What should it cost to automate order status and returns support for a small e-commerce brand?

Open your helpdesk and pull the last 90 days. Sort tickets into three buckets: where is my order, returns and exchanges, and everything else. You need your own share of each rather than an industry number, because your share is the one the build gets paid out of. Do not estimate this from memory. Count it.

Then time the work honestly. Sit with whoever answers tickets and watch five of each type end to end, including the part nobody counts: opening a second tab, checking the carrier, checking the warehouse export, deciding whether this return is inside policy, then writing the reply. Multiply the real minutes by the monthly volume, and cost it at what that person actually costs you per hour with payroll tax and benefits included, not at the figure you carry in your head. If you use an outsourced support desk this is even simpler, because you already get an invoice with a per ticket or per hour figure on it.

Now add what the queue costs you in ways that never show up on a payroll line, which is the part most owners never total. Refunds handed out because a reply came too late and the customer got angry. Duplicate orders placed because nobody confirmed the first one shipped. Chargebacks that started as an unanswered email. Reviews that mention response time. You will not get these to the dollar, and you do not need to. You need them roughly, so they stop being invisible.

That total is the number that governs the whole purchase. Set the payback bar yourself before anyone quotes you: if a build cannot clear its own cost out of that annual figure within twelve months, it is not worth buying, and you can put that sentence in front of every vendor, us included, at the start of the call rather than the end. It also tells you the answer in advance: if the queue costs you very little a year, no quote for automating it is a good quote.

First, volume and mix. Higher volume does not make the build harder, but it makes a bigger build worth doing, and it changes what you should automate. A brand with a heavy where is my order load and simple returns wants a narrow, fast build. A brand whose pain is exchanges, warranty claims, and partial refunds is buying something more careful, because the decisions are harder and the cost of getting one wrong is real money.

Second, how many systems hold the truth about an order. If everything a customer could ask about lives in your store platform, that is one connection and one set of edge cases. If the order is in the store, the shipment is with a third party warehouse, the tracking sits with two carriers, and return authorizations live in a spreadsheet, that is four sources of truth, four ways to be out of date, and four things to keep alive after launch. This is usually the single biggest swing in a quote, and it is the one buyers underestimate most.

Third, whether your returns policy is written down and decidable. An assistant can only enforce a rule that exists. If your policy says thirty days but your senior agent quietly stretches it for repeat customers, that judgment is either written into rules someone can read, or it is not automatable at all. Writing it down is real work. It is either a week of your own time before the project starts, which is free to you, or a line on the vendor's invoice. Do it yourself and the quote comes down.

Fourth, how many brands, catalogues, or regions the answers have to cover. One catalogue has one true answer per question. Four brands have four, and the risk is not that the assistant lacks an answer, it is that it hands a customer brand two's return window for a brand three order and sounds certain doing it. Shipping restrictions, warranty terms, and return windows all diverge, and every extra policy set is more to load, more to keep current, and more to test.

Fifth, where you put the approval line. Letting the assistant answer where is my order costs less than letting it open returns, which costs less than letting it touch money. Our standing recommendation is that it answers status freely, opens returns strictly inside written policy, and leaves anything that moves money behind a one click approval by a person. A refund issued in error costs you the cash and then the time spent unwinding it, while the approval click costs a few seconds a case. That is a design decision with a price, and it belongs on the quote as a line rather than as an assumption.

A quote you can evaluate names the exact ticket types in scope and, just as importantly, the ones explicitly out of scope. It lists every system it will touch, names the connection method for each, and says who creates the credentials. It states what happens when the assistant is not sure, which means naming the person it hands the conversation to and confirming the whole thread goes across with it. And it names what done looks like on the day, in terms you can check without a developer: how many tickets it resolved without a human, how many it handed over, and how many it got wrong.

