Most mid-sized employers now have an AI pilot running somewhere. Fewer have the people to run it once it grows. The pilot works on a small data set and a modest cloud bill. Then usage doubles, the invoice triples, and nobody on staff can say why.
That is a hiring problem before it is a technology problem. Here is how to think about the team behind AI workloads.
Why AI changes the cloud job, not just the cloud bill
A conventional business application is fairly predictable. It needs a known amount of compute, a known amount of storage, and steady maintenance. AI workloads are lumpier. Training runs and large data jobs can spike usage for days, then drop off. Specialized hardware may be scarce or priced differently from standard capacity.
The person managing that environment needs skills your existing infrastructure staff may not have. They need to understand how workloads are scheduled, how data moves between systems, and what each experiment costs. A strong generalist sysadmin can learn this, but not on a deadline you set last quarter.
The four roles that matter most
You do not need a large team. You need the right few people in place early.
- Cloud engineer with AI workload experience. Builds and tunes the environment. Look for hands-on work with at least one major platform and real experience running data-heavy jobs, not only web applications.
- Data engineer. Gets clean, governed data to the models. In our experience this is the role most often missing when a pilot fails to scale.
- Cost and usage analyst. Sometimes this sits in finance, sometimes in IT. Someone must own the question of what each AI initiative costs per month and per use.
- Security and compliance lead. Models touch sensitive data. In Canada that means privacy law and, for some employers, data residency expectations. In the US it varies by state and industry.
Smaller companies often cover two of these with one person. That is fine if you are honest about which duties are being doubled up.
Questions to settle before you write the job posting
Candidates in this field can tell quickly whether an employer has thought it through. A vague posting attracts vague applicants. Work out these points first.
- Where will the workloads run? One public cloud, several, or a mix with on-site systems. The answer shapes which certifications and experience matter.
- Who owns the budget? If the new hire cannot see the bill, they cannot control it.
- What is the data situation? If your data is scattered across systems with no governance, say so. A data engineer will want to know, and the honest ones will respect you for it.
- Is this build or run? Designing a new environment and keeping an existing one stable are different jobs. People tend to be good at one and bored by the other.
- What does success look like in 12 months? A number, a system live, a cost target. Something a person can aim at.
Contract or permanent
Both have a place here.
Contract suits a defined build: migrating data, setting up the environment, standing up a first production model. You get specialist skill for the length of the project without carrying the salary after it ends.
Permanent suits ownership. Someone has to live with the system after launch, answer for its cost and fix it at 6 a.m. If you use contractors to build, plan the handover from day one. Ask for documentation as a deliverable, and pair the contractor with the employee who will inherit the work.
The common mistake is the reverse: hiring a permanent specialist for a six-month project, then watching them leave when the interesting part is done.
Pay and competition
These skills are in demand across both countries, and you are not only competing with employers your size. Large technology firms and data centre operators hire from the same pool. Check current wage data for your region before you set a range. In Canada, Job Bank publishes wage figures by occupation and region. In the US, the Bureau of Labor Statistics does the same through its Occupational Employment and Wage Statistics program.
If you cannot match the top of the market, compete on what you can offer: clearer scope, faster decisions, real ownership and flexibility on location. Slow interview processes lose more of these candidates than pay does.
Do not hire in a silo
AI, infrastructure, security and operations need to plan together. If the AI team hires without the infrastructure lead in the interview, you will get a person the environment cannot support. Put the people who will work with the new hire in the room, and have finance weigh in on the cost model.
What to do next
List your current AI initiatives and, beside each, write the name of the person who owns its data, its cost and its security. Any blank is your first hire or your first contract engagement. Then decide whether that gap is a build or a run need, and write the role around the answer.
If you want a second opinion on the role or the market, a Traverse recruiter can help you scope it and find people for it.