INSIGHTS

Business process automation: what to automate first

Business process automation for firms still running on paper: choose a useful first workflow, fix the handover and keep someone responsible for exceptions.

By Harrison Jones · · 7 min read

Picture a contractor's office on a Monday morning. A job sheet is on the desk, the delivery note is in WhatsApp and the office manager is trying to work out whether the work is ready to invoice. This is an illustrative example. It gives business process automation a useful starting point: stop making someone reconstruct the job before they can do their own.

I'd start there, well before anyone demonstrates an AI assistant. Before buying anything, decide which handover you want to make dependable. Then choose the smallest change that gets it done.

Start with the handover

My starting question would be: what happens after this document arrives?

Follow a completed job sheet from the engineer to the person raising the invoice. Ask what they need, what they check, what they copy and who they contact when something is missing. Use an actual recent job, with the people who handled it. Don't settle for the process everyone thinks they follow.

Write down the agreed route in plain English: the engineer submits the job record, the supervisor confirms completion, the office checks the purchase order, and accounts prepares the invoice. Your route may be different. The point is to make the responsibility visible before making it automatic.

If nobody can agree when a job is complete, pause the software conversation. That's a decision for people to settle, and no software setting will make it for you. Our Assess stage starts with understanding the operation rather than choosing a tool.

I'd also ask whether the document still earns its place. Don't spend money moving a form more efficiently if nobody uses the information it contains.

Choose a first workflow with a clear finish

I'd choose a repetitive task with a recognisable start, a dependable rule and a result someone can check. Avoid starting with a disputed commercial decision or an instruction that could commit the business without review.

For a paper-heavy service firm, I'd put these candidates on the shortlist:

  • A completed job record goes to the office queue, with missing information clearly flagged.
  • A purchase request reaches the named manager, who approves or rejects it before anyone places an order.
  • A supplier bill becomes a draft for accounts to review against the supporting documents.
  • An outstanding internal task produces a reminder for its owner, rather than relying on someone remembering to chase.

Those are suggested designs. Your existing software may or may not support them. Check the exact steps, access and subscription before committing.

Define the finish just as carefully as the start. For a job-record workflow, I would use something like: accounts has a complete record, or the supervisor has a visible request to resolve what's missing. Sending an email is only part of it. Someone must know what to do next.

Fix the information before connecting systems

For an illustrative maintenance business, imagine replacing a handwritten completion sheet with a short digital form. I'd ask the office manager and an engineer to agree the essential fields together: job reference, site, work completed, outstanding work and whether supervisor review is needed.

Keep it short. If you can't explain what a field is used for, question why you're collecting it.

Make the job reference consistent across the form, the office record and the document folder. Decide which record is the one people should trust when copies disagree. Also decide how someone corrects a mistake. A completed form shouldn't become impossible to amend just because a workflow has started.

Microsoft documents a Forms workflow in Power Automate that starts when a response is submitted, retrieves its details and uses a condition to choose an email action. That's a concrete building block, and you'd still have to build the job management around it.

I'd use that sort of pattern only after agreeing the information and the recipient. A perfectly delivered message with an ambiguous job reference is still an ambiguous message.

And test the form where the work happens. Ask the engineer to complete a realistic record using the device and working conditions they normally have. Keep a clear fallback for when the digital route isn't usable.

Use what you have before adding another platform

Before asking a supplier to build a connection, check what your current systems can already do. Start with the accounting package and office tools you actually use before looking at new automation products.

For example, Xero's UK accounts payable page describes uploading or forwarding bills so that key details are extracted into draft bills ready for approval. If you use Xero, investigate that existing route before commissioning a separate bill-entry workflow. Keep accounts responsible for checking the draft against the source document and your own purchasing records.

For office handovers, the useful distinction is often what starts the action. Microsoft's Power Automate trigger guidance distinguishes event-triggered, manually triggered and scheduled flows. That gives you different design choices: act when something arrives, let a person deliberately start the next step, or run a check on a schedule.

My preference would be a deliberate button press where someone needs to confirm the work is ready. Don't make a workflow automatic merely because it can be.

Keep AI out of the first version unless you can explain the job it needs to do. Moving a structured record to a named person doesn't, by itself, justify adding an AI judgement. If you later use AI to interpret free-text notes, make the checking responsibility explicit.

Design the exception route before the happy path

Here's the test I'd set before approving a pilot: show me what happens when the information is wrong.

Try a missing job reference, a repeated submission, an unreadable attachment and a record sent to the wrong supervisor. These are proposed test cases; I'm making no claim about how any particular product behaves. Agree what you want the process to do in each case, then check that the implementation does it.

For the illustrative job-record workflow, I would want incomplete records held in a visible queue. The supervisor should see what's missing, the office should know why the record isn't ready, and someone should be responsible for resolving it. I would not let an incomplete record trigger a customer invoice.

Name an owner for the automation too. Ask who checks failures, who can pause it and who updates the process when responsibilities change. Write down the manual fallback in language the office can use without the person who built the workflow being present.

Be especially clear about repeated submissions. If an engineer corrects a record and sends it again, decide whether that updates the original or creates a new item. Don't leave the business rule for the software installer to guess.

Run a small trial and judge the whole job

This week, I'd pick a live workflow, sit with the people on either side of the handover and trace recent examples. Record active handling time separately from time spent waiting. Note missing information, corrections and chasing. Treat that as a starting picture. It doesn't promise a saving.

Agree the pilot's success test before building it. I'd look for a complete record reaching the right person, fewer manual copies and a visible route for exceptions. Include the engineer's effort as well as the office manager's. If the admin has just moved out of the office, it's still there.

After the trial, repeat the same observations and ask the people doing the work what became easier and what became harder. Review the awkward cases as well as the clean demonstration. If checking the automation takes more effort than the old task, simplify it or stop it.

That's the discipline behind Mise's Assess · Design · Implement · Operate & Optimise approach. Each stage is agreed separately and has to earn the next. A useful assessment might lead to a better form, a clearer responsibility or a modest connection between existing systems. It doesn't have to lead to a bigger technology project.

My view is simple: automate a handover you understand. If it's a mess you're tired of looking at, sort the mess first. Start with the operation; use technology only where it helps.

If you'd like to work out which handover to tackle first, talk to us.

This article is general information, not legal, tax or regulatory advice. Check the linked sources before making a decision.

Want a second pair of eyes on your operation?

Thirty minutes with a founder. Tell us how the work runs and where it gets stuck, and we will tell you honestly whether we can help.

Book a conversation