AI Voice Agent for Ecommerce: Cost per Order-Status Call

An AI voice agent usually costs between roughly 25 cents and a dollar per order-status call for an online shop, from listed per-minute prices of about $0.07 to $0.31 plus telephony and a typical three-minute call. It is worth it when you take several hundred such calls a month, or miss them out of hours.

The price per call is rarely the deciding number. The deciding numbers are how many calls you really get, whether the agent can read your order data, and what share it resolves without a person. Run those three through the sums below before anything else.

Follow me on Instagram@sagnikteaches

What an order-status call looks like

Most "where is my order" calls follow the same pattern. The caller gives an order number or an email, the agent finds the order, states the status (paid, packed, shipped, delivered), the carrier and the expected date, and the caller hangs up. A smaller share needs something else: an address change after dispatch, a parcel marked delivered but not received, a damaged item, or a refund. The first group suits automation; the second needs a person.

Connect on LinkedInSagnik Bhattacharya

That split is the whole business case. If 70% of your status calls are the simple pattern, an agent that handles those reliably removes most of the phone queue. If only 30% are, you are paying for an agent that mostly passes calls on, and a tracking link in the order confirmation email would have done more.

Subscribe on YouTube@sagnikteaches

What the platforms list per minute

Prices come from vendor pages read in November 2026. They change often and the headline rate may leave out parts of the bill.

The Gorgias line is a different pricing model and worth noticing: if your help desk already handles chat and email there, a flat price per automated interaction may beat per-minute billing for long calls, though rarely for short ones. Voice is listed as an add-on whose price depends on volume, so ask for a quote.

Working out cost per call

The formula is simple: average call minutes × all-in rate per minute, plus any per-call charges, plus the share of calls that transfer to a person times the cost of that person's time. Do it with your own numbers; here is an illustration.

Example: a four-person shop with 600 status calls a month

Say the shop averages 600 order-status calls a month. Average length with the agent is 2 minutes 40 seconds, about 2.7 minutes. Use a mid-range all-in rate of $0.12 a minute, an assumption in line with the Vapi component sum and Bland's Build rate. Usage is 600 × 2.7 × $0.12 = $194.40. Add a number at $2 a month and, say, $10 for extras: about $206. That is roughly $0.34 a call. Suppose 70% are fully resolved and 30% transfer, costing the shop 180 transfers × 4 minutes of staff time at $25 an hour = $300 of time, though a person would have spent time on those calls anyway.

The comparison is with what handles those calls today. A part-time hire is the usual alternative. Illustrative: 15 hours a week at $18 an hour, with 20% on top for employer costs, is 15 × 52 ÷ 12 = 65 hours a month × $21.60 = $1,404 a month. If that person only did phones, 600 calls works out at $2.34 a call, although most of their time would be idle between calls and in practice they would also do email and returns. If you only reallocate the time of an existing team, the true saving is smaller: 420 calls resolved by the agent × 4 minutes of avoided handling = 28 hours, or $700 at $25 an hour, against about $206 of agent cost. The saving is real but modest, around $490 a month, and it depends on the 70%.

What changes the answer

Raise the call length to 4 minutes and the usage cost rises by half. Fall back to 40% resolved and the saving nearly disappears because staff still take most calls. Add a $299 monthly platform fee and the bill is $505 at the same volume. Move to 150 calls a month and fixed fees dominate: a $299 fee on 150 calls is about $2 a call before any usage. The cheapest option for low volume is usually a pay-as-you-go plan with no monthly fee, or no agent at all.

The order-lookup limits you need to plan around

The agent is a conversation engine; it knows nothing about your orders unless you connect it. Typical setups call your shop platform's API or a small automation that fetches order status by order number, then pass the result back as text for the agent to read. Several practical limits follow.

Read-only access, narrow fields

Give the connection access to status, carrier, tracking number and estimated date. Do not give it the ability to edit, refund or cancel until you have run it for months. A read-only link limits the damage from a misheard order number.

Identity checks

Phone callers can be anyone. Require the order number plus the email on the order, or the order number plus the first line of the delivery address, before the agent says anything beyond "I can't find that". Do not let it read out a full address, phone number or the items ordered to an unverified caller. If the details fail twice, hand over or take a message.

Order numbers are hard to speak

Callers read out digits fast, mix up letters and digits, and say "double five". Speech recognition makes mistakes with these, so have the agent read the number back ("I have order one-zero-four-two-seven, is that right?") before lookup. Offer to find the order by email as a second route. Test with numbers at your real length.

