AI phone order taking: voice orders straight into Shopify
A phone ordering system that takes customer orders by voice and creates them in the retailer's Shopify store.
The problem
A retail business wanted customers to be able to place orders by phone without a person handling every call. The orders had to be accurate and stay in step with the online storefront.
Challenges
Accuracy on critical values
Spoken orders go wrong on the details that matter most, such as the total and the delivery choice, where a misheard word has a real cost. The build had to make those values reliable without slowing the call, because speech recognition alone was not trusted for them.
Matching the storefront at the moment of ordering
A phone order is only useful if it agrees with the online store. The system had to check availability and current pricing in the Shopify catalogue before finalising, so a caller is never promised something the storefront could not actually supply.
Landing in normal order management
Orders taken by phone needed to appear in the merchant's usual order flow, not in a side system someone has to re-key. That meant creating real orders through the commerce API rather than sending staff a note to act on.
How the system works
Call intake
An ElevenLabs conversational voice agent answers inbound calls and walks the customer through the order, collecting the product choice, quantity and delivery details.
Keypad confirmation
At the points that matter most, such as confirming the order total or choosing a delivery slot, the caller presses keys instead of relying on speech recognition alone. This protects accuracy on sensitive details.
Product lookup
Before the order is finalised, the system queries the Shopify catalogue through the GraphQL API to check availability and current pricing.
Order creation
Once confirmed, the order is created directly in Shopify through a GraphQL mutation, so it appears in the merchant's normal order management.
Spoken confirmation
The customer hears a summary of the order before the call ends, similar to the confirmation step of a live checkout.
Design decisions
Keypad input at decision points
At the moments that matter, the caller presses keys instead of speaking. We chose keypad input for critical values because it is unambiguous, and left ordinary conversation to the voice system, where natural speech is faster and more comfortable for the customer.
Live catalogue lookup before confirmation
Rather than holding a copy of products and prices, the system queries the store when the order is about to be finalised. It costs a little time on the call but removes the risk of confirming from stale data.
Spoken summary before the call ends
Before hanging up, the caller hears a summary of the order, much like the confirmation step of a live checkout. It gives them a last chance to catch a mistake, which is far cheaper to fix on the call than after dispatch.
Result in production
Phone orders land in the same order management system as web orders, which removes manual re-entry.
Phone orders land in the same order management system as web orders, so both channels are handled in one place.
Nobody has to re-enter phone orders by hand, which removes a repetitive step and a source of transcription mistakes.
Customers can place an order by phone without a person having to handle every call.
What we would do differently
We would push more of the intent routing into a smaller, cheaper model with structured outputs, and keep the more expensive model for the final step only. It should cut cost noticeably at scale without hurting quality.
Who this pattern fits
We built this for a retailer, but the shape suits any business with similar call volumes or a need to keep phone and online channels coordinated. The stack is stable in production, can be self-hosted where it is critical, and is light to operate.
Tools used
ElevenLabs
Speech synthesis for the agent's voice, chosen for production latency and quality.
Twilio
Telephony carrier for inbound and outbound calls, SMS and call recording.
DTMF
Keypad input for menus and payment entry during live calls.
Shopify GraphQL API
Commerce backend integration for product lookups and order creation.
n8n
Workflow runtime and orchestration, with version-controlled logic and observability.
Related
Frequently asked questions
Does the caller confirm the order before it is placed?
Yes. Key points such as the total and delivery choice are confirmed by keypad, and the caller hears a spoken summary of the order before the call ends. The order is created in Shopify only once it has been confirmed.
Does voice ordering only work with Shopify?
This build uses the Shopify GraphQL integration. The same pattern can extend to any commerce backend that exposes product and order APIs, including headless setups.
How does the caller pay?
Card details can be captured by keypad through Twilio, or an SMS link can be sent during the call so the customer completes checkout on their phone.