How we pick what to automate first

Every business has dozens of processes that could be automated. The temptation is to start with the one that annoys you most, or the one a vendor pitched you last week. Both are bad starting points.

We've helped enough businesses through this that we now have a reliable framework for picking where to start. It's not complicated, but it does require being honest about how your business actually works rather than how you think it works.

Why the order matters

Starting with the wrong process doesn't just waste time. It damages your team's confidence in AI, makes the business case harder for the next project, and often means you spend months on something that delivers modest savings when a different starting point would have paid for itself in weeks.

According to McKinsey's research on automation, companies that take a structured approach to prioritising automation see returns three to five times higher than those that pick projects opportunistically.

The order you automate in is genuinely one of the highest-impact decisions you'll make.

Our four-factor framework

We score every candidate process against four criteria. Each one matters, and the combination tells us where to start.

1. Frequency and volume

How often does this task happen, and how many times per week or month? A task that happens 200 times a week offers far more automation value than one that happens twice a month, even if the individual task takes longer.

This is where the boring, repetitive work shines. Invoice data entry. Customer email triage. Order status updates. Stock level checks. These aren't glamorous, but they happen constantly, and that's exactly what makes them valuable targets.

We look for processes where you can reduce the time your team spends on high-volume repetitive work, because that's where the hours add up fastest.

2. Complexity and variability

How much does the process vary from one instance to the next? A task that follows the same steps every time is straightforward to automate. A task that requires judgment calls, exceptions, or creative thinking is harder and riskier.

That doesn't mean complex processes can't be automated. It means they shouldn't be first. Start with the predictable stuff. Build confidence. Learn what works. Then tackle the trickier processes with a team that already understands how AI operates in your environment.

The UK government's guidance on AI adoption makes a similar point about starting simple and building capability before attempting more complex applications.

3. Current cost

What does this process cost you today? Not just in direct labour, but in errors, delays, and opportunity cost. A data entry task might take one person three hours a day, but if the errors from that manual entry cause invoicing disputes that take a week to resolve, the true cost is much higher.

We map the full cost chain for each process. Direct time. Error correction time. Customer impact. Delays to downstream work. The process that looks modest on paper sometimes turns out to be the most expensive when you follow the knock-on effects.

This is particularly important for ops and finance teams building a business case. The numbers need to reflect reality, and reality is usually messier than the first estimate.

4. Integration difficulty

How hard is it to connect AI to the systems involved? Some processes live entirely within one application, with clean data and a decent API. Others span three legacy systems, a shared inbox, and a spreadsheet that Dave maintains on his personal laptop.

We're not afraid of messy integrations, but we'd rather tackle them second or third, once you've already seen results and your team is comfortable with the approach. It works alongside your existing systems, but some systems are easier to work alongside than others.

How the scoring works in practice

We rate each process from 1 to 5 on each factor:

• Frequency: 5 means daily/constant, 1 means rarely
• Simplicity: 5 means highly predictable, 1 means every instance is different
• Cost impact: 5 means expensive (time, errors, delays), 1 means minimal cost
• Integration ease: 5 means simple to connect, 1 means deeply entangled systems

We multiply frequency by cost impact (that gives us the potential value) and multiply simplicity by integration ease (that gives us the feasibility). Then we plot them on a simple 2x2 grid.

Top right (high value, high feasibility): start here. These are your quick wins.

Top left (high value, low feasibility): plan for later. Worth doing, but needs preparation.

Bottom right (low value, high feasibility): maybe. Do these if they're nearly free, skip them otherwise.

Bottom left (low value, low feasibility): don't bother. At least not yet.

A real example

A distribution company we worked with had a list of 14 processes they wanted to automate. The managing director was keen to start with customer demand forecasting, which was the most exciting option and the one he'd read about in an industry magazine.

When we scored everything, demand forecasting came out as high value but low feasibility. Their data was scattered across three systems with no consistent product coding.

What scored highest? Purchase order processing. It happened 150 times a week, followed nearly identical steps every time, cost the team roughly 20 hours of work weekly, and lived entirely within their ERP system.

We automated purchase order processing first. It took four weeks. The team got 20 hours a week back. The confidence that built meant the data cleanup project for demand forecasting (the harder, bigger prize) got enthusiastic support from the same team that might have resisted it three months earlier.

Most clients see results within 8 weeks when they start with the right process. Starting with the wrong one can take twice as long and deliver half the value.

What usually wins

After doing this exercise with dozens of businesses, certain categories come up repeatedly as strong first candidates:

Document processing

Invoices, purchase orders, delivery notes, expense claims. High volume, predictable format, clear rules, and usually painful to do manually. The Federation of Small Businesses highlights document handling as one of the most common areas where SMEs benefit from automation.

Email triage and routing

Customer enquiries, supplier communications, internal requests. If someone in your business spends hours sorting and forwarding emails, that's almost always a strong first project.

Data entry and reconciliation

Moving information between systems, matching records, checking for discrepancies. Tedious, error-prone, and exactly the sort of work AI handles well.

Status reporting

Pulling data from multiple sources into a weekly or monthly report. The data exists; it just needs assembling. AI can do this in minutes rather than hours.

What usually shouldn't go first

Some processes are genuinely better left until you've built some automation experience:

Anything customer-facing without human review is a harder first project. Automated responses, chatbots, personalised marketing: the stakes are higher and the judgement calls are trickier. Get your internal processes sorted first.

Processes that depend entirely on undocumented knowledge are also better left until later. If only one person knows how the process works and the rules live in their head, you need to capture that knowledge before you can automate it. That is a documentation project first, an automation project second.

Highly regulated processes (compliance, financial reporting, anything where an error has legal consequences) can be automated, but they need more careful testing and oversight. Not ideal for a first project.

The conversation we have with every client

We'll show you exactly where to start. That's literally what our initial engagement is designed to do. We map your processes, score them against this framework, and give you a prioritised list with estimated timelines and savings for each one.

It's not theoretical. It's based on what your business actually does, with your systems, your team, and your data. We handle the technical side entirely. You tell us how the business works, and we tell you where AI will make the biggest difference fastest.

Get your starting point

Our free AI opportunity report does exactly this analysis for your business. You'll get a clear picture of which processes to automate first, what the likely savings are, and how long it would take.

No commitment. No sales pitch. Just a practical look at where AI fits in your business.

Get your free AI opportunity report and find out where to start.