This process does not need AI. It needs fewer steps.
Before asking where to add AI to a process, count the steps. In many workflows the real cost is not the work itself. It is the handoffs and the waiting between them.
A customer asks for a quote.
Someone opens the email. Then a spreadsheet, to check what was quoted last time. Then the price list, because the spreadsheet is three months old. Then a document, to write the quote. Then the email again, to send it. Then a reminder in the calendar, so somebody follows up next week.
Six tools. One quote.
And this is usually the moment when somebody asks:
Where should we add AI?
Wrong question.
Not because AI is useless. It is genuinely good at some of this.
It is the wrong question because we are trying to improve a process nobody has properly counted yet.
Count the steps before you improve any of them
Take one real request that arrived last week. Not a hypothetical “typical” one — an actual request.
Then write down every point where somebody had to do something for it to keep moving.
Not only the work.
The moving.
A process people casually describe as three or four steps often gets much longer once every lookup, copy, approval, handoff and follow-up is included.
And the more useful discovery is usually not how many steps there are.
It is where the time went.
The work is rarely the expensive part
Writing a quote may take a few minutes.
Finding the correct numbers can take longer.
But neither necessarily explains why the quote reaches the customer two days later.
Often, it simply sat.
It sat in an inbox until somebody opened it.
It waited for a colleague to confirm a price.
It remained in drafts until somebody returned to it.
The actual work might add up to twenty minutes while the elapsed time stretches from Thursday to Monday.
That distinction matters because the two problems have different fixes:
- Steps are things somebody does. You reduce them by removing unnecessary work.
- Handoffs are the gaps between those actions. You reduce them by removing waiting and unnecessary transfers.
Automating an individual step makes that step faster.
Removing a handoff can make the whole process faster.
And that often requires no sophisticated technology at all. Sometimes the information simply needs to be available in the right place when the person needs it.
If you want a quick estimate of what a repetitive task is costing before deciding whether to automate it, our task calculator does the arithmetic in about a minute.
What the shorter version looks like
The quoting process above does not need six tools and a trail of manual transfers.
It can be four steps.
1. Request. The customer sends the details through a form, an email that can be read automatically, or whatever channel they already use. Nobody re-types the same information somewhere else.
2. Pricing. The existing rules are applied automatically: current price, quantity, discount tier, or whatever else the business already uses to calculate the quote. The rules are not new. They are simply written somewhere software can read them instead of living in somebody’s memory, spreadsheet or inbox.
3. Review. A person checks the exceptions.
4. Send. The quote goes out and the follow-up is scheduled.
Four steps.
And the human step in the middle is deliberate.
The third step is the one that matters
We keep a person in the process on purpose.
Not because software cannot calculate a price.
Because the interesting ten per cent of cases are often the ones where the normal rule should not decide the outcome.
The customer who has worked with you for nine years.
The order that looks ordinary but is clearly the beginning of something larger.
The request that technically fits the rules but should probably be declined.
Automating the predictable work creates the time to handle those cases properly.
That is the actual trade.
And it is worth making explicit because:
Automation is not about removing people. It is about removing work that never needed a person.
So when is AI the right answer?
Sometimes it clearly is.
AI becomes useful when the input is genuinely unstructured and the variation is real. For example:
- reading specifications that arrive in different PDF layouts;
- summarising a long email thread before someone takes over;
- understanding what incoming messages are actually about rather than matching a fixed keyword;
- extracting useful information from text that cannot be handled reliably with a simple rule.
Those are problems where writing all the rules becomes the difficult part.
What AI does not fix is a process with eleven handoffs.
Put a model in the middle of that process and you may simply end up with eleven handoffs and one faster step.
The quote can still sit in drafts over the weekend.
The order matters more than the technology:
- Count the steps.
- Remove the ones that exist only because information is somewhere else.
- Automate what remains and is genuinely repetitive.
- Then ask whether anything left requires judgement that a rule cannot express — and consider AI for that.
Many businesses discover there is less left for step four than they expected.
That is not a disappointing result.
It is a cheaper one.
Where to start
Pick the process people complain about most.
Annoyance is useful evidence. It often points directly to places where someone is repeatedly doing something the system should already have done.
Take one real example.
Count the steps.
Mark which ones are work and which ones are waiting.
Then decide what actually deserves technology.
If you want a second pair of eyes on that process, that is what our 20-minute check is for.
We look at one process with you, identify what is worth automating and what is not, and send you the list afterwards.
And if the answer is that nothing is worth automating yet, we will say so.
This started life as a carousel on LinkedIn. That is the short version; this is the one with the reasoning in it.