Pfizer's Prescription for Data
Michael Linhares, Ph.D. and Research Fellow, Pfizer
Ever since the invention of aspirin in 1897, the pharmaceutical industry has been among the most complex and competitive arenas in the corporate world. Today, billion-dollar outcomes rest on the success of unique drug breakthroughs, their shelf life and the pipeline of new products that will replace them as old patents expire.
But the constant undertow of any major project is time to market, the core competency that becomes critical where projects are large, extremely expensive and intertwine scientific and business interests over extended time frames.
Nowhere is this process more intensive than at Pfizer, the world's largest drug manufacturer, where $7.5 billion is poured annually into research and development. At Pfizer, as many as 100 commercial projects are taken up annually, of which about five are likely to eventually reach market. Moreover, the timeline for a single successful drug project might be anywhere from eight to 10 years, and a blockbuster might arrive on pharmacy shelves with a 10-figure accumulated cost.
Regulatory and safety approvals take their course, but operational efficiencies are always under a spotlight. "If you bring a $12 billion product like Pfizer's Lipitor to market a month early, you have potentially captured $1 billion in lost revenue," says Dr. Alan Louie, research director at IDC subsidiary Health Industry Insights. "But no one actually does that because most products fail in development. For the very few that succeed, you've pretty much done everything you can, but you probably haven't done it in the most efficient manner."
At Pfizer, efficiency was hobbled by the fact that updated research data was isolated in spreadmarts, tables of Excel data on individual computers that were inaccessible for sharing. So, the company's Business Information Systems team devised a plan to federate Pfizer's data with enterprise information integration tools.
Time to Information
Time to market can be expressed as time to information, where consistent, accurate and timely data is essential to avoid costly missteps and delays, which, by estimates from Pfizer, can add up to $10 million per day.
For Michael Linhares, these challenges are all in a day's work. Linhares arrived at Pfizer in the mid-1990s as a drug metabolism scientist (see Sidebar: The New IT Specialist, below). Today, his title is Ph.D. and research fellow, but his job is to head up Pfizer's BIS team of three scientists/information delivery specialists who expedite information and assist business units with portfolio management services for the many projects underway at the company.
Linhares' customers are Pfizer's Pharmaceutical Sciences (PharmSci) teams of businesspeople - who are also scientists, chemists, clinicians or engineers in their own right - who develop commercial processes for drugs, support manufacturing processes and manage a multitude of budgets and activities as drugs advance through development.
Dry as this might sound, it's an interesting mess that Linhares walked into. Blessed with scientific skills and an intellect that led him into informatics and back, Linhares lately found himself among colleagues confronting legacy data traps common in the pharma industry. The most obvious of these was a proliferation of spreadmarts, one-off, manually entered Excel builds for ad hoc queries. These silos of information were necessary to projects but unshared among working teams across the business.
"Even as a scientist, part of me has always felt an experiment is not done until your data is in the database where you can share it with everybody," say Linhares. In his BIS role, not just sharing data, but ensuring data quality, consistency and transparency became the goals. "We had a variety of scattered experts in different areas, but what we needed was a single integrated data source so, for example, everybody who needs to know where a budget stands has access to the same numbers, which cuts down on confusion in a big way."
As a matter of business, Pfizer maintains a global information factory and data repositories for financial reporting and for FDA-regulated manufacturing, supply chain and packaging processes.
Formal, repeatable reports are also part of PharmSci team resources, but were inevitably backed up by spreadmarts. Linhares' task became to hack away at those desktop silos and replace them with agile processes that would let teams react and change direction as new facts were uncovered and incorporated into the complex analysis required for drug development. Pfizer's data warehouse, enterprise project management, inventory and supply chain, portfolio and project management systems are all sources of information used by PharmSci teams, but were often inaccessible to them in any timely way.
This presented a challenge very unsuited for static reporting. "What really benefits our business is simply listening to people who are in desperate need of information," Linhares says. "You find that many of their questions have never been asked before and are unlikely to be asked again."
One of Linhares' customers who already recognized the pitfalls of spreadmarts is Dr. Karl Bratin, a research fellow at Pfizer who leads a PharmSci team in the areas of commercial process development and manufacturing.
In some cases, Bratin's team develops whole new technologies that might be inhalation devices, drug delivery patches or new formulations for biological processes that require a great deal of learning and development. His role makes him a data provider, a data tracker and a data consumer.
"Sometimes we have to invent science, and things can change very rapidly in that environment," Bratin says. "In my world, spreadsheets are only good at the moment you close them. Information is almost immediately out of date because somebody else has a new piece of data."
Making the problem worse, each spreadsheet user maintained his/her own naming convention, so even centralized information was worth little. In the absence of common dynamic information, for example, purchase orders quickly fall out of date and lead to errors in planning and forecasting.
