What to Automate First in a Small Business: A Simple Scoring Method
Most owners know they "should automate more," but the list of candidates is long and vague. Picking wrong is expensive: you spend weeks on a clever workflow that saves twenty minutes a month. This is the method we use on first calls to pick the right starting point.
Step 1: List the work that repeats
For one week, have everyone jot down any task they do more than once that involves copying information, sending a routine message or checking something. Typical entries:
- Replying to new web enquiries
- Typing invoice details into accounting software
- Sending appointment reminders
- Updating the CRM after a call
- Pulling numbers together for a weekly report
Don't filter yet. You'll usually end up with 15 to 40 items.
Step 2: Score each task on four questions
Give each task a 1 to 5 score on each question:
- Frequency. How often does it happen? Daily scores high; quarterly scores low.
- Clarity. Could you write the rules on one page? "If the form says X, send Y" is clear. "Decide whether this customer is a good fit" isn't.
- Low risk. If it goes wrong once, how bad is that? A late reminder is mild. A wrong payment is not. Score safe tasks high.
- Pain. Does it cause missed leads, errors or after-hours work? The more it hurts, the higher the score.
Multiply the four scores. The highest totals are your shortlist. Multiplying rather than adding means a task that scores 1 on clarity or risk drops down the list, which is what you want.
Step 3: Check the plumbing
Before committing, confirm the tools involved can be connected. Most popular apps work with Zapier or Make, or have an API. If the data lives in a desktop program with no export, the job is harder, and that should lower its rank.
What usually wins
For service businesses, the first winner is almost always lead follow-up: it happens daily, the rules are clear, a delayed text is low risk, and slow replies cost real money. In a well-known Harvard Business Review analysis, companies that contacted leads within an hour were nearly seven times as likely to qualify them as those that waited longer.
Close behind are appointment reminders and invoice data entry. Reporting comes up often too, though it usually saves less time than people expect.
What to leave for later
- Tasks that need judgment every time (pricing a custom job, handling complaints)
- Processes that change every month
- Anything you can't describe without saying "it depends" a lot
AI can help with some of these by drafting or sorting, but keep a person making the call.
Fix the process before you automate it
If a task has five steps and two of them exist only because of an old workaround, remove those first. Automating a messy process just makes the mess happen faster.
A worked example
A hypothetical plumbing company lists 22 tasks. "Reply to web leads" scores 5 × 5 × 4 × 5 = 500. "Enter supplier invoices" scores 4 × 4 × 3 × 3 = 144. "Quote large remodels" scores 2 × 1 × 2 × 4 = 16. They start with lead replies, schedule invoices second, and leave remodel quotes to people.
How to run the one-week task log
The log only works if it's easy. Don't hand people a spreadsheet with twelve columns. Give them one shared note or a simple form with three questions: what did you do, how long did it take, and what tools did you touch. Ask them to log anything that took under fifteen minutes and that they've done before. Those small, repeated jobs are where automation pays off, and they're the ones people forget to mention in a meeting.
At the end of the week, merge duplicates. "Emailed the customer back," "replied to the website form" and "called the lead from Facebook" are usually one task: responding to new enquiries. Group them, then add up how often each one happened.
A scoring sheet you can copy
Set up a sheet with these columns:
- Task (one line, in plain words)
- Who does it today
- Tools involved
- Times per week
- Minutes each time
- Frequency, Clarity, Low risk, Pain (1 to 5 each)
- Score (the four numbers multiplied)
- Notes: anything that makes it awkward
Times per week multiplied by minutes each time gives you a rough weekly cost in minutes. It's worth having next to the score, because a task can score well but only save ten minutes a week. That might still be worth doing if it removes errors, but you should know.
Scoring guide: what a 1 and a 5 look like
Frequency. 5 means many times a day. 3 means a few times a week. 1 means monthly or less.
Clarity. 5 means a new hire could follow written rules and get it right. 3 means mostly rules, with a few judgment calls. 1 means it's different every time.
Low risk. 5 means a mistake is mildly embarrassing and easy to fix. 3 means a mistake costs a customer some time or annoyance. 1 means a mistake moves money, breaks a legal obligation or affects someone's health.
Pain. 5 means it regularly loses you work, causes errors or keeps someone late. 1 means it's a little tedious and nobody really minds.
Score as a group if you can. The person who does the task knows how messy it really is. The owner knows what a mistake would cost. You need both views.
Checklist before you build anything
- Can you write the steps on one page, including what happens when something is missing?
- Do all the tools involved have a connector or an API?
- Who gets told when it fails, and how?
- What will you measure to know it worked (reply time, errors, hours)?
- Is there a manual fallback for the first few weeks?
- Have you removed any steps that only exist because of an old workaround?
If you can't tick most of these, the task isn't ready yet. Spend an hour tidying the process and score it again.
Common mistakes
Starting with the most impressive idea. An AI agent that reads every email and decides what to do sounds exciting. A form that sends an instant reply and creates a CRM record is dull, and it will probably save more time in its first month.
Automating for one person's habits. If only one team member does a task a certain way, check that it's the right way before making it permanent.
Forgetting the exceptions. Most of the build time goes into what happens when a field is blank, a customer replies in an odd way or an app is down. Ask "what if" questions early.
No owner. Every automation needs a named person who checks it's working and fixes it, or calls whoever can.
Questions owners ask
How many tasks should we automate at once? One, then another. Running two or three projects in parallel makes it hard to tell what's working and splits everyone's attention.
Should we use AI for the first project? Only if the task involves reading or writing messy text, such as sorting free-form enquiries. Plenty of good first projects need no AI at all.
What if the top task is something we can't automate? Then move to the next one. The scoring is meant to give you a ranked list, not a single answer.
Next step
If you'd like help scoring your list, book a free call. We'll go through it with you and tell you honestly which items are worth automating.