Carrier data is only as fresh as the carrier

The agent reports what your system holds. If tracking updates arrive twice a day, a caller may hear "in transit" for a parcel that was delivered an hour ago. Train the agent to give the date of the last update ("last scanned yesterday at 4pm in the sorting centre") rather than a guarantee, and to offer a text message with the tracking link.

A sample call and what to fix in it

Agent: Thanks for calling [shop]. I'm an automated assistant. Do you have your order number?
Caller: Yes, it's 10427.
Agent: I have order one-zero-four-two-seven. And the email address on the order?
Caller: [first name] at example dot com.
Agent: Thanks. Your order shipped on Tuesday with the carrier and was last scanned yesterday afternoon. The estimated delivery is Friday. Would you like the tracking link by text?
Caller: It says delivered but I never got it.
Agent: I'm sorry about that. I'll pass you to a team member now who can look into it, and I'll give them the details so you don't need to repeat them.

This illustrative call does the right things: states it is automated, verifies with two details, reports a date and the last update, and hands over the "delivered but not received" case immediately. What to fix: read back the email address too, because speech recognition of email addresses is error-prone, and confirm whether the handover happens live or by call back if the team is out.

Handover rules that protect customers and staff

Write these as hard rules in the instructions. Hand over on: delivered-but-missing parcels, damaged or wrong items, refund requests, address changes after dispatch, anything legal or about a payment dispute, an angry caller, and any request for a person. Hand over with a summary: order number, what the caller said, what the agent already told them. Choose a method for when no one is in: a message with the same fields and a promised call-back time, which the team must actually meet. If your agent promises a call back and nobody returns it, the agent has made things worse than a voicemail.

Review the transfers weekly. A handover that was unnecessary means the agent is under-trusted, and one that should have happened but did not means a rule is missing. For the wider chatbot version of these flows, see order status, returns and sizing flows, and for copy-ready conversations seven store conversations to copy.

Peak season: when the maths changes

Status calls are not evenly spread. A shop that takes 600 calls in a normal month may take 1,500 in the fortnight after a sale, when carriers are slow and customers anxious. That is where an agent earns its keep, because it answers as many calls at once as your plan's concurrency allows, while a person answers one. Retell's page lists 20 concurrent calls free on every account, and Bland's plans list 10 and 50 concurrent calls by tier. Bland's Start plan also caps calls at 100 a day, so a fortnight of 1,500 calls, about 107 a day, would already break it. Check every limit against your worst day, not your average.

Illustrative peak: 1,500 calls at 2.7 minutes and $0.12 a minute is $486 of usage in the month, about $0.32 a call, near the normal rate, because per-minute pricing scales smoothly. A part-time hire does not: covering the peak means overtime, a temporary hire or missed calls. That asymmetry, not the average, is the best argument for an agent, and it is worth writing a note of the worst week in last year's call log before you decide.

A delayed-order script, with the wording that works

Delays generate the angriest calls, so give the agent a specific script rather than letting it improvise sympathy. An illustrative instruction block:

If the estimated delivery date has passed or the carrier shows no update for 3 days:
1. Say: "I can see your order [number] hasn't moved since [date of last scan]. I'm sorry for the wait."
2. Do not promise a delivery date.
3. Offer: "I can pass this to our team to open a trace with the carrier. They'll call you back by [time window]."
4. Collect a confirmed phone number and best time to call.
5. Transfer if staff are in; otherwise log the message as "delayed order, needs trace".

The rule against promising dates is the important one. A carrier's estimate is the carrier's, and an agent that repeats it as a guarantee causes the next complaint.

A second scenario: a larger shop

Take an illustrative shop with 2,000 status calls a month, six staff and a help desk that already handles chat and email. At 2.7 minutes and $0.12 a minute, usage is $648. On Bland's plans the choice is a lower rate with a fee or a higher rate without one: 2,000 × 2.7 = 5,400 minutes × $0.12 = $648 plus $299 = $947 on Build, against 5,400 × $0.14 = $756 on Start. On price alone the no-fee plan is cheaper here by $191, and the fee plan only wins above $299 ÷ $0.02 = 14,950 minutes a month. But Start caps calls at 100 a day and 10 at once. Two thousand calls a month averages about 67 a day over 30 days, so a shop whose busy Mondays run past 100 calls needs Build whatever the sum says. That is a good example of why you compare plans at your volume and not at the vendor's example volume. It also shows the limits of any rule of thumb: re-run the sum when your volume changes.

At this size the per-interaction model also deserves a quote. Gorgias lists no voice prices, but if resolved calls were charged like its AI Agent interactions at the $0.90 Basic and Pro rate, 1,400 of them would cost $1,260, which is more than usage billing at these call lengths. Per-interaction pricing wins when calls run long or complicated, and loses on short ones, so the call length decides.

