Is this process worth automating? How to run the numbers before hiring an AI process automation company in Mallorca

Almost every automation conversation starts the same way: "this takes up so much of our time". It may well be true, but "so much" isn't something you can decide on. Before you speak to an AI process automation company, it pays to arrive with something firmer: what the process costs today, what it would cost once automated, and which part of the work will still need a person.

This post isn't about choosing tools or about what not to automate. It's about running the numbers with your own data, in a simple spreadsheet, so the decision is yours rather than the excitement of the moment. It works just as well for a business in Palma as for one in Ibiza, Menorca or anywhere else in the Balearic Islands.

Why the numbers usually come out wrong

Three mistakes keep coming up:

• Time is estimated from memory. Nobody really knows how long a task takes when they do it in bits between phone calls.
• What's left after automation is forgotten. Reviewing, correcting and handling the odd cases still takes time.
• The only comparison is the project price. AI automation has costs that carry on afterwards: model usage, maintenance and adjustments when something changes.

Fix those three and most decisions become fairly obvious.

Step 1: measure the process as it is today

Pick one process and measure it over two normal weeks. In the Balearics, avoid measuring during the quietest or the busiest week of the season, or you'll get a distorted picture. For each case, note:

1. How often it happens, per day or per week.
2. How long it takes each time, with a timer or by noting start and finish.
3. Who does it and what an hour of that person costs the business.
4. How often it goes wrong and what happens then: a wrong order, a duplicated invoice, a customer left waiting.
5. How many cases are unusual: missing data, different formats, requests that don't fit.

You don't need special software. A shared spreadsheet with five columns is enough. What matters is that the data comes from real cases, not from what you think happens.

Step 2: work out today's cost

With your measurements, the monthly cost of the process looks like this:

• Time: cases per month × minutes per case ÷ 60 × hourly cost.
• Errors: errors per month × the cost of fixing each one (correction time, discounts, repeat deliveries).
• Delays: if a slow process loses you sales or bookings, note it separately. It's harder to measure, so don't add it in blindly; treat it as a supporting argument.

Step 3: estimate the cost after automation

This is the part almost nobody does. A process automated with AI doesn't drop to zero. Ask about, and write down:

• Human review: who checks what the AI does, and how long per case? At first it makes sense to review everything; later, only a sample or the doubtful cases.
• Exceptions: the unusual cases you counted in step 1 will still land on someone's desk. Cost their time as you do today.
• Running costs: the AI model, integrations and hosting are paid every month. Ask the company to estimate them with your real volume, not a generic one.
• Maintenance: when a form, a supplier or your management software changes, the automation has to be adjusted. Ask how that's charged.

Step 4: compare and find the payback point

Monthly savings are today's cost minus the cost after automation. Divide the project price by those savings and you get the number of months to pay it back. Do it three ways: optimistic, realistic and pessimistic (for instance, with more exceptions than expected). If it only works in the optimistic version, wait or narrow the scope.

A hypothetical example

Picture a drinks distributor in Inca that receives orders from bars and restaurants by email and WhatsApp. Someone in the office types them into the management software by hand. The figures below are made up for the example; yours will differ.

• Over two weeks they count around 40 orders a day, about 6 minutes each, plus the odd quantity or product-code mistake every week.
• With automation, the AI reads the message and drafts the order, and the person confirms it in under a minute. Orders with new products or confusing messages (say one in ten) are still done by hand.
• In high season volume rises and so do the savings; in winter it drops. So they run the numbers on the average volume across the year, not on August.

With that sheet, the conversation with the supplier changes. The question is no longer "how much to automate our orders?" but "with this volume and these exceptions, what does it cost, and what will we pay each month to run it?"

When it's usually worth it

• The process repeats often and almost always the same way.
• Errors cost money or customers.
• Exceptions are a clear minority.
• The payback period is one your business considers reasonable, even in the realistic version.

When it usually isn't, or not yet

• The process changes every few weeks.
• It only happens a few times a month.
• The data needs sorting first (duplicate customers, products without codes, loose spreadsheets). Sometimes that's the profitable job to do first.

What to ask for once you have your numbers

• An estimate of monthly running costs at your volume, including high season.
• How doubtful cases are reviewed, and who decides what counts as doubtful.
• A trial on your own real cases before committing to the full project.
• What report you'll get to check the savings: cases processed, reviewed, corrected and sent to a person.

If a proposal can't answer these, you don't yet have anything to compare.

Next step

At Derek Solutions we work with businesses across Mallorca and the Balearic Islands on AI automation and, when it needs to connect to your existing systems, on custom software. If you've already measured a process, or would like help measuring one, email derek@dereksolutions.com or call +34 680 283 973 and we'll run the numbers with you before proposing anything.