Ian Thomas helps companies turn data into action and results, from day-to-day marketing decisions all the way up to senior leadership. He has enabled teams across Microsoft and Publicis Groupe to unlock product and marketing innovation through data, working with some of the largest, most complex datasets in the world.
His experience encompasses all parts of the data value chain: Data collection, data management and governance (including GDPR compliance), analytics and reporting, data visualization, and automated decisioning and Machine Learning/AI.
I.T.: One lesson I have learned from building and running successful data & analytics teams at Microsoft and Publicis is the importance of nurturing and enabling diverse talents. The skillsets required to deliver effective data projects are so broad and diverse that a team approach is essential – no single individual can possess all of the skills needed across the entire scope of a data project’s lifecycle. Instead, it works better to build a team with a blended set of data skills.
This recognition of diverse talents has important implications for talent and career management within the data field. At Publicis we’ve been thinking about data skills as a portfolio of related capabilities across data strategy, analytics, research data science, data engineering, and data operations; each individual practitioner will have their own unique blend of these capabilities. By capturing this information across the organization, we’ve made it easier for managers and project leaders to resource the right talent onto projects. The individual practitioners have benefited by understanding that they can grow their career in a non-linear fashion, and that a breadth of skills is as valuable as depth in a particular area.
Outside of data teams themselves, the key to embedding a data culture more broadly in an organization is to recognize that data only exists to help solve business problems – either to decide what should be done, to determine whether something was successful, or to help in the execution of a project itself.
I.T.: The best way to empower teams with data is to put the data in their hands and build a culture of data-driven decision-making that means that they are using data every day. This trend to “data democratization” is evident in the success of platforms such as Tableau, Power BI and Google Data Studio, but it is not without its detractors: Some people believe strongly that giving everybody access to data is a recipe for creating a chaotic mess of conflicting viewpoints based on bad numbers, and in the era of GDPR, generating significant privacy risk as data is shared across the organization.
To avoid this, organizations need to put some key foundations in place: First, they need to agree on and define a core set of success metrics for the business and ensure that these metrics are calculated and shared in a structured fashion. Second, they need to share data in a managed way, focusing on discoverability and governance, so that users know what they are working with, and privacy/security risks can be managed. And third, they must work with individual teams to help them understand how they can formulate focused business questions that can be answered with data, rather than just setting them loose in a vast pool of numbers without a good idea of the questions that they are trying to answer.
I.T.: Outcome-driven marketing, as its name suggests, is the practice of identifying (upfront) the specific outcome that a particular marketing activity or program is trying to generate, and working towards that outcome through all phases of the marketing lifecycle.
Thus, the first and most obvious health-check for outcome-driven marketing is that all marketing activity should have a measurable outcome that has been defined at the beginning of the process – before audience/targeting definition, before creative build and selection, and certainly before execution. This outcome should be as far down the funnel as possible – it is much better to measure revenue generated by a campaign rather than clicks, for example – as this is much more proximate to true business impact.
Of course, marketers all over the world know that they should have metrics for their campaigns (they can hardly go to their bosses after the campaign and say that they have no idea how the campaign performed), but typically a marketer will measure a basket of metrics, and when reporting on the campaign will focus on the metrics that moved the most. This is a classic example of “bad science” – by not focusing sufficiently on the outcome they were trying to drive, and then picking the metric post-hoc, the marketer essentially destroys the value of the experiment. Speaking of experiments, the mindset of the modern marketer should be that every campaign is an experiment – as well as trying to drive an outcome, the campaign should be a valuable source of insights and learning that can be rolled into future campaigns and programs. Far too much marketing is executed just because it’s a recurring investment and “we’ve always run this campaign”. This mindset needs to change.
I.T.: The simple answer to this question is that without data, you cannot use advanced techniques like Machine Learning to power marketing, let alone adopt the outcome-driven marketing approach we were just talking about. ML/AI marketing techniques offer up the opportunity to deliver significant ROI improvements based on automation, optimization and targeting; the more data an organization has about its audience and customers, the better optimization algorithms will work (though there are limits, which we’ll come to).
