What does AI automation deliver? Work it out yourself

With any automation plan, the conversation turns to price within five minutes. What it delivers usually stops at “working more efficiently”. You cannot do anything with that. You cannot base an investment on it, and afterwards you cannot read from it whether it worked.

We covered the cost side separately in the article on what AI automation really costs. This article is about the other half of that sum: what stands against those costs, how you work it out yourself up front using figures you already have, and why part of the return leaks away if you do nothing with it.

Do not count in percentages, count in hours on one process

“Twenty percent more efficient” is a sentence from a sales meeting. Twenty percent of what? Of whose working day? Measured how? Nobody in your business can check that sentence, which is exactly why it is said so easily. A year later there is nothing to hold it against.

So drop down a level. Pick one process, not your whole business, and express the return in hours per week on that process. At a wholesaler: retyping supplier documents currently costs around four hours a week, and after the integration we expect one. At an installation company: sorting and forwarding the shared info mailbox costs five hours a week now, and roughly an hour and a half afterwards.

That is a sentence you can put in front of your own people. And they can contradict it. That last part is the whole point: a return nobody can contradict is also a return nobody can check.

A return you cannot express in hours on one process is usually a return you have not investigated.

What does AI automation deliver? Four figures you already have

You do not need a financial model or a consultant to get a first answer. You need four figures, and all four come out of your own business.

1. How often does it happen? Count it over a normal week, not your busiest one. How many documents, how many emails, how many job sheets, how many quote requests. If you cannot count it, you cannot promise it either.
2. How many minutes does one instance cost now? Do not ask, time it. Sit next to someone for an hour and count along. If you ask, you almost always get a figure that is too low: people do not count the searching, the waiting for an answer and the starting over.
3. How much is left after automation? Never zero. There is still checking, handling exceptions and someone pressing approve. Work with an honest remainder, even if the supplier says it runs fully automatically. If you end up at zero, you have missed something.
4. What is an hour worth at your business? Not gross salary. Use what an hour actually costs you, including employer contributions, holiday pay and the hours nobody writes down. If this work would otherwise go to a temp or an external office, use that rate. If the hour could have been billable, use your own hourly rate.
5. What does it cost to build and to run? That figure comes from the quote: build hours, monthly usage costs and a realistic line for maintenance. Put it next to the return, not underneath it.

Here is an example with invented but recognisable figures, so you can see the shape of it. A wholesaler processes forty supplier documents a week. Each document currently takes about six minutes to retype and check, four hours a week in total. After an integration that prepares the lines as a proposal, a minute and a half of checking per document is left: one hour a week in total. The gain is three hours a week, well over a hundred and thirty hours a year.

Now put the costs next to it. Say the build takes twenty hours of straightforward work at €110 an hour, so €2,200, plus a core session at €596 and some monthly costs. Now the question is concrete: are those hundred and thirty hours a year worth that, and what is an hour worth at your business? That is a different conversation from “twenty percent more efficient”, and you can have it at your own kitchen table.

You only cash in saved time if you give it a destination

This is the part missing from virtually every business case. Saving three hours a week does not mean three hours of work disappear from your business. It means three hours come free. What happens to them is a decision, and no system makes that decision for you.

If you do not make it, the time leaks away. It goes into work that was already lying around, into longer meetings, and into the things that quietly expand when there is room. Afterwards everyone says it has become “a bit calmer” and nobody can point to where the gain is. That is not a technical problem. That is a decision nobody made.

So agree before the build where the hours go. There are only three honest destinations.

• More work with the same people. The freed-up hours go into work that earns money: more quotes out the door, more jobs scheduled, more customers served. This is the only destination you see back in your revenue.
• Doing work that keeps being postponed. Job costing, credit control, updating price lists, overdue maintenance on your own admin. Not a euro of extra revenue, but it prevents damage you do not see now and pay for later.
• Not replacing one for one when someone leaves. If someone leaves or retires, you do not have to put the whole role back. This is the only destination that genuinely reduces payroll costs, and it takes a long time.

What is not on the list: making people redundant because an integration took over their work. At SMEs that virtually never happens, and anyone who builds their calculation on it will be disappointed. Saved hours rarely become saved payroll costs. They become moved hours, and moved hours are only worth something if you know where to.

The returns nobody counts

Alongside the hours there is a second layer of return that rarely makes it into a calculation, while business owners often name it afterwards as the most important one.

• Fewer mistakes. A wrongly retyped article number becomes a wrong delivery, and that costs you a trip, a credit note and an irritated customer. How often that happens and what one instance costs, you know better than anyone.
• Responding faster. A quote that goes out the same day instead of three days later wins more often. You do not know in advance how much more often, but you can put your win rate before and after side by side.
• Work that no longer depends on one person. As long as only one colleague knows how supplier prices get updated, every holiday that colleague takes is a business risk. A process that is written down and actually followed removes that risk. You notice it the moment someone falls ill.
• Less waiting in the chain. An engineer who submits the job sheet that same evening instead of on Friday means the invoice goes out days earlier. That is working capital, and you can see it in your own books.
• Calmer work. Hard to put a figure on and real all the same: fewer Friday afternoons that overrun, less friction about who forgot what, less work piling up on one desk.

You do not have to force those things into euros. What you do have to do is pick one countable figure alongside them before you start, and measure it once. Number of credit notes per month. Days between finished work and the invoice going out. How often a week a customer calls to ask where their order is. Skip that baseline and the conversation afterwards becomes a matter of gut feeling, and gut feeling rarely beats an invoice.

The honest side: at first it costs you time

The first few weeks you pay in. Someone has to explain how the work really goes now, sit in on the tests, check the output and flag exceptions. Count on a few hours a week from exactly the person who knows the process best, and that is usually the person who is already busiest. Those hours appear on no invoice and they are real all the same. If you do not schedule them, the project stalls at the worst possible moment.

On top of that, a new process rarely runs well straight away. In the first weeks you correct more than you save, because you do not trust it yet and you check everything. That comes with the territory. The real gain starts once the trust is there and the checking gets lighter, and that often takes longer than the build itself. Promise yourself time savings from week one and you will disappoint yourself.

And then there is the outcome nobody wants to hear: sometimes it simply does not pay off. A process that happens twice a month and takes ten minutes is four hours a year. You do not build an integration for that, however tedious the work is. A process that does happen often but runs differently every time is just as poor a candidate: the time you save goes into straightening out the exceptions.

If we work that through during the core session and the answer is no, we say so. That is a cheap no: €596 to find out you should not make an investment of several thousand. That is the cheapest return in this whole article.

When you need us and when you do not

For the calculation above you do not need us. Four figures, an afternoon watching on the shop floor and a calculator get you further than any consultant’s report. In fact: if you make it yourself, you also believe it, and that saves a lot of discussion later.

We are useful when the figures do not come from one system, when the process runs across several packages that do not talk to each other on their own, or when you want an independent view on whether the gain is really there. In a half-day core session (four hours) we follow one process from start to finish, count the actions and work the return through out loud, with your figures. Building afterwards is charged per hour: €110 for straightforward work, €165 for complex development. If the sum does not add up, that is the answer and we leave it there.