Many operations-heavy employers are now being asked the same question by their boards: where can AI take cost out of the work we do every day? Telecom, utilities, logistics and manufacturing all have the same shape of problem. Repetitive configuration work, ticket handling, asset ordering and reporting eat hours, and errors are expensive.
The technology is moving quickly. The harder part is staffing it. Most projects stall because the wrong people were hired first, or because nobody owned the process the AI was meant to improve.
Start with the process, not the model
The projects that hold up begin with a plain description of a workflow that costs too much. Good first candidates share a few traits:
- The work is high volume and follows rules most of the time
- Errors are costly, so a second set of checks has value
- The data already exists in tickets, logs, purchase orders or spreadsheets
- A manager can measure the result in time, cost or quality
Call handling times, change windows, stock-outs and report turnaround all qualify. If you cannot name the number you want to move, you are not ready to hire for the project.
The five roles that carry an AI workflow project
You rarely need a large team. You do need these jobs covered, whether by employees, contractors or a mix.
- Process owner. Someone from operations who knows how the work really gets done, including the exceptions. This person is usually already on your payroll.
- Data engineer. Gets the logs, tickets and records into a state a model can use. Without this person, everything else slows down.
- Applied AI or machine learning engineer. Builds the retrieval tools, agents and integrations. Look for people who have shipped something into production, not only prototypes.
- Quality and test lead. Checks outputs against requirements and catches the confident wrong answer. This role is the one most often skipped.
- Governance and security lead. Decides what data the tools may touch, who reviews outputs and how decisions are logged. Often a part-time duty shared with IT or legal.
Contract first, permanent second
Most mid-sized employers do not know yet how many AI people they need in two years. That argues for a staged approach.
- Bring in contract engineers for a defined build of 3 to 6 months, with a pilot scope and a target metric.
- Keep the process owner and quality lead as permanent staff who stay with the tool after launch.
- Convert or hire permanently once the pilot proves out and you know the ongoing workload.
This limits your exposure if the first use case does not pay off. It also gives you time to see how your contractors work before you commit to a salary. The risk is knowledge walking out the door at the end of the contract, so require documentation and a handover period in every agreement.
Capitalizing the spend is a finance conversation
Some software development costs can be capitalized rather than expensed, which changes how a project looks on the income statement. The rules differ between US GAAP and the Canadian standards that apply to your company, and they depend on how the work is structured and documented. Involve your controller before the project starts, not after the invoices arrive. Time tracking by task and clear project stages make the treatment easier to defend with your auditors.
What to test in interviews
AI titles are inflated. A few questions separate people who have done the work from people who have read about it:
- Describe a workflow you put into production. What broke in the first month?
- How did you measure whether it was better than the manual process?
- What did you do when the model gave a wrong answer with full confidence?
- Which data did you decide not to use, and why?
For the process owner and quality roles, ask about the exceptions. Strong operators can list the five cases that never fit the standard procedure. That knowledge is what makes automation safe.
Plan for the people whose work changes
Automating ticket handling or configuration changes shifts jobs rather than deleting them overnight. Technicians become reviewers. Analysts spend less time pulling data and more time acting on it. Tell your teams early what is changing and what training you will provide. Employers who skip this step see quiet resistance and workarounds, and the pilot numbers never reach the rest of the business.
What to do next
Pick one workflow with a number attached, name the process owner, and write a one-page scope before you post a single job. Decide which roles you will fill on contract and which you need permanently. Then hire the data engineer first, since everything else depends on that work.
If you want help finding AI, data or operations people for a project like this, a Traverse recruiter can help.