News / All News / Article

The Challenges Facing Pharmaceutical Data Teams—and Weiming Penguin's Solutions

Article images may contain Chinese text.




Start with this scene.

On Tuesday morning, the sales director asks, "Which prescription drugs in East China saw sell-through fall by more than 10% last month?"

The meeting room falls silent for three seconds. No one can answer on the spot.

After the meeting, the question becomes an email to IT. Working across three systems and five tables, IT writes SQL, combines results, and reconciles metric definitions. Only two days later does the answer reach the director.


This is not any one person's problem. It is a problem with how the data is organized. Within one company, the same problem torments four very different groups of people.


Managers: Ask a Question, Wait Three Days for an Answer



You need to keep up with business developments at all times, but every question spanning multiple systems has to go through IT. Ask a question → wait for a slot → wait for SQL → wait for metric definitions to be reconciled → wait for a report → get your answer.


By the time you get it, the question may no longer matter.


What you lack is not more data, but relationships between the data.


The solution: Replace a "table-centric" approach with an "entity-centric" one. The system first maps the relationships among retail outlets, drugs, people, and regions. Whatever you ask, it can follow that network to answer, without writing SQL across tables. Managers should be using data to make decisions, not waiting for data.


Analysts: 80% of Your Time Joining Tables, No Time to Analyze



Your job is analysis, but your daily routine involves opening five Excel files with different metric definitions, fields, and levels of granularity. You spend half a day standardizing them and half a day joining tables—only to discover that the day is over.


You spend 80% of your time preparing data and only 20% thinking. That ratio should be reversed.


The problem is still the "table-centric" approach. Every table is an island, and the analyst has become a human bridge.


The solution: Let a knowledge graph automatically connect the business relationships between tables as data is ingested. Only when the data "grows legs of its own" can analysts free up their hands.


IT: A New Wide Table for Every New Request



Marketing wants retail outlet tiers? Build a table. Sales wants regional target attainment? Build another. Finance wants spending efficiency? One more. Over three years, that adds up to more than a hundred tables. Maintaining them costs more than building them—but the cycle never stops.


Wide tables are painkillers, not a treatment plan.


The solution: Instead of building a wide table for every question, use a knowledge graph to build a network of entity relationships. One graph, every question. IT goes from "table builder" back to "architect."


New Hires: Unfamiliar Table Structures, Two Weeks to Get Started



The system contains dozens or hundreds of tables, all with different fields, metric definitions, and dimensions. You work through them one by one and barely get up to speed after two weeks. Then your manager suddenly asks a question combining things in a new way—and you are back to square one.


How quickly new hires get up to speed is a test of how well your data is organized.


The solution: Open the system and ask a question in natural language: "How did prescription drug sell-through perform in East China last month?" The semantic modeling layer translates automatically, so the system understands everyday language. New hires can create value in their first week instead of memorizing field names.


Final Thoughts



Four roles, four kinds of pain, one question: Has your data "grown together"?


Managers wait because the data has no relationships. Analysts join tables because the data is siloed. IT builds tables because there is no underlying relationship network. New hires memorize fields because the system cannot understand everyday language.


We have built this solution into our product: an underlying knowledge graph connects entity relationships, a middle semantic modeling layer translates business language, and natural-language Q&A at the top drives a closed-loop decision-making process.


The challenges facing these four groups are an architecture problem. And people should not have to shoulder the burden of an architecture problem.