Automation for ecommerce: order to doorstep and back again
The processes an online store actually runs on — order routing, stock, fulfilment exceptions, returns and reviews — and which of them repay automating first.
Rebecca Pearson11 min read
On this page
- What we'll cover
- Where the time actually goes
- Order routing and the exceptions
- Stock, and the oversell problem
- Fulfilment exceptions, which are the real job
- Returns
- Reviews and the post-purchase window
- What order to build in
- The systems that have to talk to each other
- Frequently asked questions
- Related reading
Ecommerce automation is usually sold as marketing automation, which is the part with the most vendors and the least operational pain. The work that actually consumes a store's week is further back: orders that need routing, stock that disagrees between channels, deliveries that go wrong, and returns that arrive with no paperwork.
Those are the processes worth building, because they are high frequency, rule-based, and they fail expensively when nobody is watching. This page covers them in the order a store usually meets them.
What we'll cover
Where the time actually goes
Ask a store owner what eats the week and the answer is rarely campaigns.
Checking on orders that have not moved. Something was placed four days ago and has not shipped, and finding out why means opening three systems.
Reconciling stock across channels. The store says eleven, the warehouse says nine, the marketplace listing says fourteen, and somebody is about to oversell.
Answering "where is my order". The most common support message in ecommerce by a wide margin, and almost always answerable from data the store already holds.
Chasing suppliers and couriers about the specific items that are late.
Processing returns that arrive with a handwritten note or nothing at all.
All five are rule-based, frequent, and currently done by a person opening tabs. That combination is what makes them the right things to automate, and it is why a store that automates its operations usually gets more back than one that automates its marketing.
Order routing and the exceptions
The ordinary order needs no attention. The automation earns its place on the ones that do.
Route by rule, not by inspection. Which warehouse, which supplier, which shipping method, based on destination, weight, stock location and service level. A person applying those rules manually is a person applying them inconsistently by Friday afternoon.
Flag the orders that need a human, rather than reviewing all of them. High value, unusual quantity, a mismatch between billing and delivery country, a customer with a history of chargebacks, a delivery address that looks like a freight forwarder. That is a short list per day rather than every order.
Split orders deliberately. When part of an order is in stock and part is not, decide in advance whether to ship partially or hold, and encode that rather than leaving it to whoever is looking.
Push the confirmation immediately, with a realistic date rather than an optimistic one. Most "where is my order" messages are caused by a confirmation that promised something the fulfilment side never agreed to.
Get the order into the accounting system without anyone retyping it. This is the unglamorous one and it is a genuine recurring cost in stores that have not done it.
Stock, and the oversell problem
The failure with the worst customer experience and the most avoidable cause.
Sync in one direction, from one source of truth. The most common cause of overselling is two systems both believing they own the stock number. Decide which system is authoritative and make everything else follow it.
Sync on the event, not on a schedule. A fifteen-minute polling cycle is fifteen minutes in which the same last unit can be sold three times. Where a channel supports webhooks, use them.
Hold a buffer on fast-moving lines. Showing nine when you have eleven costs you two sales occasionally and prevents the failure that costs a customer permanently.
Alert on the reorder point, not on zero. By the time stock is zero the decision is already late. The threshold should reflect the supplier's lead time for that item rather than one number for everything.
Reconcile daily and report the differences rather than silently correcting them. Discrepancies between the store and the warehouse are information about a process problem, and a sync that quietly fixes them hides it.
Fulfilment exceptions, which are the real job
Happy-path fulfilment needs no automation; the courier's own systems handle it. The work is the exceptions, and they are entirely detectable.
Nothing has shipped after N days. The single most valuable check in ecommerce operations. An order sitting unfulfilled is a complaint forming, and nobody notices because nothing failed — it simply did not happen.
Tracking has not updated in N days. A parcel that stopped moving is lost, and the customer usually notices before the store does. Detecting it first turns a complaint into a proactive message, which changes the conversation entirely.
Delivery failed or was refused. Needs action now rather than at the end of the week.
Address looks wrong. Missing a line, a postcode that does not match the city, a suspiciously short street. Catching this before dispatch is far cheaper than after.
Tell the customer before they ask. A message saying a parcel appears delayed, with what you are doing about it, prevents the support ticket and reads as competence rather than failure.
These checks are a set of conditions over data you already hold, which is straightforward to describe and tedious to wire together across a store, a warehouse system and a courier. On CodeWords you describe what counts as an exception and what should happen for each, and Cody, the automation builder, builds it, connects it to the systems you already use, and deploys it. Automations connect to more than 3,000 integrations. The free plan covers light use, with Pro at $39 per month and Business at $100 per month as usage grows; details are on the pricing page.