"You can just imagine all the questions that come up at a huge, distributed global business like Pfizer," says Gartner analyst Ted Friedman. "The research, finance and marketing guys are all siloed with their own spreadsheets, facts change quickly, and you can see quickly how federated approaches to integration really hold value in that scenario."
Indeed, Pfizer had already decided to support its PharmSci teams with a federated data integration model intermediated by Linhares' BIS team of domain experts to quickly track down and answer discrete information requests.
In January 2008, Linhares set about revamping Pfizer's operational data integration and came up with a framework that contained tools including SAP BusinessObjects for business intelligence, TIBCO Spotfire for analytics and light integration through Web services within Microsoft SharePoint portals for data delivery.
But the centerpiece of the framework that would allow rapid deployment of information access was Composite Software's Information Server, which Pfizer had been using since 2006. "We had evaluated EII [enterprise information integration] tools, as they were called then, and did a head-to-head against our existing Oracle Views," says Linhares. "We came to the conclusion that EII offered a big benefit in caching, virtualization and the fact that a business user could use it without special skills."
It was a classic case for EII technology, which first came to market as a means to manage the "last mile" of specialized information inquiries much more quickly and cost-effectively than traditional data warehouse/BI systems.
"Federation is a really useful complementary technique that makes it much easier to iterate and turn around ad hoc reporting requests," says Forrester Research analyst Rob Karel. "What Composite and other vendors like them provide is a way to componentize and deliver targeted insights without having to deal with the whole infrastructure."
With domain knowledge of where data resides, Linhares now handles ad hoc queries from PharmSci teams with simple data flows. "When somebody comes to my office needing something, I usually pull out a blank sheet of paper and draw them a data flow of what they're looking for and then put it into a virtual view they can access on the Web."
In the federated model, Linhares extracts virtual views - smaller tables with data volumes consisting of thousands of rows, compared to the tables with millions of rows in the information factory. "When you think about that and the time it takes to repopulate and rebuild that, it's seconds or minutes versus days," he says.
Along with rapid time to information, a good case for eliminating spreadmarts rested on making data persistent and accessible at any time. A downside of this kind of virtualization is that when source databases go offline for maintenance or to conserve power, virtual views no longer function. With Pfizer's systems scattered in locations that include Sandwich, U.K., St. Louis and Groton, Connecticut, this was almost a probability.
Addressing this, Linhares discovered a "trick" in the Composite server's caching that surprised even the engineers who designed the system. "Composite caches data for a virtual view, but you can also tell the server to save the data to a table in the database and run stored procedures and triggers that let you move data from one database to another," he says. What this meant was that he could use Composite as a "light" extract, transform and load tool to move data into a SQL server so access and drilldown is always available.
Fail Fast and Move On
What makes the solution groundbreaking in Linhares' opinion is that it instantiates continuous process improvement from the get-go. As information quests become more complex, "probability dictates against getting it right the first time," he says. It might be that a business request wasn't specific enough or was misinterpreted by the deployment team. With so much at stake, Pfizer cannot afford many missteps because of delays and costs that pile up so rapidly.
"When we're trying to do something difficult, there is usually a 50/50 chance that we'll get it wrong the first time," says Linhares. "If we're in a low-cost, quick iteration mode, we think it's mostly okay to get it wrong the first time. Even if I do three iterations, it's a matter of hours versus days or months, and by the third iteration we feel we're closer to a 95 percent chance of getting it right." Overall, Linhares says Pfizer has gained a 5 percent improvement in R&D project delivery through fast deployment of planning data to key executives and managers.
In a perfect world, all solutions would succeed even more quickly, and Gartner analyst Friedman warns that a downside of federated systems is that they perpetuate silos and a lack of common definitions that would allow information to be universally shared. Nevertheless, the analyst says, "We're living in economic times where business needs to move faster and can't necessarily get funding for multimillion dollar efforts to build massive centralized data stores," another reason why approaches like federation are gaining traction.
Linhares isn't particularly swayed by this shortfall. "Everything has its purpose, but when you build a big data warehouse, you're in the mind-set of very long development cycles, building sophisticated ETL, getting your data mart and your data model just right. It can be comforting to be able to throw something away and quickly move on."
Linhares' greater concern at the moment is more psychological and goes back to persuading people to let go of their desktop data stores and embrace open sharing. "It's how you try to convince people they make themselves more valuable this way, but for some, it's like a 'fight or flight' reaction."
Forrester analyst Karel says change management becomes a huge issue whenever companies move to new technologies. "Getting away from spreadmarts is difficult," he explains. "It's hard even if you're just trying to move to hub and spoke or an integrated data warehouse because you still have to ask someone to stop doing what they're doing and approach it in a different way."
Where customers like Dr. Karl Bratin buy into the results is proof (as once was the case with wooden filing cabinets) that a better answer has arrived. "Michael's [Linhares'] group is providing us with quicker access to uniform business data I can rely on," says Bratin. "And that's the information we will use to make business decisions."