Text after the call

A useful finishing move is to have the agent send a text message with the tracking link at the end of the call, if you have the caller's consent and a sending number. That turns the call into a record the customer keeps, and it cuts repeat calls. Texts cost money too: a message of more than 160 plain characters is split and billed in segments, and a curly apostrophe or emoji in the text can cut the segment size to 70 characters, so keep the message short and plain. Check your provider's page for current per-message prices before adding it to the sums.

A mistake that shows up in practice

A shop launches an agent without testing numbers over a bad line. A caller on a train reads out order 10427, the agent hears 10247, finds a different customer's order and tells the caller it has been delivered. The caller is confused, and the other customer's order details have been disclosed. Two fixes close the gap. The agent must verify a second detail, the email on the order, before reading anything. And it must read the order number back and wait for a yes. After the change, the shop adds three noisy-line test calls to its standard set. Calls where the second detail fails twice go to a person, with no information given.

What to report each month

Keep a one-page report. Calls received. Calls resolved by the agent without a person. Calls transferred, with the top three reasons. Average call length. Total spend divided by calls, which is your real cost per call. Complaints that mention the agent. And a column for the unanswered question: what callers asked that the agent could not answer, which feeds the next change to your site or the knowledge base. If the cost per call is below the cost per call of the person handling the same calls, and complaints are not rising, extend the agent's hours. If not, narrow its job or switch it off for a month and compare. A trial that can be reversed is the safest way to learn what the agent is worth to you.

Callers should also know they are speaking to an automated system, and call recording and disclosure rules vary by place, so check the ones that apply to your customers and tell callers at the start of the call. The knowledge base the agent reads matters as much as the platform; how to train a voice agent on your business covers writing it, and logistics voice agents for status calls looks at the same job from the carrier's side.

Cheaper alternatives to try first

Many status calls exist because the shop has not answered the question elsewhere. Before paying per minute, check these. Put a tracking link in the dispatch email and on a status page. Add the tracking link to a text message at dispatch. Put a "Where is my order?" box on the site with an order-lookup form. Let a chatbot answer the same question in writing; can an AI chatbot handle order tracking and returns looks at what that involves. Each of these removes calls instead of answering them, which is cheaper than any per-minute rate. After you have done them, count the calls that remain; that is the number to cost out.

A four-week trial

Week one: log every status call for a week with minutes spent and whether it was simple. That gives you volume, average length and the simple share. Week two: build a read-only lookup and run 25 test calls with real order numbers, including bad ones. Week three: take after-hours calls only and review them all. Week four: calculate cost per call (usage plus platform fee plus number, divided by calls), resolved share, transfers, and complaints about the agent. Decide using the sums: if the saving is under a few hundred dollars a month and the setup took a week, you may prefer the cheaper alternatives. If the number is larger or your calls pile up after hours, extend to daytime overflow.

Keep a note of anything callers say that your website does not answer, since those are the questions to fix at source. For a wider view of AI across orders, stock and suppliers, see best AI for ecommerce operations. If you want help running the call count and the sums, that is the sort of thing my AI implementation consultation covers.

Questions about order-status voice agents

How much does an order-status call cost with an AI agent?

At listed rates of roughly $0.07 to $0.31 a minute plus telephony, a three-minute call usually costs between about 25 cents and a dollar. Your figure depends on the platform, the models and how long callers stay on the line. Measure it over a month.

Can the agent look up orders itself?

Only if you connect it to your shop or order system through an integration. Without one, it can take a message but cannot see tracking. Decide which fields it may read and read-only access is the safe starting point.

How does it know the caller owns the order?

Require two details the caller would know, such as the order number and the email on the order, before reading anything out. Never read out full addresses, and hand over any call where the details do not match.

Is a voice agent worth it for a small shop?

Usually only if you take more than a few hundred status calls a month, or miss calls out of hours. Below that, a tracking page and an email reply cost less. Compare cost per call with the time your team spends today.

Further reads

• AI Voice Agent for a Hotel Front Desk: Calls to Hand Over and Keep — Which calls to hand over and which to keep.
• AI Voice Agent for Plumbers: A 15-Call Test Before It Takes Real Jobs — A 15-call test plan you can adapt for a shop.
• AI Tools for Ecommerce Websites: What a 50-Orders-a-Month Store Needs — A need-versus-skip checklist for a store shipping about 50 orders a month, with free tiers, break-even sums and the volume at which each skipped tool pays.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Retell AI, Vapi and Bland AI pricing pages, Gorgias pricing page, Twilio documentation on SMS character limits, read in November 2026.