Launch day is a fixed point. Everything else is measured back from it.
The milestone that matters most is the day field reporting goes live, which DataKitchen plans for six months before launch. Reps need those months to learn whether the numbers hold. If you are reading this at T-12 months, you are already halfway through a build that starts around month 18.
The good news: a first launch does not need a big team. In one published launch case study, one data engineer supported the work for the first 1¾ years, and staffing went to two nine months before launch. What it needs is an order of operations.
Work back from field reporting
| Milestone | Data in place | Decision made | Acceptance check |
|---|---|---|---|
| T-12 months | Warehouse in your cloud account; a first one-time data cut loaded; a list of every feed with its cadence | MDM for the prescriber list; CRM; call planning; incentive compensation system | Each feed profiled before anyone writes a report on it |
| Build week 4 | A first star schema on your highest-priority feed | Which feed the brand team asks about most | Tests on the star from its first build |
| Build week 6 | An AI context layer: schema, grain, business definitions | Definitions for equalized prescriptions and payer hierarchy | Validated example queries return the expected answer |
| T-9 months | Syndicated Rx, claims, alignment and zip-to-territory, EDI 852, 844, and 867 | Staffing: in the case study, one engineer became two here | Extract-to-load reconciliation on every feed |
| T-6 months | The full field report set in production | Alignment version the field and the comp plan both use | Prescriber keys resolve to one master record; territory totals tie to the national total |
| Launch day | Weekly launch tracking; specialty pharmacy dispense and patient status | Which specialty pharmacy files drive the access reports | Volume bounds learned from each feed’s own history |
| After launch | New datasets as questions arrive | What the next business question needs | A diff of overlapping periods on every refresh |
Milestones from the launch timeline. The T-9 staffing step and the feed names come from the launch case study. Which feed lands by which milestone is our recommendation, not a record of that launch.
The full list of 24 reports a launch analytics team owns, with cadence and audience for each, is on the launch page. Read it as your requirements document. Every report on it names the feeds it joins.
Sources come in the order the questions do
The case study team built its sources up in steps. It started with over a decade of claims data, tens of billions of rows, to understand the market. Then it tried IQVIA and Symphony Health for NRx and TRx, explored claims options from Komodo Health and Fingertip and formulary and market-access data from MMIT, and added sales force alignment, zip-to-territory mappings, and EDI 852, 844, and 867 data. The claims data had tests from day one, and a data dictionary published to a wiki gave analysts a common language for every set.
That sequence matches how analysts work. Celgene’s team built a scrappy star first, so analysts had something real to react to while questions to the data supplier were still open, then hardened it over several sprints.
Each feed fails in its own way. The data sources reference lists 37 of them with their characteristic failure. Check it against your feed list at T-12, while there is still time to change a contract or a file spec.
Decide the systems before you build on them
The slow decisions at T-12 are the systems, because each one shapes data the reports depend on. The case study launch evaluated IQVIA OneKey, Reltio, and Veeva Network for its physician master. It weighed Beghou, ZS, and KMK for call planning and sales force design. Its incentive compensation system computed rewards from data in Snowflake that had already been transformed and tested.
Split ownership the same way. Your commercial team owns incentive compensation, territory alignment, call planning, and brand strategy. The data engineering lane owns the feeds, the stars, and the tests those systems read. DataKitchen holds only the second lane.
Give every milestone an acceptance check
A report is ready when its inputs pass, not when it renders. Two checks belong on every feed: extract to load, and period over period. The post on the commercial data most likely to be wrong walks through both. The second one is the check most teams skip.
The T-6 checks are relational, because field reporting is where alignment and prescriber errors surface. A duplicated prescriber splits credit in the comp run, and a realignment moves a rep’s numbers with every value still valid. See how to test territory alignment and incentive comp data and how to find duplicate HCP records.
The stakes are the first year. The first 6 to 12 months set a drug’s lifetime revenue trajectory, and you get one chance at the field’s trust. Six months of live, tested reports before launch day is how you avoid spending that chance on a data problem.