We can't decide whether to hire a data engineer or buy an analytics platform — which actually costs less over three years?

• 01The two options do not have the same cost shape, so comparing headline figures proves nothing. A hire is 36 months of fully loaded salary — base pay plus employer taxes, benefits, equipment and a one-off recruiter fee — and it recurs whether or not the dashboards land. Buying is an implementation paid once, a subscription, and change requests. Get the fully loaded figure from finance rather than base pay before comparing anything.
• 02Be honest about what year one looks like with one hire. A single data engineer spends the first weeks chasing read-only access to the ERP and the next arguing out what “margin” means with finance and operations, then builds a handful of pipelines, one semantic model and a small set of certified reports. Three to six months to the first dashboard the business trusts is the normal outcome, not a bad one.
• 03Four costs are usually missing from the hiring case: the gap before day one, which is search time plus a notice period with nothing being built; key-person risk, because the semantic model ends up living in one person’s head; no cover for holiday, sickness or resignation; and the tooling, training and on-call burden that comes with owning production pipelines.
• 04Buying is not a single invoice either. Microsoft bills Fabric capacity and per-consumer Power BI licences directly to you under any option, including the hire. Change requests outside the pre-built library are billable. And you still need one internal person — a few hours a week, not a full-time engineer — who can settle a KPI definition when two departments disagree.
• 05On the five criteria that actually decide it: cost shape (recurring headcount versus implementation plus subscription), speed (3–6 months versus 4–6 weeks to the first module), KPI coverage (what one person can write in a year versus 200+ pre-built definitions and 50+ connector templates), continuity (a delivery team versus a single point of failure), and control (identical, because Fabric artifacts deploy into your own Azure tenant with no data egress either way). Most mid-market teams settle on buying the layer and hiring an analyst rather than an engineer.
• 06Hire instead when the analytics are the product, or when the metrics have no industry analogue — actuarial, trading, scientific or bespoke machine-learning work. Hire when your sector has no pre-built module, since Manufacturing, Retail and E-Commerce, Warehousing and Distribution, Healthcare and Agriculture are what exists. Hire when a capable data team has already agreed the semantic model, because the hardest part is finished and a pre-built layer buys you less. And hire when policy requires employees rather than a vendor to touch the data.

For the full explainer on this topic rather than this specific question, see the detailed guide on the blog.