Most SMB owners we speak to are eager to automate their processes - they are well aware of the repetitive manual work happening in their companies. What they usually can't answer is where to start.

That's a harder question than it looks. Not only because not everything is worth automating, but because a bad pick can disrupt a process the team has spent years getting right. That was exactly the situation with one of our clients, so they paid to have the question answered before building anything.

This article is how we answer it. It's a guide to looking at the work your business already does and coming out with a ranked list: what to automate now, what can wait, and what to skip. You can run it yourself in a couple of hours, and the list is useful even if you decide not to build anything in the end.

Step 1. Write down how the work is actually done

Not how it's supposed to work, but how it works in reality. Go from employee to employee (or by role, if several people do the same things) and break every process they run into small steps. For each step, write down:

• What starts it - a customer email, a submitted form, the first of the month, someone just remembering, etc.

• Who is involved and how long it takes

• What information they need, where they get it from, which systems or files are involved

• What comes out - an invoice, a record in the CRM, a message to a colleague or a vendor, a payment, etc.

• What happens when it goes wrong, and how often does that happen? Try to remember the last time it did.

Do this at the level of individual steps (actions), not whole departments. "Onboarding a customer" is not a single process you can evaluate. It usually contains 4-5 different sub-processes (collecting the required documents, granting access, and so on), and they score very differently - which is the whole point of the next step.

Tip: talk to the people actually doing the work, not to the ones who manage it or designed it. The written procedure is rarely the same as what happens in real life.

One thing worth noticing here - by the time you have done this for your main processes, you have something you didn't have before: a written record of how your business actually runs. Normally most of that lives in a few people's heads, and once those people leave the company, the knowledge leaves with them. So whether or not you decide to automate anything in the end, this kind of map is worth having - for training new people, for handing work over, or for anyone who ever wants to buy the company.

Step 2. Ask what a computer can't do

Before you score anything, drop the steps that aren't candidates at all. Ask these questions:

• Does it happen on a screen? Anything physical comes off the list - someone driving somewhere, meeting a customer, fixing something in a room, etc.

• Can data get in and out of the systems involved? If the only way to get the information out is a person reading it off a screen and typing it somewhere else, automation is still possible, but it will be fragile and expensive. Not a step we would recommend starting with.

• Is the judgement the whole point? Some work looks repetitive, but the real value is a person making a call and maybe even taking responsibility for it. If that's the case, automation removes the value.

Everything that fails these questions comes off the list. Not forever - some of it becomes possible later - just not where you want to start.

Step 3. Score what's left

Score each step from 1 to 5 across the following categories (higher is better):

• Frequency - how often does it run? 1 - a few times a year, 5 - several times a day

• Cost - what does one run consume (time, people, paid services)? 1 - a couple of minutes of one person's time, 5 - hours, several people, or software you pay for by usage

• Importance - what happens if it's late or wrong? 1 - nobody notices, 5 - a customer notices, or money/compliance is involved

• Predictability - does it run the same way each time? 1 - every case is different, 5 - same steps, same shape of data, nearly every time

Predictability is the most underrated of the four. It's the difference between a robust system and the one your team quietly stops using because it needs adjustments too often. So even a step that scores 5 on frequency, cost and importance, but 1 on predictability, is a bad first project.

Let's take a short-term rental operator as an example:

So what do we have:

• Routine in-stay questions are the ones to prioritize for automation. Questions like how the air conditioner or the TV works, what the wi-fi password is, where the spare key is, and so on. Dozens a day, a few minutes each, and it adds up to more than a day a week of someone's time. Most of them are highly predictable, while anything unusual can be escalated to the team.

• Maintenance and complaints are expensive and very important, but they are often highly unpredictable, which makes them a bad candidate for a first automation project.

• Review requests are frequent and highly predictable, yet they shouldn't be your first choice either, because they are too simple and too cheap - it takes less than a minute to copy, paste and send one.

• Pre-booking enquiries and pre-arrival messages are the other strong candidates for what to automate next.

Step 4. Do the numbers on your top choices

Take the two or three highest scoring candidates and calculate the following:

• Time saved per month - minutes per run, times runs per month. Multiply that by what an hour of that person's time actually costs your business (salary, insurance, paid leave, etc.)

• What else it's costing you - errors that get fixed later, customers waiting, work that doesn't get done at all.

• What it costs to build, run and maintain. Ask for a fixed price before you commit.

After that, calculate the payback period - how many months until the solution pays for itself? Under six months is good for a first project, but it can vary from industry to industry.

Example: answering questions during the stay. Assuming thirty properties, roughly 120 bookings a month and around 4 routine questions per booking, that's about 500 questions per month. Even at 3 minutes per answer, that's around 25 hours a month. Depending on the salary of the support/operations manager, it costs between $200 and $400 a month. And if we take into account the cost of the distraction, of someone having to be available all the time, the number goes up even more.

What we would tell you to leave alone

Some processes are better left as they are, for example:

• Anything that runs less than several times a year - even a perfect automation will save you a few hours and will still need some maintenance.

• Anything where every case is genuinely different - you will spend more hours describing the rules and the corner cases than the automation will ever save you.

• Anything you are about to change anyway - automating a process you are planning to redesign means building it twice.

• Anything where a person deciding is the product, or your competitive advantage - if clients are paying for your judgement, don't automate it.

One at a time

Whichever process you choose to automate first, build that one and nothing else until it's working and your team is using it without being reminded. There is a temptation to start several automations at once, since you want your business to be more efficient. But we strongly recommend being patient here. One well-performing automation is better than 3 half-finished ones and a confused team - they are asked to change how they work in three places at the same time, none of which is properly tested yet, and you can't tell which of the three was worth the money and the effort. Finish one, then use what you have learned to do the next one faster.

Where to start this week

Block two hours in your calendar, pick the process you think has the most potential for automation, and break it down step by step, the way it actually runs. You will know by the end of the day whether it's worth automating, and you will have the first page of the map of your business either way.

If you'd rather have someone do this for you, that's what our audit is. It takes a few hours of your team's time, you get the ranked list at the end, and it's yours to keep whether you decide to build anything or not.