Start with a repeated moment of friction

Choose one task that happens often enough for the irritation to be real. It might be moving an enquiry into the right place, reminding a customer for missing information, preparing a weekly update or handing work from sales to delivery.

Keep the starting point concrete. ‘Our administration is too manual’ is a useful signal, but it is too broad to improve. ‘Every accepted quote is copied into three places before delivery can begin’ gives you something the team can observe together.

The best evidence is usually the workaround people have already created. A private spreadsheet, a copied email template or a daily message asking for status is not necessarily bad practice. It often shows where the current process is failing to give someone the information or prompt they need.

Use the four-question readiness check

Talk through the task with the people who do it, then answer these four questions in plain English. A ‘not yet’ answer is not a failure; it tells you where a small process change will be more valuable than rushing into a build.

  • Is the trigger clear? Someone should be able to say exactly what starts the process: a form submitted, a quote accepted, a date reached or a status changed.
  • Is the expected outcome clear? Define what ‘done’ looks like, including the record, message, handover or decision that should exist at the end.
  • Is the information reliable enough? If names, dates, stages or contact details are usually missing or differently recorded, tidy that habit before asking software to reuse it.
  • Is there a named owner for exceptions? Automation can handle routine steps, but a person still needs to notice and resolve the work that does not fit the normal route.

Separate the routine from the judgement

A useful first automation normally removes a predictable administrative step, not a decision that depends on context. Sending an acknowledgement when a complete request arrives is routine. Deciding whether a customer needs a bespoke response is judgement.

Map the flow in two columns: what should happen every time, and what needs a person to decide or check. The first column is where automation is most likely to help. The second tells you where a prompt, queue or better information might support the person without pretending the decision has become automatic.

This distinction also makes the work safer. A clear exception route prevents an awkward customer case, incomplete request or unusual commercial decision from being quietly pushed through a workflow that was only designed for the ordinary cases.

Choose the smallest helpful change

There are three sensible outcomes from the check. The process may be ready for a small automation. It may need a process fix first, such as a required field, shared naming convention or clearer handover. Or it may need a named owner and a regular review before any new technology is useful.

Pick the smallest change that removes a visible source of delay or rework. For example, instead of rebuilding the whole sales process, make sure each accepted quote creates one task for the delivery lead with the key information already attached. Let the team use that change before adding the next one.

Run it alongside the current process first

For the first few weeks, keep a simple eye on the new flow. Check whether the trigger fired at the right time, whether the information arrived complete, whether the right person could see the next action and whether exceptions were visible rather than lost.

You do not need a large benefits case to learn from this. Note how long the task took before, how often someone had to chase it, how many corrections were needed and what the team found confusing. If the change is not helping, adjust it or stop it. A small test is there to teach you, not to defend a tool choice.

Make ownership part of the design

Before you move on, agree who checks that the automation is still doing the right thing, who can change it and where the team records an issue. Use shared accounts and permissions that survive staff changes, rather than building an important routine around one person’s login or inbox.

Good automation should make the work calmer: fewer repeated requests, clearer handovers and less time spent asking what happened. It should not make the process mysterious. If the team cannot explain the trigger, the result and the exception path, it is not yet simple enough to rely on.

The aim is not to automate everything. It is to remove the repetitive steps that distract capable people from customers, delivery and the decisions that genuinely need their attention.

Make the next step useful.

Bring us the workaround, the repeated report or the AI idea that has not found an owner yet.

Discuss a challenge