Building AI Unicorns

Everyone talks about AI, but how do you really do it and how do you do it at scale? Which challenges does companies face when starting on their AI journey, how do you extract most value out of AI, and how do you drive AI@Scale?

Typical challenges when starting on the AI journey:

• “We do not understand it”: companies struggle to understand the “art of the possible” which makes it difficult for them to know where to start
• “We do not have good enough data”: many companies use poor data quality as an excuse for not getting started, and does not take into account the amount of data they already have, the fact that more and more external data is available, and the fact that you can gather new data as you go along
• “We hired a bunch of geniuses, with no impact”: once getting started many companies hire a few new data scientists, put them in a war-room, and expect results. However, without top management anchoring of the problems they solve and how to drive and scale the solutions only limited impact can be achieved
• “We have small initiatives all over the place”: typically many smaller AI initiatives pop-up here and there. Without strong top management driven governance these will easily drown in governance processes and don’t get the needed focus
•  â€œWe have a fancy algo, but cannot scale”: while algorithms are intellectually stimulating and interesting the key to scale is technology in general (incl. data platform), change management & process, and organizational set-up/readiness which often gets neglected
• “We spent eternity to build a data lake with little benefit”: do not treat building a datalake as an isolated IT project. Data platform projects needs to be built in very close collaboration with the use case/algo teams in order to gradually and value based build up the needed data and infrastructure
• “We got it, no one uses it”: creating algos alone does not drive business adoption. For business users to use algos a new breed of front-end tools are needed
• “We are not able to hire the star”: Data scientists and engineers are very sought after talent, and most companies are struggling to attract them. If not combining with external resources when needed this can become a bottleneck for an organizations ability to scale

The key to success with AI is to think big, start small and grow fast; from an algo, data platform, front-end and organizational point-of-view.

Typical AI use cases with high value potential span the entire value chain, and vary greatly by industry. Typical high potential AI use cases by industry: B2C - Personalization, Industrial Goods - Predictive Maintenance, Transportation/Telecom - Network Optimization, Financial Services - Risk and Fraud Management, Back-office heavy industries - workforce deployment design.

While its relatively easy to prove the value of AI pilots, the challenging part is to scale. To scale 4 areas are needed:

• Strategy fit for the AI era
• Selection of high value pilots and use cases
• Building use cases gradually scaling algo, data platform, devops platform, infrastructure, and front-end applications
• Transform operating model for AI@Scale. A typical good way is to start building the capability externally, gradually infuse more and more resources from the company, and finally hand over the organization to business as usual (in IT, a new algo team and daily operations)

To recap. The potential of AI is large, yet the challenges on the way are many. To get most value out of AI one need to think big, start small, grow fast! There are many typical AI use cases, yet their relevance varies by industry. Getting started: Develop strategy fit for an AI era, select 1-2 use cases to get started, scale use cases and data platform gradually, and drive organizational transformation in parallel.

 Note: Above comments are my personal opinion, and not official statements from BCG.