Follow one piece of work
Start with an enquiry, report or handover. Where does information arrive, who checks it, and where is it entered again? We map the steps and the exceptions before choosing an integration or automation. Sometimes removing an unnecessary step is more useful than making it faster.
Use rules where the rules are clear
Routine routing, reminders, data synchronisation and recurring reports are candidates when the inputs and responsibilities are understood. An example might be taking an approved record from one tool, creating a task in another and flagging missing information for review. The exact approach depends on the systems and permissions available.
Use AI where interpretation adds value
AI may help organise unstructured text, suggest a category or extract fields from varied documents. Outputs need validation and a clear review path. We compare a narrow AI-assisted workflow with a straightforward rules-based alternative; we do not treat a plausible answer as a verified result or automate consequential decisions without appropriate human control.
Make failures and results measurable
Agree a baseline: time per task, volume, errors and time spent checking. Design bounded retries, duplicate protection and an alert when a person needs to act. Include maintenance and review time when judging the result. Our capacity calculator can help frame a conversation, but its estimates are not a promise of savings.
Questions before you start
Is every repeated task worth automating?
No. Low-volume tasks, changing processes and unclear ownership may need a simpler change first. The benefit should justify implementation, maintenance and exception handling.
What should we share initially?
A plain-language description of the process, the tools involved and where work gets stuck. Use anonymised examples and leave out passwords, identity documents and confidential customer information.