A mid-sized employer decides to start an AI project and posts for a data scientist. Six months later the hire is spending most of the week reconciling spreadsheets and asking who is allowed to change a field definition. The model is not the problem. The groundwork is.
This happens in insurance, distribution, manufacturing, financial services and professional services alike. Before you open an AI or data requisition, it helps to know what you are actually hiring someone to walk into.
The symptoms show up in ordinary work
Data readiness rarely announces itself. It shows up as small delays:
- Two reports give two different revenue or margin figures.
- A simple question from the owner takes three days because someone has to reconcile the numbers first.
- A system change in one department quietly breaks a report in another.
- Nobody can say who decides which number is right.
If you recognize two or more of these, a new analyst or engineer will inherit them on day one.
Three gaps behind most stalled projects
The decision gap. The data exists, but getting an answer means manual pulling and last-minute fixes. Leaders get answers late and trust them less.
The ownership gap. When reports conflict, nobody is responsible for settling it. Definitions of a customer, an order, a claim or a billable hour differ by system and by department.
The proof gap. The team agrees on an answer but cannot show how it was produced. That is an inconvenience in a board meeting and a real problem in an audit, a customer dispute or a regulatory review.
A five-question test you can run this week
Skip the maturity model. Ask whether your team can do these without heroic effort:
- Trace a number. Follow a key metric from the dashboard back to the source system and its written definition.
- Change safely. Update a system upstream without breaking reports downstream.
- Name the owner. Say who resolves a mismatch and who they escalate to.
- Measure quality in business terms. Track errors that matter to the user of the data, not only technical completeness.
- Grant access quickly and properly. Get the right people into the data fast while meeting security and privacy obligations.
If the honest answer to most is no, the first hire is probably not a machine learning specialist.
What this means for hiring
The roles that fix readiness are often less glamorous than the AI title on the original wish list.
- Data or analytics lead with ownership authority. Someone who can approve definitions and settle disputes, with a mandate from senior leadership. Without the mandate, the title is decoration.
- Data engineer or integration specialist. The person who makes systems feed reporting reliably and tests changes before they cause damage.
- Business analyst or operations analyst. Often the bridge between the people who use the data and the people who store it.
- Finance or operations systems specialist. In many firms, the controller's team carries the reconciliation burden. Relief here pays back quickly.
- Contract data talent. For a defined clean-up of one domain, a contract engineer or analyst can do the work without adding permanent headcount.
When you do hire for AI work, write the role around the problem. "Reduce time to quote by cleaning and connecting pricing and order data" attracts stronger candidates than a list of tools, and it screens out those who want a research role you cannot offer.
Start with one workflow
The temptation is to fix everything at once. It rarely works, and few companies can pause operations to modernize every system.
Pick one workflow that matters, such as quoting, order intake, month-end close or production scheduling. List the 3 to 5 decisions it depends on. Map only the data needed to make and explain those decisions. Name an owner for each definition, set a simple quality check, and document where the numbers come from.
Once that pattern holds, repeat it on the next workflow. The second one goes faster because the habits and the people are already in place.
Treat definitions as a routine, not a project
Teams that keep data reliable do a few boring things on a schedule. They review definitions regularly, show quality problems in terms the business understands, and assess system changes for downstream impact before they go live. None of this requires a large budget. It requires someone whose job it is.
What to do next
Before posting an AI or data role, run the five-question test with your controller, your operations lead and whoever owns your main systems. Write down where the answers break. That list is your real job description, and it tells you whether you need a permanent hire, a contractor or an internal reassignment.
A Traverse recruiter can help you define the role, decide between contract and permanent, and find people who have done this clean-up work before.