What to automate first
What should your business automate first?
Start with a recurring task that has clear rules, usable information and someone who owns the outcome. This guide helps you choose a manageable first project.
A practical checklist
Five questions to ask before you build.
Does the work repeat often enough?
Look for tasks your team handles every day or week: entering enquiries, updating statuses, filing attachments or assembling the same report. A rare task can still matter, but frequency makes it easier to observe the process and compare results.
Can you describe a normal path?
Write down what starts the task and what “finished” means. If different people follow different rules, agree the process first. Automation repeats a process; it does not automatically resolve disagreement about how that process should work.
Is the information accessible and usable?
Identify where the data lives and whether the software allows an integration or export. Check missing fields, inconsistent names and duplicate records. A short data-cleaning step may be more valuable than starting with an AI model.
What happens when it goes wrong?
Consider the effect of an incorrect result. Filing a document in a review queue has different consequences from sending a customer a price or making a payment. Start with reversible actions and keep approval for consequential decisions.
Who owns the result?
Name the person who can answer process questions, review exceptions and decide whether the system is useful. Without an owner, an automated queue can become another place where work is forgotten.
Worked example
Enquiry capture is often a useful candidate.
Suppose a team copies enquiries from email into a CRM. The input is an email; the output is a record with an owner and a next action. That gives the project a clear boundary.
A first version might prepare a record and draft a reply for review. It can check for an existing contact and flag missing details. It does not need to decide pricing, promise a delivery date or send every message automatically.
The next step is to test realistic examples: repeat senders, unclear requests, attachments and a CRM outage. That is more informative than testing only one ideal enquiry. This is an illustrative scenario, not a claim about a client result.
Make a business case
Estimate capacity, then include the running cost.
A simple starting estimate is:
Tasks per month × minutes saved per task ÷ 60 = hours of capacity released
For example, 100 tasks with three minutes of repetitive work removed gives five hours a month. Those are illustrative inputs, not an expected saving. Review time, exceptions and maintenance reduce the net benefit, and freed capacity is not automatically a cash saving.
Compare that estimate with the build cost, software subscriptions, AI usage and support. Also consider quality: fewer missed handovers or more consistent records may be valuable even when the time saving is modest.
Your first brief
Bring a process, not a shopping list of AI tools.
- What starts the task, and how often does it happen?
- Which people and systems are involved?
- What does a correct finished result look like?
- What are the common exceptions?
- What should always need a person’s approval?
- How will you compare the result with the current process?
Use those answers to define a small, testable scope. Our AI implementation approach explains what happens next, from mapping the process to handover.
Let’s talk
Start with one business problem.
Tell us where work gets held up, which tools you use and what a better result would look like. We’ll discuss the practical options and agree a scope before any build starts.
Discuss an automation project