Blog · AI Calling
A voice agent is only as good as your price list.
People judge a calling agent by its voice. Retailers judge it by whether the price was right. We think the price list deserves more attention than the voice.
When teams first hear a voice agent, the conversation is about how it sounds. Is it natural? Does it handle Hinglish? Those things matter, and they are the easy part to demo. The thing that decides whether retailers keep ordering through it is quieter: did it quote the right price, apply the right scheme and respect the minimum order? A warm voice that gets the price wrong is worse than no call at all.
What a real order call looks like
Picture a distributor of bakery ingredients calling a neighbourhood bakery on a beat whose rep is on leave. The baker says, in a mix of Hindi and English, that they need two cartons of whipping cream and six packets of cocoa. In the few seconds that follow, the agent has to know:
- The price of each SKU in this zone, for this packing.
- Whether two cartons meets the minimum order.
- Which trade scheme applies this week, and what it does to the baker’s margin.
- That a carton price is not a packet price.
None of that is a voice problem. It is a data problem. AI Calling is built so that the model answers only from your knowledge base: prices, minimums, schemes and policies. It knows each SKU’s price by zone, packing and minimum order, applies trade schemes and explains the retailer’s margin, and never quotes a bundle price as a piece price. If it does not know something, the unknown goes to a callback rather than a guess.
Where calling projects actually go wrong
In our experience the risk is rarely the speech recognition. It is the gap between the price list the agent has and the price list sales is actually using. Common versions of that gap:
| What happens | What the retailer hears |
|---|---|
| A price change goes out by email but the agent’s list is not updated | An old price, confidently stated |
| A scheme ends on Friday but is still loaded on Monday | A discount the distributor will not honour |
| Minimums differ by zone but are stored as one number | An order that will be rejected later |
| Pack sizes are named loosely | Confusion between a carton and a packet |
Every one of those is a promise the company then has to either keep at a loss or break with an annoyed customer. That is why we treat the price list as the real script.
Treat the price list like master data, not a PDF
AI Calling has a knowledge desk with price list sync, minimums and schemes, a product knowledge base and a price-change review. The idea behind the review step is simple: a change to what the agent will say about money should be seen by a person before it goes live, in the same way a change to the rate card would be.
The habits we would put in place before switching on any order agent:
- One owner for the price list the agent uses, and one place it lives.
- Every scheme loaded with a start and an end date, not just a description.
- Minimums and prices stored per zone and per packing, not as one figure.
- A review step for any price change before the agent can quote it.
- A rule that anything not in the list becomes a callback, never an estimate.
Then check it on every call
Even with a clean list, you want evidence. AI Calling scores price accuracy, script, compliance and listening on every call, and saves a transcript, summary and outcome against the caller. That turns “I think the agent quoted the wrong price” into something you can look up. It also catches the case where the list was right but a SKU name was ambiguous, which is a fix to the catalogue rather than to the agent.
If you are planning to use calls to keep beats ordering while reps are away, read the absent sales rep beat coverage playbook and how AI order-taking calls to retailers work. Both assume the price list is in order; this post is the reminder to make sure it is. More on the AI Calling guides page.
Questions
What does an order-taking voice agent need to know?
Each SKU’s price by zone and packing, the minimum order, the trade schemes currently running and how they affect the retailer’s margin. AI Calling answers only from that knowledge base.
What happens if the agent does not know a price?
In AI Calling, unknowns go to a callback. The agent does not invent prices, discounts or promises.
How do you catch pricing mistakes on calls?
AI Calling scores price accuracy on every call and saves the transcript and summary, so a disputed price can be checked against what was actually said.
See it on your own data. AI Calling — Voice agents that take orders. Book a 30-minute working session with an engineer.