Improving pharmaceutical marketing using big data solutions


Published on

A presentation for SMi Big Data in Pharma conference, London, 12-13th May 2014.

Slides herein contain most content shown on stage.

1 Comment
No Downloads
Total views
On SlideShare
From Embeds
Number of Embeds
Embeds 0
No embeds

No notes for slide
  • Show of hands – who thinks this is funny (at all). Keep your hands up if you also like Star Wars. This was for you!
    Now, I took an intuitive guess that a bunch of data geeks might be interested in Star Wars and Lego and would find this humorous.
  • The best marketing, is about presenting an idea that resonates with an audience in a way that generates interest. If you know what the audience likes, really likes, you can resonate – and in some cases, change behavior.
  • And now we have the possibility of big data… to think in terms of country, city, street, or community… A scale once only possible with the budget of NASA is now possible for most organizations for a reasonable price.
  • Tapping in to big data for marketing really means going beyond what you say or think about yourself – it’s the evidence from what you actually do. Behavioral trends and indicators. Yes, evidence!
  • Perhaps surprisingly, people willingly share personal information about themselves which is a goldmine for healthcare marketers who are willing to embrace the challenge.
  • We know how some companies have developed their business model around behavioral data.
    Consider the man who stormed into a Target near Minneapolis and complained to the manager that the company was sending coupons for baby clothes and maternity wear to his teenage daughter. The manager apologized profusely and later called to apologize again – only to be told that the teenager was indeed pregnant. Her father hadn’t realized. Target, after analyzing her purchases of unscented wipes and magnesium supplements, had.
  • Consumers knowingly permit personal data to be stored and processed by companies, if that means TARGETED marketing and less ‘noise’.
    Online survey of just over 2,000 consumers and 748 senior business decision makers.
  • In recent times (May 2014) there has been an increasing number of warnings from experts in academia and big business about the ‘hype’ of big data.
  • Marketers are not always understanding how ‘big data’ might be relevant to them. Is it science, is it statistics, is it market research, or is it analytics?
    There were 896 respondents to the research request, which took the form of an online survey in Q2, 2013.
    Key facts about the sample include:
    Two-thirds of the respondents (67%) stated that they were based in the UK, with a further 18% coming from the rest of Europe. Asia Pacific, North America and the MENA region were also represented.
    Almost one in five (19%) respondents said their principal role was as a data analyst.
    Almost one third (31%) of UK respondents came from a company with a turnover over £150 million. 40% of non-UK respondents came from companies with a turnover in excess of US $150 million.
    The most well represented business sectors for in-house marketers are retail (16%), financial services (12%), travel (8%) and publishing (7%).
    For supply-side respondents, 58% stated they worked for a digital agency, with a further 17% as self-employed consultants.
  • Mostly retail, finance, travel, publishing, - Freeform responses to the question: “what effect has big data had on the web analysts in your organization?”
  • We know that the pharma industry is constrained by regulation; we know that FMCG and other consumer sectors are able to innovate and pilot technologies and techniques on the cutting edge. The ‘lag’ is not negative. It gives us the opportunity to learn from other sectors, before implementing (hopefully) best practice within the constraints of our industry.
  • The reality is that Pharma is still thinking in discrete channels, as a portfolio of channels, and trying to work out ‘Multi-channel Marketing (MCM)’ – i.e. how to make it work holistically and cohesively.
    But they don’t see that big data analytics and predictive preference-based content is the way to go for the future (and for now, where possible). Or, if they do, it is a bridge too far in terms of legal and medical approvals.
  • Yet the customer is ultimately a professional person, with behavioral traits like any other consumer. Sure, a rational and qualified professional – but at the cure, a human not a number.
  • Now let’s look at some ‘low-hanging’ fruit in two cases which show how the concepts of big data can be used to improve pharmaceutical marketing.
  • The first is a study of healthcare professionals (customers) and how to distil the signal from the noise. The population of people talking about health is as large as the population of people using the Internet.
