You have five options, and the vendors pitching each will tell you the other four are wrong. The useful question is narrower: when the people change, who still has the code and the knowledge of your feeds?
Five models, compared
| Model | Cost | Who owns the code | Domain knowledge | Continuity | Exit |
|---|---|---|---|---|---|
| Your IT department | Internal, but your requests queue behind every other department | You | Generalist; learns pharma feeds on your timeline | Shared with every other priority | Nothing to exit |
| Hire your own team | Six FTEs at $1.5M to $2M a year, plus six to nine months to recruit | You | Strong once hired | Rests on keeping specialists who are hard to keep | Nothing to exit |
| Hourly analytics or offshore vendor | Around $300 per analyst hour for consulting; offshore cut the rate, not the steps | Often the vendor | Varies by vendor | Tied to the contract | You get back what the contract says |
| Dedicated commercial data team | One to three engineers at a flat monthly rate | You, from the first sprint | Brought on day one | Tests and code in your repo outlast any engineer | No exit fee; hand over to anyone |
| Open source, run yourself | Software free; people still needed | You | Whatever your team has | Rests on whoever runs it | Nothing to exit |
Sources: hiring and consulting figures from the cash-runway post; IT and domain knowledge from the $100 billion secret; offshore economics from the post on the three ages of commercial data engineering.
Cost is mostly people, so count them honestly
Every option except open source is a staffing decision. Six commercial data FTEs fully loaded in Boston or the Bay Area run $1.5 million to $2 million a year. A multi-year analytics consulting agreement runs higher than that, by the same post’s numbers. DataKitchen typically staffs one to three engineers, and at one rare-disease biotech, consolidating two hourly vendors into that model cut the annual cost of running commercial data by about 58%.
The hourly models carry a specific trap. A developer offshore cost a third of one in New Jersey, so teams bought hours in bulk. But the hours never went down, because offshoring lowered the wage on every manual step and removed none. Same assembly line, same defects, lower labor rate.
Domain knowledge is the cost nobody budgets
Specialty pharmacy files, payer hierarchies, NDC levels, vendor restatements. A generalist IT team can learn them eventually, but the learning happens on your launch timeline.
James Royster led an analytics function at Celgene for nearly seven years. His account of why he chose DataKitchen is about method as much as people: testing at every stage, automating any task done more than twice a year, and a team that did more analytics, faster, with fewer errors than the companies it benchmarked against.
Open source solves the license, not the job
Open source removes lock-in on the software. DataOps TestGen and DataOps Observability are both Apache 2.0. In a DataKitchen engagement the data quality software runs in your cloud, so no license expires when we leave.
It does not remove the work. The open source edition of TestGen is one user, one connection, and one project, and the commercial pharma identifier tests ship only in Enterprise. More to the point, building a pipeline is the small part. Running it never ends: the schema change, the late file, the expired credential.
Ownership decides the exit
The test that separates these models is what you hold on the day you want to change. With DataKitchen, the warehouse sits in your cloud account and the code and tests sit in your repository from the first sprint. The rate scales down on 30 days’ notice. Transfer goes to your team, Accenture, or your offshore center, with no exit fee. Eisai moved one platform to Accenture that way.
The models also mix. On the BMS CAR-T warehouse, 41 contributors from five companies, including Beghou and Accenture, have worked one codebase over seven years. That works only when tests are the contract.
One limit on DataKitchen before you choose: we do not do incentive compensation, territory alignment, call planning, or brand strategy, so those need covering elsewhere. And if you hire your own team later, we hand the platform to it. The in-market cost page shows how a takeover works when you already run something today.