It also separates two kinds of money that vendors blur constantly. There is what you pay the vendor to build the thing, and there is what you pay your own suppliers afterward: model usage, helpdesk seats, any platform subscriptions the build depends on. Everything we build runs in accounts you hold, so those suppliers bill you directly and nothing passes through us with a margin added on top. A single blended monthly figure hides which part is the meter and which part is margin. Ask for it split, in writing.

Ours states the schedule plainly: milestone based, always split in half, 50% at kickoff and 50% at launch. A typical build ships in 14 to 21 business days, and a smaller single workflow project in 7 to 10. Every project includes one week of post launch support, and continuing support after that is an optional monthly retainer priced to how much you want covered. Everything runs in accounts you own and log into, so if we part ways you keep the build rather than losing access to it.

If what arrives is one line and one number, it is not a quote. It is a hope with a total at the bottom, and the distance between the hope and the actual work gets settled later as change orders, at the exact moment you have paid a deposit and lost your leverage.

Ask where each answer comes from, and whether the assistant can show its source. For order status the source should be the live order record, not a summary someone synced last night. For returns it should be your written policy, quoted. An assistant that answers from a general model without reading your data is decoration with a text box attached, and it will invent a delivery date in its first week.

Ask what happens when it is not sure. Guessing should disqualify a vendor on the spot. The correct behavior is to stop, say it does not know, and hand the conversation to a named person with the full history attached. Ask to see that path demonstrated, not described.

Be careful with integration promises, because that is where most disappointment starts. If you name your warehouse system, your carrier, or your helpdesk and the vendor says yes we support that inside thirty seconds without asking which fields you need, treat it as a warning rather than a reassurance. Any connection of this class works one of three ways: a documented API, a prebuilt app or connector, or a person or script logging into a portal because there is no other way in. The third is a workaround, and workarounds cost more to build and much more to keep alive. On our side, the systems we publish for this service are your own documents plus Slack, Teams, Intercom and Zendesk, and if your order truth lives somewhere else the honest answer is not that we already support it. It is that we check your system first, confirm your plan exposes the fields, confirm the rate limits, and confirm whether tracking is pushed to you or has to be polled, and only then say yes or no. Insist on that sequence from whoever you hire.

Last, ask for their numbers with a basis attached. Ours, for VOT Distribution, a multi-brand e-commerce distributor: two AI storefront assistants in production, which is measured from systems we operate; $120K in generated sales opportunities and $7K closed from those opportunities so far, which are client-reported from the client's own pipeline; and a 12% campaign response rate, also client-reported. At the current campaign pace that projects to roughly $1M in pipeline, and that is a projection, not a result. Round numbers with nobody's name attached to who measured them are advertising, and you should read them as such.

If your ticket volume is low, do not buy anything. A queue that takes one person a few hours a week is cheaper left alone than automated by anyone, us included, and a build quietly demands maintenance attention that never appears on a quote. Automate what hurts by lunchtime, not what merely annoys you on a Friday.

If your where is my order volume is high because your shipments are genuinely late, automation is the wrong purchase and it will make things worse, because you will deliver the bad news faster and at scale. The fix is upstream: the warehouse, the carrier, or the delivery promise on your product page. Send proactive shipping updates and cut the contacts at the source before you buy anything that answers them.

If your returns policy is not written down, write it first. There is nothing to enforce until it exists, and you will discover during the writing that two people on your team have been applying different rules, which is worth finding out for free rather than paying a vendor to discover it in week three. The same goes for your catalogue and order data. Wrong data does not stay quietly wrong once something is reading it out to customers all day at machine speed.

If the pain is seasonal and shows up for six weeks a year, price temporary help against a permanent build honestly before you commit. And if cash is tight, do not buy the audit either. Do the 90 day ticket count yourself, take the free call, or use our free intake, where an engineer writes back within two business days with a ranked map of the three places AI or automation would pay back fastest, and if the honest answer is that nothing pays yet, the map says exactly that. The $4,500 audit earns its keep when you have several candidate projects and no defensible way to rank them. If you already know that order status and returns is your one bleeding queue, skip the diagnosis and ask for a quote on that single fix.