items you've run out of | A menu sync that lags offers things you can't make | How quickly stock-outs reach the system |
| Very large orders | A 15-item office order strains any system | When it hands over and how smoothly |
The allergy row needs its own instruction for your crew: any mention of an allergy or dietary requirement is confirmed by a person, whatever the system says. Ask the franchisor to state that rule in writing for the lane, because the store is usually the one answering for a mistake, not the vendor.
What to tell the crew
Order-takers will ask two things: will this cost me hours, and what do I do when it gets stuck. Answer only what you know. If the franchisor hasn't said anything about staffing, say so and don't promise either way. Do tell them what the job becomes: watching orders as they're taken, stepping in when the system hands over, and logging remakes. Practise the takeover until it takes seconds. And ask the crew which phrases the system mishears. They hear the lane all day and spot patterns before any dashboard does. Employment terms are a matter for your adviser; the point is not to make promises on the franchisor's behalf. A set of short cards for the takeover moments is in reference cards and quizzes for fast food crews.
Mandated, invited or choosing: what each franchisee can push for
A single-store owner in a mandated rollout. She can't opt out, so her room is in timing, training and paperwork. She asks for install outside peak season, a written takeover procedure for her crew, access to her own store's accuracy data and a clear statement of the fee. Her end-of-shift log becomes the evidence behind any feedback she sends.
A multi-unit operator invited into a pilot. He chooses which stores to put forward, and picks two with different lane layouts so the comparison shows what the layout does. He asks for a fee holiday during the pilot, a no-penalty exit, and for results to go to the franchisee group. A weekly roll-up across his stores, like one weekly AI summary across locations, keeps the comparison honest.
A franchisee in a brand with an approved-vendor menu. He treats the choice as a purchase. He runs the same dozen test orders through each vendor's trial, from clear speech to a noisy lane, and picks on measured accuracy rather than the demo. Twelve test calls to run before go-live gives a set to adapt, and protecting your number, data and exit covers the contract side.
When a voice-AI vendor calls you directly
Vendors sell to franchisees as well as head offices, so you may get the call. Don't sign, and don't agree to a demo that shares your store's data. Reply that your brand controls drive-thru technology, ask for their details and pass them to your franchisor's technology team. If the franchisor says go ahead, you've lost nothing. If the franchisor says no, you've avoided a contract that may have put you in breach.
Franchisee worries about AI ordering
Can I refuse an AI drive-thru rollout?
If the system is a brand standard in your operations manual, refusing may breach your agreement, and the manual can usually be updated by the franchisor. Ask in writing whether deferral, exemptions or opt-outs exist for your layout, and have a franchise lawyer read the technology and manual clauses before you answer.
Who pays for the hardware and the monthly fees?
It varies by brand and agreement, and the public reports on pilots don't say. Technology fees exist in some franchise agreements. Ask in writing who pays for equipment, installation, subscription and any per-order fee, during the pilot and after it, and whether the fee can change.
Do I have to tell customers an AI takes their order?
Often yes, and the franchisor usually sets the wording. If you serve customers in the EU, the chatbot transparency duty in Article 50 of the EU AI Act has applied since 2 August 2026. Ask the franchisor and the vendor who is responsible for the notice and where it appears.
Can I pick my own voice-AI vendor?
Only if your brand's supplier-approval process allows it. Voice ordering has to connect to the brand's point-of-sale, menu data and payments, which is why most brands choose the vendor themselves. Ask what the approval process is, how long it takes and which vendors are already certified.
Further reads
- AI Restaurant Order Taking: Who Takes the Order, Who Texts a Link — Phone, text and web order-taking compared.
- Square AI Voice Ordering for Restaurants: Set Up and Test (Beta) — A voice-ordering set-up for an independent.
- Franchise Audit Checklist: Prepare for a Brand Inspection With AI — Prepare for a brand inspection.
- AI Chatbot Disclosure: What to Tell Customers at the Start of a Chat — Wording for telling customers it is an AI.
- How to Run Your First AI Pilot Project in a Small Business — The stages of a first AI pilot.
- Kiosks AI for Cafés and Takeaways: Self-Service Ordering Costs — Kiosks as a different kind of AI ordering.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: trade-press coverage of the 2024 end of McDonald's IBM order-taking test and of Dairy Queen's April 2026 franchisee pilot expansion, with the voice-AI vendor's press release; general franchise-agreement structure; EU AI Act Article 50. No franchise agreement terms are quoted.