Decision Sciences in the Enterprise

In my role as the Analytics Leader in a Fortune 500, I spend a lot of my time thinking about how to make Decision Sciences real in the Enterprise. The two main focus areas that Enterprises need to think about to start extracting value from Big Data and AI are:

1. Moving from Discovery to Implementation: Translating the excitement and potential that AI offers into business value. This is where AI needs to meet the real world of processes.
2. Moving from Expertise to Data: Augmenting human expertise with AI and Data to accelerate the decision making process and enable humans to focus on solving problems that have the potential to create true value.

And the first question that needs to be addressed is: what is the right Organization construct that will best enable this? Should there be a centralized Decision Sciences COE or should we let a ‘thousand flowers bloom’ across the organization – i.e. should every function build its own capability? Here too, the debate has moved beyond the ‘either-or’ question: it is a given that all organizations (beyond a certain size) need a hybrid model. Almost every function today recognizes the need to build Decision Science capability – and rightfully so, given how fundamentally the decision-making process is changing with data and AI.

And as happens often, the real challenge is not the What but the How: how do you go about building this Hybrid model? Let’s start with the 3 main challenges:

1. Data: The fuel that powers AI is Data. AI systems need to be constantly learning: instrumentation of processes; recording of observations. There needs to be a coherent strategy coupled with agile execution to ensure that the right data gets captured consistently. Who owns and executes on this strategy?
2. Talent: Perhaps the hardest challenge. Most organizations are struggling to get the right talent (Decision Scientists who have had some real experience are scarce to begin with). And in the distributed model, there is the problem of integrating Decision Scientists with the other roles within the function.
3. Policy: AI-based solutions often start with exploration, which needs to have a higher tolerance for failure. And then when some prototypes do show promise, scaling requires significant investment on building the data and solution assets. And then perhaps the single most important predictor of success: AI systems need to learn continuously – and in many cases, this requires supervised learning and/or tweaking of processes to capture data. This is where AI meets the human part of the organization – something that always throws up interesting policy challenges.

It is obviously not a trivial challenge – many, many things need to come together to make this real: big (top-down organizational changes) and small (series of bottom-up hacks). And underpinning all of this – fundamental changes to organizational culture. More on this later.