Most mid-sized employers do not have a data team, a security team, an automation team and a governance team. They have a few people who wear several hats, and an AI project lands on all of them at once.
That is where hiring plans go wrong. The job posting asks for one skill set, the project needs three or four, and the role stays open for months.
What one AI project actually touches
Take a company that wants to use a language model on customer records or plant data. Before it goes live, someone has to do each of these:
- Move and clean the data so the model gets reliable inputs.
- Control who can see that data and what the model can send back out.
- Build the workflow around the model so its output reaches a person or a system.
- Check that the results can be explained, logged and audited.
In a large company, four teams split that. In a company of 150 to 1,500 people, it is often two or three individuals and an outside vendor. Hiring only for the first item leaves the other three unowned.
Where the real gaps sit
The hard-to-find candidates are rarely deep specialists in one field. They are people who know enough of the neighbouring field to catch a problem early. The overlaps that matter most:
- Data and security: access controls on shared data stores, privacy-aware data movement, knowing what should never be fed to a model.
- Data and oversight: tracking where data came from, spotting bias, keeping inputs in a state an auditor can review.
- Security and automation: building workflows that are safe from the start, not patched after launch.
- Automation and oversight: deciding where a person must approve an AI action before it runs.
When you write a requisition, name the overlap you need. A data engineer who has worked under privacy rules is a different hire from one who has not, and the second will cost you in rework.
Two profiles worth hiring for
It helps to think in terms of two roles, even if one person fills both at a smaller company.
The connector. This person sees how a decision in one area creates risk in another. They sit in design reviews, ask the awkward question about where the data goes, and translate between technical staff and the business. Look for people who have changed domains before and can show what they learned each time.
The checker. AI tools produce output that looks right and is sometimes wrong. Developers who use these tools every day will tell you they still review everything. Someone needs to own that review: sampling results, testing edge cases, and signing off. This takes depth, not just curiosity. In finance, operations and manufacturing settings, it often takes domain knowledge too. A controller or a quality lead who understands the process can catch errors a generalist developer will miss.
Build it in as well as buy it
You will not hire your way out of this alone. Skilled people are scarce in both Canada and the US, and the overlap candidates are scarcer still. The practical answer is a mix:
- Hire for the gap you cannot train quickly. That is usually security depth or data engineering depth.
- Train for the neighbouring skill. Give your security person a working grasp of how data pipelines run. Give your analyst a grounding in access control and audit requirements.
- Put people in the same room. Cross-discipline design reviews teach more than a course. Make them a standing part of any AI project.
- Use contract help for peaks. Data migration, model testing and audit prep are bounded pieces of work. A contract specialist can cover them while your permanent team learns the system.
How to write the role differently
A few changes make a requisition easier to fill and more likely to attract the right person:
- Describe the project, not just the tools. Candidates with cross-over experience recognize themselves in a real problem.
- Separate must-have depth from trainable breadth. Do not list five required specialties in one posting.
- Ask interview questions that test judgment: how would they check an AI-generated result, or what would they want to know before sharing a data set with a model.
- Weigh adaptability. Someone who has learned two new toolsets in three years will likely learn the third.
What to do next
Before you open any AI-related requisition, write down every task the project creates across data, security, automation and oversight. Mark who owns each one today. The unowned tasks tell you whether you need a permanent hire, a contractor or a training plan.
If you want a second opinion on how to scope that role, a Traverse recruiter can help.