Quick Takeaway
Choose processes to automate by scoring each one from 1 to 5 on frequency, time cost, error cost, rule clarity, data availability and risk, then rank them. The best first candidates happen often, follow clear rules, cost real time and carry low risk. Pilot one, measure it for 30 days, then move to the next.
Key takeaways
- Frequency and time cost tell you the value; rule clarity and data availability tell you how hard it will be.
- Never automate a broken process. Fix it first, or you will just do the wrong thing faster.
- Pick one pilot with a fast payback and low risk, and measure it before expanding.
- A two-week time log beats guessing. People are usually wrong about where their hours go.
Most businesses have dozens of tasks that could be automated and no way to decide between them. So they either pick the most annoying one, which is often the hardest, or the most exciting one, which is often the least valuable. A simple scoring method fixes that. It takes an afternoon and prevents months of wasted effort.
This builds on what AI automation is and where it fits. The aim here is practical: end up with a ranked list and one clear pilot.
Step 1: Find the candidates
Step 1: Keep a two-week time log
Ask each person to jot down recurring tasks and roughly how long they take. People are usually surprised by how much time goes on copying, chasing and re-entering.
Step 2: List the repeated tasks
Look for work that happens daily or weekly and follows a similar pattern each time: lead intake, appointment reminders, invoicing, reporting, onboarding, data entry.
Step 3: Describe each in one line
Write it as trigger, steps and outcome, for example: "When a new enquiry arrives, we check it, reply, and add it to the CRM."
Step 2: Score each process
Score every candidate from 1 (low) to 5 (high) on the six factors below. For the first four, a higher score is better. For the last two, a higher score means an easier and safer project.
| Factor | The question to ask | A score of 5 means |
|---|---|---|
| Frequency | How often does it happen? | Many times a day |
| Time cost | How much time does each run take? | Long and repetitive |
| Error and rework cost | What does a mistake or delay cost us? | Mistakes are common and expensive |
| Rule clarity | Could you write the steps down as rules? | Clear, consistent steps |
| Data availability | Is the information already in a system we can connect to? | Structured data, easy to access |
| Low risk | What happens if the automation gets it wrong? | Little harm, and easy to catch and fix |
- Priority score
- The sum of the six factor scores, out of 30. Higher is a better first candidate. Treat it as a guide for discussion, not a verdict.
A worked example
The scores below are illustrative, to show how the method ranks real-world tasks. Yours will differ.
| Process | Freq. | Time | Error | Rules | Data | Risk | Total |
|---|---|---|---|---|---|---|---|
| New lead intake and first reply | 5 | 4 | 4 | 4 | 5 | 4 | 26 |
| Appointment reminders | 5 | 3 | 4 | 5 | 5 | 5 | 27 |
| Weekly reporting | 3 | 4 | 3 | 4 | 4 | 4 | 22 |
| Client onboarding documents | 3 | 4 | 4 | 3 | 3 | 3 | 20 |
| Handling complex complaints | 3 | 5 | 5 | 1 | 2 | 1 | 17 |
Appointment reminders and lead intake come out on top: they are frequent, rule-based, use structured data and are easy to check. Complex complaints score low despite consuming a lot of time, because they need judgment and the cost of getting them wrong is high. That is a task for a person, with AI assisting at most.
Step 3: Check the return
Hours saved per month = runs per month x minutes saved per run / 60
Monthly value = hours saved x loaded hourly cost
Net monthly benefit = monthly value - monthly running cost
Payback (months) = setup cost / net monthly benefitA pilot with a payback of a few months is a good first project. For how the running side is priced, see what AI automation costs.
Do not automate a broken process
If the current process is unclear, inconsistent or produces bad results, automation will repeat the problem at speed. Simplify and standardize the steps first, then automate the version you actually want.
Step 4: Run a pilot
Step 1: Choose one process
Take the highest-scoring candidate that a real person owns and cares about.
Step 2: Define success
Pick one or two measures before you build, for example time to first response, or hours spent per week.
Step 3: Build the simplest useful version
Keep a person in the loop for anything uncertain, and add logging so you can see what happens.
Step 4: Measure for 30 days
Compare against your baseline. Fix what you learn, and only then decide whether to expand.
Good first pilots often sit in CRM automation, and our 10-automation playbook lists the usual candidates.
Mistakes to avoid
- Starting with the hardest problem because it is the most painful.
- Automating without a baseline, so you cannot prove it worked.
- Trying to automate everything at once instead of one process at a time.
- Removing the human check before the logs show the automation is reliable.
- Ignoring the people who run the process today. They know its exceptions.
If you want a second pair of eyes on your list, our business process automation audit does this scoring with you, and you can book a free strategy call to start.
Which processes to automate: common questions
To compare shortlisted processes by value, follow our guide to calculating automation ROI or use the free AI automation ROI calculator.
Once you’ve picked the process, choose the platform: see n8n vs Zapier for which tool fits simple versus complex workflows. Still building your shortlist? Browse the 100 AI automation ideas checklist.
What makes a good process to automate?
A good candidate is frequent, follows clear rules, uses data that is already in a system, costs real time or causes costly errors, and is low risk if it goes wrong. Lead intake, reminders and routine reporting are typical examples.
How many processes should I automate at once?
One to start. Get a single pilot working reliably and measure it for about 30 days before adding another. This keeps risk low and teaches you what your team needs.
Which processes should I not automate?
Processes that are unclear or broken, tasks that depend on nuanced human judgment, and anything where a mistake is expensive and hard to undo. These may still benefit from AI assistance, but with a person making the final decision.
How do I estimate the time an automation will save?
Multiply how often the task happens by the minutes it takes each time, using your time log rather than guesses. Then subtract the time still needed for review and exceptions.
Do I need AI to automate these processes?
Often not. Many of the best first candidates are rule-based and need only a workflow tool. AI helps where the input is unstructured text, such as reading enquiries, summarizing documents or drafting replies.