A pilot works. Leadership likes the demo. Six months later it still serves eight users and nobody can say who owns it.
That pattern is common, and it is usually a people and decision problem, not a technology problem. Before you buy another tool or sign another consulting contract, work out which of the five gaps below you actually have. Each one points to a different hire or a different fix.
1. No one owns the outcome
If the pilot sits between IT, operations and a business unit, it will stall. Everyone contributes and no one is accountable for results, budget or rollout.
The fix is one named owner with authority over scope and a line to an executive sponsor. In mid-sized companies that is often a director of operations, a head of data or a technical product lead. It is rarely a committee.
Signs you have this gap:
- Three departments run separate experiments with separate budgets.
- Nobody can say which pilot deserves more money.
- Every decision waits for a meeting that includes everyone.
2. The data is not ready
AI work exposes problems that were already there. Customer records live in two systems. Production numbers are keyed by hand. Finance and operations argue over which report is right.
If your team already spends days reconciling reports, a model will not fix that. It will produce confident answers from unreliable inputs.
The hire here is usually not a data scientist. It is a data engineer or an analyst who knows your systems well enough to clean, connect and document the sources. Plenty of employers hire the glamorous role first and then watch that person spend months doing plumbing.
3. The skills are missing and you do not know which ones
"We need AI people" is not a job description. The title matters because the work differs a great deal:
- Data engineer: builds and maintains the pipelines.
- Machine learning engineer: puts models into production and keeps them running.
- Data scientist or analyst: tests ideas and measures what works.
- AI or solutions architect: decides how tools fit your existing systems.
- Governance or risk lead: sets rules for data use, privacy and review.
Most mid-sized employers need one or two of these, not all five, and not all permanently. This is where contract talent earns its place. A 6 to 12 month contractor can build the first working version, document it and train your staff, while you decide what belongs on the permanent payroll.
The mistake to avoid is hiring a senior specialist into a company with no data foundation and no sponsor. They leave within a year, and you are back where you started.
4. Risk questions end every meeting
Security, privacy and accuracy come up in nearly every serious AI discussion. If those questions never get answered, the project never moves.
You do not need a 40-page policy to start. You need a short document that says which data can be used, which tools are approved, who reviews outputs before they reach a customer, and who signs off on exceptions. In Canada, that includes how you handle personal information under federal and provincial privacy rules. In the US, state privacy laws and any industry requirements apply.
Often this is a part-time responsibility for someone in finance, legal, IT or operations who has the standing to say yes or no. Name that person.
5. Nobody agrees what success looks like
Teams drift when the measure of success is vague. Leadership wants growth, operations wants efficiency, IT wants stability.
Pick one or two numbers before the project starts and write them down. Good ones are specific and tied to work you already track:
- Hours of manual reporting removed per month
- Order-entry error rate
- Time to quote
- Days to close the books
- Scrap or rework rate on a production line
If you cannot measure it today, you will not be able to prove it later.
A quick way to diagnose your situation
| Symptom | Likely gap | First move |
|---|---|---|
| Many pilots, no winners | Ownership | Name one owner and sponsor |
| Results nobody trusts | Data | Hire or contract a data engineer |
| Works in a demo, fails in practice | Skills | Bring in an ML or integration specialist |
| Constant security objections | Risk | Write a one-page usage policy |
Hiring versus bringing in outside help
Outside consultants are useful for strategy and for a first assessment. They are less useful when the real gap is a missing person inside your team who will still be there in two years.
A simple rule: if the work is a one-time decision, such as choosing priorities or a vendor, a consultant fits. If the work recurs, such as maintaining data pipelines or reviewing model output, you need an employee. If it is a defined build with an end date, use a contractor.
What to do next
Take your current AI efforts and write down, for each one, the owner, the data source, the person doing the technical work and the success measure. Wherever a line is blank, you have found your stall point, and usually your next hire or contract role.
If you want a second opinion on which roles to fill first, and whether they should be permanent or contract, a Traverse recruiter can help with that.