Many mid-sized employers start their AI work by looking for a data scientist. Six months later the project is stuck, and the reason is rarely the model. It is usually that the data is scattered, inconsistent or locked inside systems nobody fully owns.
That is a hiring problem as much as a technology problem. The people who make data usable are different from the people who build models, and most companies are short on the first group.
What data infrastructure means in plain terms
Data infrastructure is the set of systems and routines that collect, store, move, secure and serve your data so people can rely on it. Databases, pipelines, storage, access controls and monitoring all belong to it.
It is different from data architecture. Architecture is the design, meaning how data is organized. Infrastructure is the working machinery that makes the design run every day. A good design with weak machinery still produces late reports and numbers nobody trusts.
For most employers this spans a mix of cloud platforms, older on-premise systems and, in plants and warehouses, equipment that streams data from the floor.
Where AI projects actually get stuck
In our experience, and consistent with what analysts have reported for several years, the common blockers are:
- Data that is incomplete, duplicated or defined differently by each department
- Pipelines built by hand for one report that cannot be reused for the next use case
- No clear owner for data quality, so errors are found by whoever notices first
- Security and access rules added late, which forces rework
- A gap between the technical team and the people who use the output
None of these is fixed by buying another tool. They are fixed by people with the right skills in the right seats.
The roles that matter before the model builders
If your AI plans are early, these are usually the first hires or contracts to consider:
- Data engineer. Builds and maintains the pipelines that move data reliably. This is often the most important and hardest-to-find role.
- Analytics engineer or BI developer. Turns raw data into clean, defined tables the business can use without arguing about the numbers.
- Data platform or cloud engineer. Runs the storage and compute environment, including cost control.
- Data governance or security lead. Sets who can see what, and keeps you inside privacy and regulatory rules.
- Data product owner or operations analyst. Connects the business question to the data, so the work answers something real.
A machine learning engineer or data scientist added before these roles are in place often spends most of their time cleaning data. That is an expensive use of a scarce hire, and it is a common reason they leave.
Why this matters beyond the office
If you run a plant or a distribution operation, the same point applies on the floor. Production monitoring, quality tracking and maintenance prediction all depend on data coming off equipment reliably. Someone has to connect operational systems to the analytics side and keep that connection running. That person often sits between IT and operations, and many employers do not have one.
Finance teams feel it too. A controller who waits 10 days for month-end figures because three systems do not agree is paying a data infrastructure tax every month.
Cross-border considerations for Canada and the US
If you operate in both countries, where data is stored and how it moves is a real question. Canadian privacy law, provincial rules such as Quebec's Law 25, and a patchwork of US state laws all affect system design. Hire or retain someone who understands these before you choose a platform, not after. Rebuilding a system for compliance costs far more than planning for it.
How to staff it without overbuilding
You do not need a large team. A practical approach for a mid-sized company:
- Audit one use case. Pick a single business question, such as forecasting demand or reducing scrap, and trace the data it needs from source to report.
- Find the break points. Note where the data is missing, manual or disputed.
- Hire for the break points. If the gap is pipelines, hire a data engineer. If it is definitions, hire an analytics lead.
- Use contract talent for the build. A defined migration or pipeline project suits a contract specialist, while the ongoing ownership belongs to a permanent hire.
- Name an owner. Someone should be accountable for data quality, with a title and a mandate.
This keeps spending tied to a result and avoids a large platform purchase with nobody to run it.
What to do next
Before you post an AI role, write down the one decision you want better data to support, then list who currently owns each step between the source system and that decision. The empty spots are your first job descriptions. If you want help sorting out whether those are contract or permanent roles, a Traverse recruiter can help with that.