Returns
The process most stores handle worst, because it is unpleasant and irregular.
Give the customer a route that produces data. A form or portal rather than an email, so the reason, the order and the items arrive structured rather than as prose somebody has to read and retype.
Approve the straightforward ones automatically. Within the window, unopened, standard reason. Reserve attention for the ones outside policy, which are the ones that actually need judgement.
Tell the warehouse what to expect before the parcel arrives, so it can be matched on receipt rather than sitting in a corner while somebody works out whose it is.
Refund promptly once it is received, and tell the customer. The gap between receipt and refund is where most return-related complaints come from, and it is usually a queue rather than a decision.
Report the reasons monthly. Returns data is product feedback arriving free. A line with a return rate three times the average is telling you something about sizing, photography or description, and nobody will notice without the report.
Reviews and the post-purchase window
The highest-return marketing automation, and the one closest to the operational work.
Trigger on delivery, not on dispatch. A review request arriving before the parcel is the most common own goal in ecommerce email.
Wait long enough for them to have used it. A few days for most products, longer for anything that takes setting up.
Suppress for anyone with an open support issue or a return in progress. Asking a customer who is currently unhappy to rate you is actively harmful, and the suppression rule is one condition.
Ask about the product, not the store, if what you want is product reviews. Two different questions frequently merged into one form that answers neither well.
Route the unhappy replies to a person rather than into a reporting dashboard. A poor review answered quickly is often withdrawn or amended; one nobody responded to is permanent.
What order to build in
First: the unfulfilled-order check. Cheap, catches the most expensive failure, and works before anything else is in place.
Second: stock sync from one source of truth. Prevents the failure customers remember.
Third: order routing and the exception flags. Once volume makes manual review impractical.
Fourth: the tracking-stalled check and proactive delay messages. This is where support volume drops noticeably.
Fifth: returns intake as structured data.
Sixth: reviews. Last, because it depends on delivery data being reliable, which the earlier work provides.
Building reviews first is the common sequence and it is backwards. It is the most visible and the least valuable of the six.
The systems that have to talk to each other
Worth mapping before building anything, because ecommerce automation is mostly a question of which system is authoritative for what.
The storefront owns the customer-facing catalogue, the order record and the payment. It is authoritative for what was sold and to whom.
The warehouse or fulfilment system owns physical stock and dispatch. It is authoritative for what actually exists and what actually left. Where these two disagree about stock, the warehouse is usually right and the storefront is usually the one being believed.
The courier owns the parcel's journey. Authoritative for tracking, and frequently slow to admit a problem, which is why a stalled-tracking check beats waiting for an exception notification.
The accounting system owns the financial record. It should receive orders and refunds rather than originate them.
The support tool owns the conversation, and it needs read access to all of the above to answer anything without a person switching tabs.
The single most useful hour in an ecommerce automation project is writing down which of these is authoritative for each piece of data. Most sync problems are two systems both believing they own the same number, and that is a decision rather than a technical fault.
Frequently asked questions
What should an ecommerce store automate first?
The check for orders that have not shipped after a few days. It takes an afternoon, it catches the failure that generates the most support contact, and it needs nothing else to be in place first.
How do we stop overselling across channels?
One authoritative source of stock, everything else following it, updates on events rather than on a polling schedule, and a small buffer on fast-moving lines. Overselling is nearly always two systems both believing they own the number.
Is marketing automation worth it for a small store?
Less than the operational work, initially. Abandoned cart and post-purchase flows are genuinely effective and most platforms include basic versions already. The operational automations save time every day and prevent the failures customers remember.
What about marketplaces with restricted APIs?
Many will provide scheduled exports even where the API is limited, and picking those up works. The important thing is that the marketplace does not become the authoritative stock source simply because it is awkward to write to.
How do we handle pre-orders and backorders?
Explicitly, and in the confirmation. The damage from a backordered item is almost always the customer not knowing, rather than the delay itself. Flag those orders separately so they are not caught by the unfulfilled-order check and treated as a problem.
Should returns be fully automated?
The straightforward ones within policy, yes. Anything outside policy wants a person, because that is where the judgement and the goodwill live. Automating the intake so the data arrives structured is worthwhile regardless of who makes the decision.
How much does this cost to run?
For a store with a few hundred orders a month, usually under fifty dollars monthly on a usage-priced platform, plus whatever your existing systems include. Check whether your store platform and shipping tool already do parts of this before adding anything.
What about peak trading periods?
Everything above matters more and the thresholds need loosening. A stalled-tracking check tuned for normal weeks will alarm constantly in December when every courier is slow. Review the thresholds before peak rather than muting the alerts during it, because muting is how a genuine problem gets missed in the week it costs most.