Choose a specific workflow with a recognised business problem, an accountable owner, and information you can use. Compare candidate use cases on value, feasibility, review effort, consequences of error, and adoption. Establish the current baseline and define the decision a pilot must support before choosing tools or committing to a build.
Your first AI initiative sets expectations for everything that follows. A useful starting point should address a problem people recognise, fit the information you can actually access, and give your team a clear way to judge the result. An impressive demonstration can help secure attention. A well-chosen piece of work can help secure a better business decision.
We recommend starting with a specific workflow and its owner. The question is where AI could improve the work enough to justify the effort of changing it.
Describe the work before choosing the technology
Replace broad ambitions such as “use AI in operations” with a description of what happens today. What starts the process? Who receives the request? Which information do they consult? What do they produce, and who checks it? Include the handoffs and exceptions that make the work difficult.
“Help account managers prepare for customer reviews” is still broad. “Prepare a review brief from approved account notes and service records for the account manager to check” gives a team something concrete to explore. It identifies the inputs, output and person responsible for using it.
Look for a worthwhile, manageable starting point
Build a shortlist with the people who do the work. Ask them where time disappears, where requests wait and where information has to be assembled repeatedly. Then examine each candidate against the same questions:
- Business value: What would improve for the customer, employee or organisation if this worked?
- Information: Are the required sources available, current and suitable for the intended use?
- Judgement: Can a qualified person recognise a good result and identify a bad one?
- Consequences: What could happen if the output were incomplete or wrong?
- Ownership: Who can make decisions, provide examples and change the workflow?
- Adoption: Will the result arrive where people already work, in a form they can use?
These are discussion prompts, rather than a universal scoring formula. A high-value idea with inaccessible information may belong on a later roadmap. A modest opportunity with a committed owner can make a more informative first trial.
Establish what improvement would mean
Observe the current process before setting a target. Include preparation, waiting, checking, correction and follow-up. A faster first draft has limited value if it creates more work for the reviewer.
Choose measures that reflect the actual job. Depending on the workflow, these might include total handling time, completeness, missed information, escalation quality or whether employees can find the answer they need. Record what you know and which assumptions still need testing. Treat estimated benefits as estimates until the workflow has been evaluated.
An illustrative choice
Consider a fictional equipment distributor comparing two ideas: an assistant that answers customers directly, and an internal assistant that prepares product information for a sales specialist. Both might use similar reference material, but they create different responsibilities.
The internal option lets the specialist check sources and correct the draft before anything reaches a customer. It also exposes gaps in the product information. The direct-response option requires additional decisions about permitted answers, unusual requests, escalation and service ownership.
The internal assistant could therefore be the better first investigation, provided preparation is a meaningful burden. That conclusion would come from the workflow and its constraints. It would not establish that internal tools are always preferable or that this particular idea will deliver a return.
Give the first initiative a clear boundary
A short decision brief should state the problem, intended users, permitted inputs, expected output and review process. Include what the trial will leave out. For example, drafting an account brief need not include updating customer records or sending messages.
Name the business owner and the person responsible for evaluation. Identify the decision the trial must support: proceed, adjust the scope, investigate another option or stop. This gives the implementation team a useful brief and gives leadership a basis for reviewing the evidence.
For a broader approach to considering trustworthiness throughout an AI system’s lifecycle, the NIST AI Risk Management Framework is a useful voluntary reference. The selection approach above is our practical recommendation, rather than a certification checklist.
Authentia’s AI strategy advisory helps leadership teams turn a list of ideas into a prioritised set of decisions, with a clear brief for the people who will deliver the work.