Most mid-sized employers do not have an AI shortage. They have a time problem. Skilled people spend hours each week on invoices, status reports, record updates and hunting for a document they know exists. That is the work AI handles best right now, and it is a better starting point than any grand strategy.
The harder question is who makes it work once the software is switched on. That is a hiring question, and it is where many projects stall.
Start with friction, not with a tool
The most common mistake is buying a product first and looking for a use second. Reverse it. Walk through a normal week with the people doing the work and write down where they lose time.
Five questions usually surface the answer:
- Which reports get rebuilt by hand every week or month?
- Where is the same data keyed into two systems?
- Which requests arrive by email and have to be sorted and forwarded?
- What information do people regularly struggle to find?
- Which approvals sit in a queue because nobody is watching it?
Pick one. Define success in a number you can check, such as days to process an invoice, hours spent on month-end reporting, or time to first response on a customer request.
Where the time savings are real
These are the uses we see mid-sized employers get value from without rebuilding their business.
- Document handling. Pulling fields from invoices, purchase orders, delivery slips and onboarding forms, then routing them to the right person.
- Internal search. Letting staff ask a question and get the policy, procedure or past answer instead of digging through shared drives.
- Reporting. Drafting recurring summaries from data that already exists, so a controller or operations lead reviews instead of builds.
- Forecasting support. Spotting trends in demand, scheduling and sales data faster than a spreadsheet review allows.
- Early warnings. Flagging inventory gaps, equipment drift or growing backlogs before they hit production or service levels.
None of this removes the need for judgment. It moves the judgment to the front of the process, where a person decides what to do with the output.
Why many projects underdeliver
The pattern is consistent. A tool gets bought, a pilot runs, and a few enthusiasts use it. Six months later the process looks the same because nobody was accountable for changing it.
Three gaps show up again and again.
- No owner. The project belongs to IT, the business, or everyone, which means no one.
- Messy data. The tool works fine on clean inputs. Your records may not be clean, and somebody has to fix that.
- No change in the job. If the role, the checklist and the targets stay the same, people go back to the old way.
In plants, warehouses and shared service centres, a fourth gap appears: the people closest to the process were never asked. They know where the real delays are, and they will quietly work around anything that ignores them.
The roles that make it stick
You rarely need a large AI team. You need a few well-chosen people who can connect the technology to daily work.
| Role | What they do | Where they sit |
|---|---|---|
| Operations analyst or process improvement lead | Maps the current process, sets the measure, tracks results | Operations |
| Data analyst or data engineer | Cleans and connects the data the tool depends on | AI and data |
| Finance systems or reporting analyst | Automates close, AP and reporting steps with controls intact | Finance |
| Project or change lead, often contract | Runs rollout, training and adoption for a defined period | Contract |
The contract option matters. A 6 to 12 month engagement for a process improvement or data lead lets you prove one use case before committing to permanent headcount. If it works, you will know exactly what the permanent role should look like.
Treat adoption as the real work
Training is not a one-hour demo. Supervisors need to know what the tool does and does not do. Staff need clear rules on what data can go into it, and who checks the output before it reaches a customer, an auditor or a production line.
If you are in a regulated area, such as food, finance or healthcare supply, write those rules down early. It is far easier to set them at the start than to retrofit them after an incident.
Measure it like any other operating change
Track the baseline before you start. After 60 to 90 days, compare it with the new numbers. If the gain is small, find out why before you add more tools. Often the answer is a data problem or a team that was never shown how the new step fits its day.
Public productivity data is a useful reminder of the stakes. Statistics Canada and the US Bureau of Labor Statistics both publish labour productivity measures, and the gap between employers who improve output per hour and those who do not shows up in margins over time.
What to do next
Choose one process this month, name one owner, and set one measure. Then decide whether the skills to run it already sit on your team or need to be added, on contract or permanently. If you need an operations, data, finance or contract resource to get a first use case moving, a Traverse recruiter can help you scope the role and find the person.