Which business processes should you automate first?

Why does the choice of first process matter?

The first automation sets the tone. A quick, visible win builds support and funds the next step. A slow project on a messy process makes people sceptical of everything that follows.

It also matters because adoption is broad but deep use is still rare. Stanford HAI's 2026 AI Index reports that organisational AI adoption rose to 88% of surveyed organisations in 2025, "though AI agent use remains early", with agent deployment in the single digits across nearly all business functions (Stanford HAI, AI Index 2026, Economy chapter; report page). Most organisations are still choosing where to start. Choosing well is the advantage.

Governments are setting expectations too. On 9 April 2026 Mauritius launched its National AI Strategy and FAIR Guidelines, which encourage organisations of all sizes, including SMEs, to adapt the guidance to their own context, scale and level of risk (Ministry of ICT; FAIR Guidelines). A risk-based choice of first process fits that approach.

How do you score candidate processes?

List ten to fifteen processes your team complains about, then score each from 1 (low) to 3 (high) on the factors below. Score risk the other way round: a process where mistakes are severe and hard to catch scores lower as a first candidate.

Factor	Question to ask
Volume	How often does it run each week or month?
Value	How much time, cost or customer impact is at stake?
Rule clarity	Could you write the steps and decisions on one page?
Stability	Has the process stayed the same for the past year, and will it next year?
Data quality	Is the input data structured, accurate and in a system?
Risk of error	If automation gets it wrong, is the damage small and easy to spot?

Add up the scores. The top two or three are your first candidates. Anything that scores low on rule clarity or stability goes to process redesign, not automation.

Which processes are usually good first candidates?

• Finance: matching supplier invoices to orders and receipts, routing purchase approvals, bank reconciliations, sending payment reminders.
• Sales and customer service: logging enquiries, sending quote follow-ups, routing requests to the right team, status updates to customers.
• HR: onboarding and leaver checklists, leave requests, collecting documents.
• Operations: stock alerts, delivery confirmations, recurring reports pulled from several systems.

These share a pattern: many repeats, clear rules and data that already exists in a system. They are also where staff time is most visibly wasted.

Update, 1 September 2026: the Mauritius Revenue Authority's e-invoicing stage for taxpayers with turnover above MUR 40 million is now listed as launched (MRA). For businesses in scope, invoicing moves to the top of the list, because e-invoicing exposes weak master data and approval steps. See is your business ready for e-invoicing?

What are the signs a process should be fixed before automating?

• Different people describe the steps differently.
• More than a handful of cases are "exceptions".
• Work waits for someone to chase it by email or phone.
• The same data is typed into two or more places.
• Nobody can say how long it takes end to end.

In that case, map the process as it really runs, remove steps that add no value, agree who owns each stage, and then automate the cleaner version. This is often where most of the benefit comes from. Our guide to replacing email and spreadsheet approvals shows the approach on one common case.

Update, 22 July 2026: BCG's July 2026 CEO research found that AI high performers are roughly seven times more likely to redesign workflows and reshape the business end to end with AI (BCG, 22 July 2026). Redesign first, automate second.

Who should be involved in choosing?

The people who do the work every day, the manager who owns the outcome, and someone from finance or risk who can see the controls. Staff know where time is lost; managers know what matters to customers; finance and risk spot where a fast process could become an unsafe one. A short workshop with all three, using the scoring grid above, gives a more balanced shortlist than any one of them alone. Record why each process was chosen or rejected, so the decision can be revisited when circumstances change.

Should the first automation use AI?

Not necessarily. Many good first candidates need only rules-based workflow automation. AI is useful where inputs are unstructured, such as reading documents or classifying emails, and should be added where it removes a real bottleneck. If you are considering an AI agent, check the use case against our AI agent readiness checklist.

Whatever you choose, take a baseline before you start: how long the process takes, how many errors occur, and how much staff time it uses. Without it, you cannot show the result.

Linkgrow's process redesign service scores and redesigns candidate processes before automation. The free gap finder gives a quick first view.

Frequently asked questions

What business processes can be automated?

Most processes with repeated steps, clear rules and data in a system can be partly automated: invoice matching, approvals, onboarding, reminders, reporting and routing of requests are common examples.

Should you fix a process before automating it?

Yes, if the steps are unclear, disputed or full of exceptions. Automating a broken process makes errors faster and harder to see.

How many processes should we automate at once?

Start with one or two, measure the result against a baseline, and use what you learn before widening the programme.

Do we need AI to automate processes?

No. Many processes need only rules-based workflow tools. AI helps where inputs are unstructured, such as documents and free-text emails.