Where to start with AI: how to choose your first AI workflow
Start with one workflow that is frequent, measurable and slow today, has a clear owner, and lives in a system AI can connect to, such as enquiry handling, invoice processing or company knowledge.
The best first AI project is one workflow that is frequent, measurable and slow today, has a clear owner, and lives in a system AI can connect to. Replying to enquiries, processing invoices, finding answers in documents and preparing reports are the most common starting points.
Why start with one workflow?
Company-wide AI strategies stall when they try to change everything at once. A single live workflow proves the value in numbers, shows your team how AI works alongside them, and connects the systems the next workflows will need. Each success funds and de-risks the next.
Four tests for a good first AI workflow
- Frequent: it happens many times a week, so small time savings add up.
- Measurable: you can count hours, response time, errors or conversions before and after.
- Owned: one manager is responsible for the result and can approve how AI works.
- Connected: the work happens in a system with data AI can reach, such as a CRM, ERP, inbox or document store.
Five workflows that pass the tests
- Enquiry handling: AI replies, qualifies and updates the CRM. Measured by response time and qualified leads.
- Invoice processing: AI reads, validates and matches invoices for approval. Measured by hours saved and errors caught.
- Company knowledge: AI answers staff questions from your documents with sources. Measured by time to answer.
- Management reporting: AI writes daily summaries of what changed and why. Measured by decisions made faster.
- Customer service: AI answers common questions on WhatsApp and email. Measured by resolution time and volume handled.
What to measure before you start
Record a baseline for two to four weeks: how many cases, how long each takes, how quickly customers get a reply, how often mistakes happen. Without a baseline, even a successful AI workflow is hard to prove.
How the first workflow goes live
- Discover: map the workflow and its systems with the people who do it.
- Design: decide what AI does, what people approve, and what success looks like.
- Build: connect AI to the systems and run it with approvals on.
- Optimise: measure against the baseline, improve, then relax approvals where AI has proven reliable.
Our AI readiness assessment takes a minute and suggests where your first workflow could be. For a full opportunity map, see AI transformation.
Frequently asked questions
What is the best first AI project for a business?
A workflow that happens many times a week, can be measured, has one owner and runs in a system AI can connect to. Enquiry handling and invoice processing are the most common first projects.
How long does a first AI workflow take to go live?
A focused workflow can usually go live within weeks once access to the systems it needs is in place, starting with human approval on its actions.
How do we prove an AI project worked?
Record a baseline before you start (volume, time per case, response time, error rate) and measure the same numbers after go-live.