Unlocking Advanced Analytics Value With A Designer’s Desirability Lens
Dan Feldman, Design Director, McKinsey, Sydney; Maksud Ibrahimov, Jr Principal Data Scientist, QuantumBlack, Melbourne; Justin Hevey, Expert Designer, McKinsey, Sydney; Cris Cunha, Analytics Expert Associate Partner, QuantumBlack, Perth and James Deighton, Partner, McKinsey, Melbourne
This is the second of our three-part series, Exploring The Intersection Of Design & Advanced Analytics. In this article, we will take a deeper look at how effective collaboration between designers and data scientists can unlock greater value in advanced analytics (AA) projects.
As we referenced in our previous article, failure in advanced analytics projects occur for a host of reasons including technical issues, misalignment with business priorities, and insufficient user adoption. While technical feasibility and business viability failures are less prevalent than they were a few short years ago, the disconnect between the analytic solutions proposed and their up-take by end users represents the latest bottleneck for many advanced analytics projects. The analytics-driven solution, no matter how technically advanced or valuable, simply is not adopted and does not deliver the full potential impact.
Why does this happen? Data science is often suitable to the problem and, as McKinsey has previously noted, the growing prevalence of ‘translator’ roles means that organisations are increasingly pointing analytics at the right business problems. The technology is feasible and the challenge exists in the organisation, so the solution is commercially viable — so why is this solution left on the shelf by those for whom it was created?
This situation is often the result of failure to integrate the “human desirability” lens throughout these projects. Developing a model with adoption in mind from the very beginning will mitigate against the need for a time-consuming and expensive change management process later — but it requires extensive collaboration between data scientists, engineers, designers, and the end users.
Design and analytics on the surface are quite different enterprises: Analytics is highly technical, includes calling on expertise in machine learning and artificial intelligence, and identifies patterns at scale that are anchored in data. Design is highly empathetic and creative, focuses on motivation and mindsets, and looks at deep patterns in experiences that influence people and their environments. After bringing these disciplines together to develop a more comprehensive approach to applying data to complex problems, we’ve started to think about Design and Analytics as Yin and Yang — seemingly opposing forces that bring balance to one another. Analytics looks at patterns in quantitative data. Design looks at patterns in qualitative human interactions and decision-making. Together they create a whole-brain approach bringing together the analytical and the creative by combining deep human insight and broad analytical validation to deliver greater business value.
When we consider the role of design in advanced analytics today, it most frequently focuses on user interface and data visualisation. That’s a mistake. By properly incorporating designers’ expertise from the start of a project, they add value in the following ways:
- Understanding the end user’s journey. Designers bring a different perspective on the role a model will play in a person’s process and the environment in which it will operate. This perspective provides vital context for data scientists which can surface potential bias or gaps in the data that may limit the model’s impact or even highlight a completely different model architecture required to capture value.
- Decoupling product and model creation. Designers do product design in parallel with model creation. This establishes iterative and collaborative streams that continuously inform each other so that by the time the model is productionised, it’s built in a way that meets the user’s needs. Users are then eager to pick it up and use it.
- Managing change from the beginning. Designers can inject an element of change management right from the start of the process, thereby lowering barriers to adoption. Discussions around ensuring users see value in the tool, usability requirements, and barriers to use occur far earlier, informing the model creation process and de-risking adoption.
Maintaining collaboration between data scientists and designers throughout every stage of a project is crucial, as expertise from designers can drive impact throughout the lifecycle of an analytics use case. Of course, the onus for this collaboration should not be on data practitioners alone. Designers must proactively take steps throughout to take data scientists on the customer journey. They must not just interpret what customers see through the desirability lens but help practitioners see through it too, all in order to inform a better outcome.
Alongside feeding into the data science workflow, designers can build in time to help their colleagues understand a user’s mindset, needs and the process that led them to the tool in the first place. This richer understanding of users and context helps inform which hypotheses to test as features, reducing cycle times in feature engineering.
While there are many ways this collaboration yields value, data scientists and designers should prioritise the following three elements to maximise the value gained from communication:
- Objective function and target function. Understanding the user journey is important at this stage. For example, if we want to predict churned customers in the next two weeks, while it may be a good model, it won’t be usable since there is no time for business to act on these customers as the time frame is too short.
- User driven model configurations. Giving users the ability to configure constraints and inputs into the model is essential for use case adoption. Users usually want control of the model in addition to the automation. For example, in one of the scheduling studies, end users wanted to configure product specifications manually, overriding the values in existing systems. This was useful for what-if scenarios, as well as cases where users had better data based on field conversations.
- Model explainability. By working with designers, data scientists can uncover what kind of hints/intuition the model should provide to build trust in recommendations, making it less of a black box.
The value of design and analytics working hand in hand recently came together on a QuantumBlack project with an energy company. The analytics team was tasked to build a model to effectively predict the optimal use of a maintenance tool to improve the productivity and extend the life of an energy producing asset. The data science team dove into the data, exploring the features and beginning to build a technically sound model.
Learning from challenges in the past, the team brought in designers to work directly with the end operators to ensure they were creating a useful, usable and desirable tool. The designers rapidly understood operators motivations, modelled their mental model for making critical decisions about maintenance, and uncovered constraints created by the complex context in which they worked. Through this process the designers found new data features that would increase model performance as well as established trust with the oerpators, changed the format of output to limit users’ cognitive load by fitting it seamlessly into their workflow, and opened critical dialogue between teams that was critical for preemptively solving barriers to adoption.
Across McKinsey and QuantumBlack, we have found that multidisciplinary teams are key to driving greatest value from advanced analytics. A range of expertise — and a variety of opinions — can be crucial in tackling problem-solving issues and generating the creativity required from increasingly ambitious data science projects. We hope you have enjoyed this article — please do check out the final entry in this three part series, where we will provide more detailed advice on how multidisciplinary teams can thrive together throughout a project’s lifecycle.