The five tasks to automate first in a ten-person company
Not the most impressive ones. The most profitable: the ones that come back every day, look the same as yesterday, and run through text.
In short Start with what comes back every day, looks like yesterday and runs through text. The five candidates: sorting and first replies to incoming requests, the quote from a brief, the weekly report, extraction from documents, the follow-up. The first is almost always sorting incoming requests: the gain is visible from day one.
When I ask a business owner "what would you like to automate?", the answer is almost always the most ambitious thing: prospecting, strategy, the thing that would double revenue. That's rarely where to start.
Good first candidates share three traits: they come back every day or every week, they look the same each time as the time before, and they run through text (emails, messages, documents, spreadsheets). They're boring tasks. That's precisely why they pay off.
Here are the five I find in almost every company this size.
1. Sorting and first replies to incoming requests
Emails, website form, WhatsApp: requests come in, someone reads them, sorts them from memory, and often replies with the same thing. An AI can read each message, classify it (quote, question, complaint, spam), extract what matters (who, what, by when), and prepare a draft reply based on your past replies. The human reads and sends.
This is almost always the first thing I put in place, because the gain is visible from day one and everyone in the company understands it.
2. The quote or proposal from a brief
A call with a client, some notes, and half a day to turn that into a quote. With your past quotes as templates and one page of rules (your prices, what you always include, what you never accept), a first version comes out in ten minutes. Not the final version: the version you spend twenty minutes on instead of four hours.
3. The report, the status update, the weekly check-in
Everything that consists of gathering information that already exists (tickets, calendar, tracking spreadsheet) and formatting it for someone else. It's the work people hate most, and it's the cleanest to automate, because the data is already there.
4. Extraction from documents
Supplier invoices, purchase orders, CVs, enrolment files, contracts: PDFs and scans from which ten fields need to be pulled into a table or a system. It's tedious, error-prone, and an AI does it well, with a human check on the cases it flags as doubtful.
5. The follow-up
Missing documents in a file, the quote with no answer, the unpaid invoice. What doesn't get done, because nobody has time to check every day what's pending. A simple automation that looks every morning at what's stalled and prepares the follow-ups, which you approve in one go, recovers money and files that would have been lost.
How I choose, concretely
For each of these tasks, I ask the person doing it three questions:
- How many times a week, and how long each time?
- What changes from one time to the next? (If the answer is "almost nothing", it's an excellent candidate.)
- Where is the information needed, and can we access it without asking someone?
I multiply time by frequency, check whether the data is reachable, and take the top two or three on the list. Not the most impressive. The one that gives someone an hour a day back starting next week.
The rest, the ambitious part, comes after. And it comes much more easily once the data has been tidied and the rules written down for the five boring tasks.