Every merchant we talk to wants automation, and almost none of them can say what they want automated. The word has drifted into the same vague territory as "AI", a thing you are supposed to have rather than a problem you are trying to solve. This post is the version of the conversation we actually have in discovery: what automation costs to build, what it saves once it runs, and how to tell the difference between the work that pays for itself and the work that just looks impressive in a demo.
The hidden cost is the manual work, not the software
The biggest line item in running a digital business is rarely the storefront licence or the hosting bill. It is the human time spent moving data between systems that were never introduced to each other. An order arrives in Shopify or Magento. Someone re-keys it into the ERP. Someone exports a stock file and uploads it to a marketplace. Someone reconciles yesterday's payouts against the accounting tool by hand, on a Friday, with a spreadsheet that one person understands.
None of that work shows up as a cost on a dashboard, which is exactly why it grows unchecked. It hides inside salaries and "that's just how we do it." A useful first exercise costs nothing: for one week, have the ops team write down every task they do that involves copying information from one screen to another. The list is always longer than anyone expects, and it is the real automation brief, far better than "we want to automate things."
What automation actually is
Stripped of the marketing, automation in commerce is three boring capabilities. First, integration: getting two systems to share data so a number is entered once, not three times. Connecting Magento or Shopify to an ERP like Sage, to a PIM, to a 3PL, to the accounting stack. Second, event-driven workflows: making something happen automatically when something else changes, instead of a person checking and reacting. An order ships, the customer gets the right message, the stock decrements, the invoice is raised. Third, internal tools: giving the ops team a button that does the safe, correct thing, instead of a raw database or a fragile spreadsheet.
That is the whole field. It is not glamorous, and it is not magic. It is plumbing. But plumbing is what decides whether a growing business scales smoothly or spends every peak season hiring temps to copy-paste.
The trade you are actually making
Automation is a trade of engineering hours now against manual hours and human error forever. The maths is simple enough to do on a napkin, and we encourage clients to do it before we quote.
Take a task done daily that costs a person 30 minutes. That is roughly 125 hours a year. If automating it takes two engineering weeks and the workflow runs for three years, the build pays for itself comfortably inside the first year and returns free time every year after. Now take a task done once a quarter that takes an hour. That is four hours a year. Automating it almost never pays, and we will tell you so rather than bank the work.
The second half of the trade is error cost, and it is the half people forget. A manual stock sync that is wrong one time in fifty sounds fine until the fiftieth order is an oversell on your best product during a campaign. The value of automation there is not the minutes saved, it is the oversells, the chargebacks and the support tickets that never happen. When a manual mistake costs real money or real trust, automation is worth it even when the time saved is small.
Where automation pays, and where it does not
It pays when the task is frequent, stable and rule-based. Order-to-ERP sync, stock reconciliation across channels, catalogue publishing from a PIM, transactional messaging, returns workflows, scheduled reports the finance team currently builds by hand. These have clear inputs, clear outputs, and they do not change shape every week.
It does not pay when the task is rare, or changes every time, or needs real judgement. Negotiating with a supplier, handling the genuinely weird customer complaint, deciding next season's range. Automating those either fails outright or produces a brittle system that breaks the first time reality does not match the rule. The honest answer for that work is a better internal tool that makes the human faster, not a robot that replaces them.
On one B2B relaunch we worked on, the integration that mattered was the link between the storefront and the client's ERP, the unglamorous backbone that let a century-old parts distributor actually operate online at the volume the relaunch brought in. The headline was a double-digit lift in online revenue. The thing that made it survivable was the plumbing underneath it.
How we scope it
We start every automation engagement with a data-flow map, not a tool choice. One or two weeks to draw every system that holds a copy of your data, every place a human moves it by hand, and every point where the same number lives in two systems that can disagree. That map is a deliverable you keep whether or not you build with us, and it usually surfaces two or three quick wins worth more than the cost of the audit.
Then we sequence by payback, not by ambition. The first thing we automate is the highest-frequency, highest-error task on the map, because that is the one that funds the rest of the programme in saved hours. We resist the temptation to build the grand unified integration platform up front. Most businesses need five sharp automations, not one cathedral.
We also build for the day it breaks, because every integration eventually does. An API changes, a file format drifts, a third party has an outage. The difference between an automation that helps and one that becomes a liability is whether it fails loudly and safely, an alert and a clean retry, or silently corrupts data for a week before anyone notices. That defensive engineering is unglamorous and it is most of the real work.
Three questions to ask before you automate anything
First, how often does this actually happen, and how long does it take? If you cannot answer in numbers, measure for a week before you spend anything. The numbers decide whether there is a case at all.
Second, what does it cost when it goes wrong by hand today? If the answer is "nothing, we catch it," the automation is a convenience. If the answer is an oversell or a mispriced order, it is a risk control, and it is worth more than the time saved.
Third, will this task still look the same in a year? Automate the stable spine of your operations, not the part that is still changing every month. Automating a process you are about to redesign just means building it twice.
Automation done well is invisible. Nobody celebrates the order that synced correctly or the stock figure that was simply right. The payoff shows up as an ops team that grows revenue without growing headcount, and a peak season that does not require all hands on a spreadsheet. That is the work we like: not the magic button, but the quiet machinery that means the business runs while everyone sleeps.
Where to dig deeper
- Kilat Labs: Automation + Integrations, how we scope and build the integration layer
- Shopify Next Gen Events end the webhook glue era, the event model that makes modern automation cheaper
- Kilat Labs: E-commerce engineering, where storefront and integration work meet