Start with a repeated task that has a clear finish line and a person who can check the result. Measure how the job works today, test AI on a small sample, and include review time when you compare the two. A task that saves time after checking is a stronger candidate than a flashy demo.
Choose a job someone already does
Look at the work between your existing tools: collecting information, drafting routine replies, preparing reports and following up on missing details. Choose a specific piece of that work with a clear input and output.
A useful first scope might be preparing a draft reply from an incoming enquiry. ‘Automate customer service’ leaves too many decisions undefined. Who receives the draft? What information may it use? Who handles a complaint or a request outside the normal process?
Write down the starting point
Use recent work to establish a baseline. Record how often the task happens, the time it takes to finish, and the corrections people already make. Name the person responsible for deciding whether a new output is acceptable.
- Input: what starts the job, and where does that information live?
- Output: what must exist when the job is finished?
- Quality: what makes a result usable, and what makes it fail?
- Exceptions: which cases need a person to decide?
- Owner: who checks the work and maintains the instructions?
Compare a few candidates
Use this worksheet to compare real tasks in your operation. It is a discussion tool, not a universal scoring formula.
| Question | Stronger first candidate | Needs more preparation |
|---|---|---|
| Does it repeat? | Enough examples to test | Rare or constantly changing |
| Can you explain the result? | A clear, checkable output | Success depends on unwritten judgment |
| Are the inputs available? | Known, accessible records | Missing or conflicting information |
| Can you catch a mistake? | A reviewer checks before action | An error reaches a customer immediately |
| Does it free useful capacity? | Time goes to a named priority | Nobody knows what the time is worth |
Run a small test before connecting everything
Reprise’s implementation approach starts with an agreed scope, a goal and the people doing the work. Nick’s five-day exercise gives you a simple way to test a task before investing in an integration.
Try ordinary inputs and an awkward case. Keep the original records so you can compare outcomes. Count preparation, review and rework as part of the new process. If the result needs rebuilding by hand, narrow the task or improve the instructions before expanding it.
Make the next decision from the test
At the end, you should have a short process description, examples of acceptable work, a list of exceptions and a measured comparison. That is enough to decide whether to automate, keep testing manually or leave the job with your team.
Use the time recovered for a specific purpose: responding sooner, clearing a backlog or giving a specialist more room for higher-value work. Hours freed are capacity; they become a business result when that capacity is used.
Sources behind this guide
Explore Nick’s original material and the Reprise approach this guide draws on.
- WebsiteReprise AI: implementation approach Reprise AI
- Newsletter · August 12, 2026Give AI one task for five days Nick Puru