The short answer
The most repetitive task is not always the best first AI project, because frequent work is often quick and already half handled. The real cost usually sits in slower work: waiting between steps, information searched for again and mistakes fixed weeks later. Ranking tasks by minutes, frequency and rework shows where automation pays back fastest.
Ask a team what to automate first and most people point at the task they do most often. It is a natural instinct. If something happens fifty times a day, surely taking it away is the biggest win.
Sometimes it is. Often it is not. Frequency tells you how often a task happens. It does not tell you how much it costs, how often it goes wrong, or what it holds up downstream.
01Why is frequent work often cheap?
Plenty of repetitive work takes a few seconds each time and is already half handled by the tools in place. Filing an email, tagging a ticket, copying a reference number. Automating it is satisfying, but the hours you get back are small, and the system you build still has to be maintained.
Feels like the big win
- Happens many times a day
- Everyone mentions it
- Easy to picture automated
Where the hours often are
- Work that waits for one person
- Information found again each time
- Mistakes fixed weeks later
02Where does the real cost hide?
The expensive work is usually messier. It involves a judgement, a search or a hand-off. A follow-up that slips because nobody owned it. A quote that needs three systems open at once. A question from a customer whose answer exists, but in a document from last year that someone has to dig out.
These tasks happen less often, so they do not come up first in a conversation. But each one can take a long time, and the delay they cause spreads. A late reply loses a lead. A wrong number in an order shows up at month end, when fixing it costs far more than catching it would have.
Something being frequent does not make it expensive.
03How do you tell frequent from expensive?
When we map a workflow, we put three numbers next to every step, even if they are rough estimates: how long it takes, how often it happens, and how often it has to be redone. Multiply the first two and add the cost of the third, and the ranking often looks very different from the one people started with.
- Minutes per occurrence, measured on a normal day rather than remembered.
- How many times a week it happens, across everyone who does it.
- How often it goes wrong, and what fixing it involves.
- What waits on it: the steps, people or customers held up until it is done.
04An example, in outline
Take a common setup: enquiries come in through a website form and land in a shared inbox. Ask the team what slows them down and the answer is usually the volume of emails to sort. So the obvious first project is an assistant that tags and files them.
Map the work end to end, though, and sorting is often a small part of it. The time goes into what happens next: looking up the company, deciding who should reply, writing a first answer and remembering to follow up when nobody hears back. The slow step is not the filing. It is the hours between an enquiry arriving and a person replying to it.
The first idea
- Tag and file every email
- Saves a few seconds each
- The reply is still late
After mapping the work
- Research the sender on arrival
- Draft a reply for approval
- Follow up if nobody answers
Both projects are about the same inbox. Only one of them changes how fast a customer hears back.
05When is repetitive work the right place to start?
None of this means repetitive work is a bad starting point. When a task is frequent, slow and rule based, it is often the best first project there is: data entry from documents that always look alike, routing requests by type, chasing approvals that follow the same path every time.
The point is to check before committing. A week spent mapping the work is cheap. A quarter spent automating the wrong job is not, and it makes the next AI project harder to argue for.
FAQCommon questions
- What should you automate first with AI?
- Automate first the workflow with the fastest payback, not the one that happens most often. Multiply minutes per occurrence by how often it happens each week, then add the cost of rework. Tasks that are frequent, slow and rule based, such as data entry from similar documents, are often the best first project.
- Is repetitive work always worth automating?
- No. Repetitive work that takes seconds and is already partly handled by existing tools gives back little time, and the automation still needs maintaining. A useful test is to ask what would get faster if the task disappeared tomorrow. If the answer is nothing much, it is not the place to start.
- How do you measure the cost of a task before automating it?
- Put three rough numbers next to every step of a workflow: how long it takes on a normal day, how many times a week it happens across the team, and how often it has to be redone. Add what waits on it, such as customers or other steps held up until it is done.
Written by
Abdallah, consulting and ai automation
Abdallah starts with the work itself: sitting with your team, mapping where the hours go and deciding what is worth automating and what is not.
