How to choose which process to automate first
The most expensive mistake in an automation programme is not choosing the wrong technology. It is choosing the wrong first project — spending a quarter on something visible but low-volume, then losing the appetite to try again.
What makes a process a good first candidate?
A good first candidate is high-frequency, low-variability, sits on systems with usable APIs, and fails safely. Those four properties matter more than how painful the task feels. Frequency creates the return, low variability keeps the build small, API access decides the effort, and a low cost of error keeps the approval burden light.
Why frequency beats difficulty
Automation value is arithmetic: time saved per run, multiplied by runs per period. This sounds obvious and is almost universally ignored, because people nominate processes by how much they dislike them.
Consider two candidates. The first is a monthly reconciliation that takes a skilled person three hours and that everybody resents. The second is a two-minute data-entry task performed roughly ninety times a week that nobody mentions. The first consumes about three hours a month; the second consumes around twelve. The unglamorous one is worth four times as much, and it is also the easier build because a task repeated ninety times a week has, by necessity, become standardised.
How do you judge variability?
Ask the person who does the job how often they have to stop and think. A process where every instance follows the same path is cheap to automate. A process where one instance in five needs a judgement call is not a harder version of the same project — it is a different project, because the exception handling becomes the bulk of the work.
The trap is that people describe processes as more standard than they are. The documented version has three steps; the real version has three steps plus a shared spreadsheet plus a rule about a particular customer that only one person knows. Watching the work happen, rather than asking about it, is the only reliable way to find that out.
Scoring candidates side by side
Score each candidate one to five on the four axes below, then sort. The numbers are not precise, but forcing the comparison surfaces disagreements early, while they are still cheap to resolve.
| Axis | Score 5 looks like | Score 1 looks like | Why it matters |
|---|---|---|---|
| Frequency | Runs many times daily | Runs a few times a year | Determines total hours returned |
| Consistency | Every instance follows the same path | Most instances need a judgement call | Drives how much exception handling is required |
| System access | All systems have documented APIs | Key system is closed or on-premise legacy | Often the single largest effort multiplier |
| Cost of error | Mistakes are caught and cheap to reverse | Mistakes reach customers or the ledger | Sets the level of review and approval needed |
Does the payback arithmetic actually work?
Before committing, do the sum explicitly. Multiply hours saved per week by the loaded cost of the person doing the work, annualise it, and compare against the build cost plus a year of running costs. If payback lands beyond about eighteen months, the surrounding systems will probably have changed before the project repays.
Include the running costs honestly — model API usage, platform subscriptions, and the maintenance time when an upstream system changes its format. These are modest, but a business case that omits them is not a business case.
When the right answer is not to automate
Sometimes the process should be eliminated rather than automated. If a report is assembled every week and nobody can name a decision it has influenced, automating it just produces an unread report faster. If two systems hold the same data because of an old organisational split, consolidating them beats building a bridge.
A good assessment names these cases explicitly. Our own assessment process uses the same four axes, and the deliverable includes what we recommend leaving alone. If you would rather not score your own processes from the inside, a workflow automation audit is the paid version of this method, run by someone with no stake in the answer and ending in a ranked backlog. For the cost side of the decision, the companion guide on what AI automation actually costs breaks down where the money goes. The vocabulary for risk and consequence-of-error follows the NIST AI Risk Management Framework.
Questions we get about choosing a first project
Should we start with the process people complain about most?
Not automatically. Complaint volume tracks how unpleasant a task feels, not how much it costs. A tedious monthly reconciliation generates more grumbling than a two-minute task repeated ninety times a week, while the second one consumes far more hours. Score both and let the arithmetic decide.
How much time does a process need to consume to be worth automating?
As a rough filter, a process worth a first project usually consumes at least four to five hours a week in aggregate. Below that, the payback period stretches past the point where the surrounding systems are likely to change anyway, which quietly erodes the return.
What if the process runs across a system with no API?
Treat it as a substantially larger project rather than an impossible one. Workarounds exist — scheduled exports, screen automation, vendor support — but they are more fragile and more expensive to maintain than an API integration. If a comparable candidate sits entirely on modern systems, do that one first.
Can we automate a process we are about to change?
You can, but it is usually wasted money. Automating a process that is scheduled to be redesigned, or that runs on a platform due for replacement, means building against a target that is about to move. Change the process first, let it settle, then automate the version you intend to keep.
Free consultation
Tell us what is costing you the most time
Describe the process. You get a straight read on whether it is worth automating, roughly what it would take, and what we would tackle first — on a call that costs nothing.
- Under an hour, free, and no obligation
- Fixed price per stage, agreed before any work starts
- You own everything we build, documented and in your accounts
- If you do not need us, we will say so on that call
The form is the fastest way to reach us — we reply from a real person, usually within one business day.