  • Healthcare professional conversations may typically contribute between 2-5 % of the content in a given therapy area. Of course, the use case is very different.
    You may look at the 100,000 sources in Creation Pinpoint and think that this is not ‘big data’. True.
  • Obtaining these individuals and analyzing their behavioral signals – that big data!
  • It is also their geo-location data – when and where they are contributing intelligence to the healthcare conversation.
  • Importantly, we can also understand the evolving networks of content distribution and influence.
    Relationship marketing or influencer marketing is not new to Pharmaceutical companies. However, the rise of the DOL (digital opinion leader) is a paradigm shift. This requires big data solutions.
  • You will have all by now seen a social network analysis i.e. nodes and edges. Graphing and visualization of data still creates good opportunities for human analysts to spot outliers and areas of interest. Once the scale increases sufficiently, we end up with a hairball – something which is not quite as useful. Even here, just 10% of a data set looks colorful, but hardly insightful
  • We can also build a profiles database of individuals – sure, there are concerns around privacy. However, much of the social data is publicly available and according to the terms and conditions accepted by each user of the social platform, permissible for use.
  • It’s is real-time, ongoing ‘buzz’ monitoring, but the more powerful and regularly commissioned work is about historical analysis to inform future plans. Here we see a demonstration dashboard from Brandwatch, modified by Creation Pinpoint technology and Creation Healthcare analysts. We can detect emerging treatment types that no one could have pre-empted or expected. It was not part of our ontology, but will be from now on. With such tools we can discover the unknown unknowns.
  • The accuracy, relevance, and timeliness of medical information is critical to the success of treatments, and of course pharmaceutical companies in general. Managing risks, concerns, and reputation can determine the value of the stock price for a company. It is fundamental and important!
  • Doctors are having a hard time keeping up with the vast reams of medical information which is now available. Watson may not be a doctor, but is so far proving to be a good diagnostician!
    What are the applications if such machine learning systems are turned onto our own datasets?
  • Still, we don’t have to be that advanced to start improving pharmaceutical marketing. In one project, we tool a look at contact records from a medical information database. So, calls, emails, and other contacts from mainly healthcare professionals seeking answers to medical and technical product questions. The ultimate dataset includes around 1 million records per annum in ~27 languages.
  • Starting with the structure data, and extrapolation of data into geocoded records, it is possible to create a summary dashboard that instantly gives more insights than the typical service oriented KPIs like turn-around-time. Here, our dashboard is interactive.
  • So, if I click on the category ‘pharmacist’ I can quickly identify that there are four products which have repeated queries. If not already, I can have my marketing team produce a campaign directed at resolving these queries proactively, potentially alleviating unnecessary concerns and helping to build the knowledge needed at the point of dispensing.
  • Or in another dashboard, I can investigate unstructured data by focusing in on human identified anomalies. What happens when I click ‘fridge’ – well I see which products are having repeated questions relating to this topic and I can see through context sensitive charts the other tokens and specific verbatim concerns. This may lead to a new SRD (standard response document) or in a more data-driven model, my HCP portal may dynamically update to ensure that this topic is served as a higher priority in search.
  • Mapping verbatim tokens over time, I can determine anomalies relative to ‘normal’. Automated triggers and ‘intelligence events’ can notify my cross functional team to prevent unnecessary resource wastage or reputational damage. In this case, a simple packaging change resulted in numerous queries because it was no longer clear whether the product could be stored out of the fridge.
  • Identifying anomalies can lead to better customer service through medical representative and the field force. Here, the data prompts an investigation of Newport and Lincoln – why is there a disproportionate leaning on medical information? Is this good? Does it expose an unmet need or information gap?
  • A final model for helping our marketing teams to think about the possibilities of big data.
  • Improving pharmaceutical marketing using big data solutions

