Run intake with the people who live in the work
Bring the hiring manager, process owner, frontline user, technology partner, and risk or quality reviewer into one 45-minute session. Ask each person for one recent example where work stalled or needed correction. Request a real, de-identified input and the accepted final output. You are not asking them to imagine an AI future; you are looking at the current handoff.
Record volume, variation, approved sources, who can sign off, and what happens after the answer is produced. If the work changes weekly or exceptions dominate, the hire needs stronger judgment and process design than prompt-writing speed.
Illustrative example
A service team says it needs a chatbot. Intake shows the real bottleneck is that five policy documents disagree. The person to recruit first may be a service-operations lead who can settle policy ownership.
Where this fails
Only the executive sponsor attends intake, so the job description omits the system where work is actually recorded.
Trade buzzwords for observable behavior
“AI visionary” is hard to assess. “Can turn an ambiguous service request into an approved answer path, with a reviewer and a stop rule” is observable. Write five or six essential actions. Separate what the candidate must do on day one from what can be learned with coaching.
Do not make a degree, a list of model names, or a long history with one vendor a proxy for ability unless the work truly requires it. A good process owner may have a modest technical vocabulary and excellent judgment; a deep engineer may need a business partner to make adoption happen.
Illustrative example
Replace “expert in generative AI orchestration” with “can connect approved knowledge to a workflow, show where an answer came from, and route uncertain cases to a person.”
Where this fails
The requisition combines enterprise architect, trainer, product manager, security lead, and subject expert into one mid-level salary band.
Write a candidate profile with two edges
The first edge is business credibility: the person understands the queue, the customer, and the consequences of a mistake. The second is change behavior: they can run a small trial, invite dissent, and make it easier for colleagues to use the new path than a private spreadsheet. Neither requires them to be the best model builder in the market.
List likely partners and boundaries. Does this person own production changes, or recommend them? Can they access customer data? Who approves an exception? The candidate deserves to know whether the organization will give them authority to do what the title promises.
Illustrative example
For a regional bank, a strong profile might be a lending operations manager who has improved exception handling and can work with compliance and technology. The first AI pilot remains narrow and reviewed.
Where this fails
Promising a transformation mandate while every operational decision still needs approval from a manager who is not part of the hiring plan.
Use the job ad as a realistic preview
Open with the first job, not the company’s AI ambition. Name the workflow, the team, the first 90-day assignment, and how success will be judged. Describe the required work sample and offer an accessible alternative format. The U.S. Equal Employment Opportunity Commission explains that applicants may need reasonable accommodations to participate in hiring; an AI-enabled screen does not remove the employer’s responsibility.
Give candidates enough truth to opt in or out. If the source data are untidy and the culture is skeptical, say so. The right person may be energized by that challenge; the wrong person may have expected a clean lab.
Illustrative example
“You will improve our 1,000-case monthly claims intake. In the first 90 days you will map exceptions, test a reviewed draft workflow, and report accepted work and customer corrections.”
Where this fails
A vague “AI transformation” ad attracts impressive resumes but hides that the actual job is persuading a busy claims team to change a daily routine.
Decide whether to hire, train, or borrow expertise
Before posting, ask whether a respected internal operator could lead the workflow with short-term engineering support. Internal credibility is difficult to import. Conversely, if secure integration or evaluation is the gap, giving the job to an enthusiastic subject expert without support sets them up to fail.
A short project with a specialist can test the role design before a permanent hire. Make ownership explicit: the outside expert may build a trial, but the organization must own source records, acceptance tests, and the decision to expand.
Illustrative example
Assign an internal clinic operations lead for a six-week documentation trial and bring in a technical contractor for the data connection. Reassess the permanent role after seeing where work stalls.
Where this fails
Hiring a permanent “AI lead” to solve a problem that was really missing permission to change the workflow.
Use it at work
Ten questions for the intake meeting
- 01Which job, not department, will change first?
- 02How many eligible cases arrive in a normal week?
- 03Show one input and one accepted output.
- 04Where does the work get stuck or corrected?
- 05Who approves the final answer or action?
- 06What information is off-limits to a tool?
- 07Which spreadsheet or side channel do people use now?
- 08What can the hire decide without another committee?
- 09Which skill can the existing team teach?
- 10What would make us stop the trial?
Use AI to help with the hiring work
Draft a role from approved intake notes.
Use an approved tool and de-identified or public material. Keep applicant information out of an unapproved system.
Human check: A process owner checks the work, human resources checks level and accessibility, and the hiring manager removes any invented requirements before posting.
Evidence and limits
Check the sources behind the method
Sources were reviewed in October 2026. The examples and operating frameworks here are editorial teaching tools, not documented company results. Occupational wage figures are national medians for named occupations and do not establish an AI-specific pay rate.