The more interesting question is how organizations can place data at the heart of a virtuous cycle of marketing execution and measurement, so that data doesn’t just feed advanced marketing techniques, but is fed by the results of the work. The response data generated by digital marketing is supremely useful to organizations that are aiming to build a more complete picture of their audience, but is frequently thrown away rather than being fed back into targeting and optimization systems.
I.T.: The key tactic to unlocking innovation through data is to foster a culture of curious experimentation within the organization – not just within marketing, but throughout the company. During my time at Microsoft, I saw the organization transform itself from a company of people who considered themselves the “smartest guy in the room”, where highly-paid leaders felt they knew instinctively which direction to take the company’s products (often with disastrous results, as in the case of Windows 8), to a “growth mindset” culture where testing and experimentation is considered a first-order competency. This transformation is far from complete and has not been either smooth or easy, but it has enable the company to compete in some very challenging markets (such as web search with Bing) and build very important new lines of business (such as Azure and Office 365).
For product and marketing people, the key mindset change is to move from fixating on coming up with the “one big idea” that will suddenly win the day, and instead placing a value on the creation and testing of many ideas, and to separate one’s ego from the ideas themselves. For creative people this is a substantial challenge: it’s certainly the case that big creative ideas still have an important role to play, but even big ideas can be made much more effective with test and learn techniques in their execution and evolution. The original iPhone was a big, bold idea, but only became truly popular in its third or fourth iteration, with Apple making many changes based on user feedback; likewise, Compare The Market’s Meerkats have become an enormous cultural phenomenon through the intelligent execution and optimization of their message across multiple iterations.
I.T.: A big challenge in the industry now is the rise (and mythical status) of the Data Scientist. Seduced by the prospect of magical algorithms, which will unlock huge riches, many organizations are hiring Data Scientists without properly identifying what they will work on, or indeed whether they’re needed.
This is creating two challenges: Data Scientists are finding themselves taking jobs that turn out to contain very little actual data science (building machine learning models, deploying algorithms, and so on) and instead mainly comprise data sourcing, cleaning, pipeline building and reporting/dashboarding; all tasks where an analyst or data engineer would be a better fit. At the same time, the vogue for data science is creating a “second class citizen” issue among analysts – they see Data Scientists being hired alongside them for 30 – 50% more pay, while they themselves could perform many of the Data Scientist’s duties.
An upshot of this is that many analysts aspire to become Data Scientists, which has led to a proliferation of low-quality Data Science programs that seem mostly designed to part analysts from their money. In reality, analysts can learn many of the core data science skills on the job, supported by online learning resources. The path to data science is made even easier by the growth of self-serve/automated data science tools coming to market, creating the new category of “Citizen Data Scientist”, which describes an individual who may not be able to write a machine learning algorithm, but is able to deploy such algorithms in service of familiar objectives (such as generating a lookalike model for campaign audience targeting).
I.T.: In the last ten years, marketers have learned that if they don’t have a report that describes the performance of their marketing, they are likely to get fired. But most marketing performance reports are, effectively, useless, and serve only to give everyone involved a way of justifying the money that was spent on the campaign, rather than delivering any actionable learnings that can be used in future. We covered some of the core reasons for this earlier in our discussion of outcome-driven marketing – most marketing reports today contain a blizzard of metrics which allow marketers to avoid the hard conversation about whether the marketing delivered the objective that it set out to achieve (a conversation made even harder if the objective wasn’t even clear in the first place).
So my core piece of advice to marketers is to decide up-front what outcome their marketing is trying to drive, and to stick to it. Additionally, an individual campaign should only have one key (measurable) outcome – too often, marketers will hedge their bets by saying something like, “this campaign is designed to drive new subscription sign-ups and renewals”, even though these are two different things and could even pull in opposite directions. The lack of focus will then muddy all aspects of the campaign, such as creative choice and targeting.
The reason I’m repeating this point is that I’ve seen time and time again that once someone understands what it is they’re trying to measure, and the business decision they want to make, they are able to get a huge amount more value out of data, and a lot of uncertainty falls away. So many unsuccessful data projects come about because the stakeholders don’t know what they want going into the project.