    1. 1. Improving Pharmaceutical marketing performance using big data solutions Paul Grant Chief Innovation Officer @paulgrant
    2. 2. Photo credit: pasukaru76 / Foter / Creative Commons Attribution 2.0 Generic (CC BY 2.0)
    3. 3. Photo credit: Eva Rinaldi Celebrity and Live Music Photographer / Foter / Creative Commons Attribution-ShareAlike 2.0 Generic (CC BY- SA 2.0)
    4. 4. Photo credit: NASA Goddard Photo and Video / Foter / Creative Commons Attribution 2.0 Generic (CC BY 2.0)
    5. 5. Photo credit: josullivan.59 / Foter / Creative Commons Attribution 2.0 Generic (CC BY 2.0)
    6. 6. Photo credit: AndYaDontStop / Foter / Creative Commons Attribution 2.0 Generic (CC BY 2.0)
    7. 7. A sizeable proportion of consumers are happy for companies to use their personal data, providing they benefit through more targeted marketing Photo credit:$FILE/ey-bigdata_v3.png
    8. 8. “There are a lot of small data problems that occur in big data. They don’t disappear because you’ve got lots of the stuff. They get worse.” David Spiegelhalter Winton Professor of the Public Understanding of Risk at Cambridge University
    9. 9. 'don’t care – big data is a pointless marketing term’ Online Measurement and Strategy Report 2013 by Econsultancy, July 2013 8%Marketers say…
    10. 10. Time available to analyse data in Google Analytics is too little, so adding more data to the 'pile' to analyse will only lead to less insight, not more. Little to none. We know we need to gather and analyse the available date to run our marketing and our business better, but 'big data' is not the driver of this. We have tonnes of data and sometime it's difficult to analyse, but this has always been a problem and always will be as data acquisition will keep growing. Not sure what "big data" means. Online Measurement and Strategy Report 2013 by Econsultancy, July 2013
    11. 11. Photo credit: What about Big Data in Pharma^? marketing
    12. 12. Photo credit:
    13. 13. Photo credit: j.reed / Foter / Creative Commons Attribution-ShareAlike 2.0 Generic (CC BY-SA 2.0)
    14. 14. Opportunities for ^ improvement 1. Observational (Passive) inputs – Non-solicited, non-structured, non-validated – Basis of a hypothesis – indicative insights, trends 1. Direct engagement (Active) inputs – (somewhat) structured, solicited etc. – Tied (hopefully) to business questions – Still has a ‘human’ component In both cases we are exploiting real-time data marketing
    15. 15. 70M+ websites Full Twitter ‘Firehose’ feed All major social media • >100,000 verified healthcare professional (HCP) sources covering websites & social media Typically 2-5% of all public social media conversation for a health topic coming from HCPs
    16. 16. 89,824,885 processed social profiles 377,744 algorithm selected profiles 88,569 human validated profiles 24,519 HCP authored blogs & sites 208million tweets 152thousand tweets per day Source: Creation Pinpoint, data correct at Jan 2014
    17. 17. Photo credit:
    18. 18. The digital world changes the model of influence Traditional KOL model: Emerging DOL model: KOL relationships are different to digital opinion leader (DOL) relationships Hierarchy typically based on seniority, experience, publications etc. Collaborative ‘flattened’ relationships, not ordinarily common in real-world
    19. 19. HCP community network Cardiologist Academic Physician Academic Surgeon Anesthetist Dental surgeon Hospital Director Media Physician Medical Biologist Medical student Neurologist Neurosurgeon Nurse Oncologist Orthopedic registrar Pediatrician Pharmacist Physician Psychiatrist Neurolaw Rheumatologist Sports Therapist Trauma Anesthetist Trauma Physician Various Nodes: 13,781(4.35%) Edges: 35,886(9.05%) Note: This diagram represents ~10% of the HCPs connected to those talking about the study topics (shown as colored circles)
    20. 20. Detailed HCP profile information
    21. 21. Creation Pinpoint sample study, inflammation among conversations of UK healthcare professionals 01 Dec 2012-30 Nov 2013
    22. 22. Photo credit: National Library of Ireland on The Commons / Foter / No known copyright restrictions
    23. 23. Proof-of-concept real-time NLP A: Initial data insights B: Future strategic approach Analysis of an anonymized sample dataset to determine the visual outputs and information insights that are possible. An exploratory exercise to find ways that medical information can potentially service commercial strategy development. Key components include: Based on learning from the sample data set, and the evaluation of various tools and processes for developing these insights, recommendations to be made for how PharmaCo might use this type of data in an on-going implementation. Key components include: • Assessing data opportunities • Pricing and feature comparison • Analysis and experimental approaches • Handling of languages other than English • Types of outputs possible • Metrics and potential success indicators • Presentation of findings • Potential real-time integration
    24. 24. 28Drill-down by area of interest i.e pharmacistDrill-down by area of interest i.e pharmacist Four+ clear ‘problem’ products for pharmacistsFour+ clear ‘problem’ products for pharmacists
    25. 25. What happens if we focus on a word like ‘fridge’What happens if we focus on a word like ‘fridge’
    26. 26. 30 Clear issue already detectable week one, escalation within business to avoid week two peak Clear issue already detectable week one, escalation within business to avoid week two peak Normal Issue
    27. 27. Rank City Population MI requests Ratio 1&2 London/City of London(England) 7556900 818 0.011% 3 Birmingham(England) 984333 216 0.022% 10 Manchester(England) 395515 177 0.045% 4 Glasgow(Scotland) 610268 159 0.026% 6 Leeds(England) 455123 143 0.031% 22 Nottingham(England) 246654 141 0.057% 5 Liverpool(England) 468945 139 0.030% 18 Belfast(Northern Ireland) 274770 135 0.049% 9 Bristol(England) 430713 115 0.027% 31 Newcastle upon Tyne(England) 192382 108 0.056% 8 Edinburgh(Scotland) 435791 105 0.024% 44 Dundee(Scotland) 151592 77 0.051% 7 Sheffield(England) 447047 66 0.015% 62 Newport(Wales) 117326 65 0.055% 12 Leicester(England) 339239 63 0.019% 16 Cardiff(Wales) 302139 62 0.021% 23 Southampton(England) 246201 57 0.023% 38 Walsall(England) 172141 57 0.033% 26 London Borough of Harrow(England) 216200 52 0.024% 90 Lincoln(England) 89228 52 0.058% 43 Oxford(England) 154566 47 0.030% 17 Bradford(England) 299310 45 0.015% 24 Reading(England) 244070 45 0.018% UK population data source:
    28. 28. Know Know Don’t Know Don’t Know What we know we know What we don’t know we know What we don’t know we don’t knowWhat we know we don’t know Customer Information Source:Adapted from
    29. 29. Thoughts (and some tools) 1. Getting started: need education for marketing departments to develop understanding of the power of indicative insights… – what data do we already have, or could we have – how to ‘munge’ it to answer behavioral or segmentation questions – beyond the obvious, in real-time 1. Create content (dynamic?) for specific segments/needs 2. Allow customers to set their own preferences (then learn!) 3. Once you have the basics, start to explore machine learning algorithms and predictive analytics Can Pharma be as ‘clever’ as Amazon or Netflix? Of course! 1. Online HCP insights research: Creation Pinpoint 2. Social Network Analysis: Gephi/Anaconda 3. Integrations and data scraping: 4. Location visualization: CartoDB 5. Natural language processing: Brandwatch, Lexalytics, Semantria, Clarabridge 6. Structured and unstructured data: Omniscope