Case Studies
Three examples of what fixing the data foundations has made possible. In each, the result came from getting the data and the process right first, then building on top. Examples from my work in pharma commercial data, including previous roles.
Launch planning
A planning assistant that went from right 3 times in 10 to 19 in 20
The situation
A team wanted an AI assistant to draft the launch plan for a new product, using the budgets of earlier launches. It worked in principle, but even the most advanced AI models gave wrong figures, so the plans couldn’t be relied on.
What changed
I broke the job into smaller steps with a check at the end of each, and organised the underlying budget data so the assistant worked from clean, consistent figures instead of a pile of spreadsheets.
Accuracy: from right 3 times in 10 to 19 times in 20.
Time to produce a plan.
Less AI usage for the same work.
Forecasting
Forecast errors cut tenfold, updated weekly instead of quarterly
The situation
A forecast ran on spreadsheets with complicated formulas. Different versions of the file were in circulation, so the wrong numbers and assumptions kept creeping in, and the forecasts suffered for it.
What changed
I moved the forecast onto one organised source of data, so everyone worked from the same numbers, and brought in external market data so that competitor events could be reflected in the forecast.
Lower forecast error rate.
Updates, instead of quarterly.
Quicker, smarter decisions after competitor events.
Competitive intelligence
Five times less manual work to map the competitor pipeline
The situation
Competitor pipeline analysis relied on clinical trial records from ClinicalTrials.gov. Because trial descriptions are written as free text, the same disease or drug could appear under several different names, and sorting trials by disease area and drug meant a lot of manual effort.
What changed
I built a tool that recognised the different names for the same thing, picked apart how each trial was designed, and flagged anything unclear for a person to review. People stayed in control of the judgment calls, and the routine sorting was done for them.
Less manual input.
Took it up.
Understanding of the target market.
Recognise any of this?
Each of these started with a clear look at the data and the process behind it. The Data Readiness Review does that in four weeks, before any build is agreed.