Key Takeaways
- Start with one workflow people already run by hand, not with a tool or a model.
- A named person owns the data and the sign-off. The model owns neither.
- The perimeter decides what AI may draft and what stays human, before anything is connected.
- A pilot is done when the workflow runs for real users against a measure you wrote down first.
Pilot scope worksheet
Fill one worksheet per workflow. Empty lines mean the pilot is not scoped yet.
- Workflow in one sentence: Which job changes? (For example, enquiry intake, handoff between teams, or a weekly report. Not “use AI across the business.”)
- Today’s steps: List how the work runs now, step by step, and who does each step.
- Where it hurts: The one step that is slow, repeated, or error-prone. This is the pilot target.
- Inputs and data: What the workflow reads (forms, emails, documents, a CRM) and where that data lives.
- Data owner: The named person who can approve access to that data and withdraw it.
- Perimeter: AI may: Draft, sort, summarize, or route inside this workflow only.
- Perimeter: AI may not: Send to customers, publish, sign, make legal or financial decisions, or touch data outside the list above, unless a person approves each case.
- Human checkpoint: Where a person reviews output before it leaves the workflow.
- Measure: The one thing you will compare before and after, written down before the pilot starts.
- Done looks like: Who uses it, for how long, and who accepts it or stops it. Also name what happens next: expand, change, or switch off.
One workflow per worksheet. A second workflow is a second pilot, not a bigger first one.
Why AI projects stall before they start
A stalled AI project usually starts with a tool, not a job. Someone buys a subscription or connects a model, and then looks for work to point it at. Nobody owns the data, nobody drew the perimeter, and nobody agreed what done means. That is the same naming problem we mapped in AI-native work vs common buyer pains: it shows up as a technology problem, but it starts as a missing scope.
Start with the worksheet. If the job is to change one internal workflow, it is an AI Integrations & Solutions job with one pilot and one owner.
If the real goal is more enquiries through search or ads, AI is not the product. Write a Digital Marketing brief for the channel and KPI first (see Brief a digital marketing job before you spend). If the real goal is a new site or app, that is Websites & Apps, with screens under UI & UX (see Scope a website or app before you build).
A model is not the owner of the work
The habit is the one in What an AI-native agency is, and how Voyuma works: people own the scope, the job is named, and then it ships as a Final Product. AI can speed up drafting, sorting, and summarizing inside a written perimeter. It does not decide the workflow, own the data, or sign off done.
When you are weighing a freelancer, a tool subscription, or a firm, the deciding question is still who owns the scope and the data. See AI-native agency vs freelancers and tool stacks. Fill the worksheet before you choose the shape of help.
How Voyuma runs that model
We specialize in four services: Digital Marketing, Websites & Apps Development, UI & UX Design, and AI Integrations & Solutions. For an AI pilot, we name the one workflow, agree the data owner and the perimeter with you, build inside that perimeter, and ship it to the people who run the work. A named person owns the scope and the sign-off. Scoped, then shipped.