Stuck in an AI pilot rut? Here’s how to get it right
“…much of what we do with machine learning happens beneath the surface. Machine learning drives our algorithms for demand forecasting, product search ranking, product and deals recommendations, merchandising placements, fraud detection, translations, and much more. Though less visible, much of the impact of machine learning will be of this type — quietly but meaningfully improving core operations.”
~Jeff Bezos, Founder and CEO of Amazon, Letter to Shareholders(2017)
Leo Tolstoy famously wrote in Anna Karenina, “All happy families are alike; each unhappy family is unhappy in its own way.” After talking with over 250 companies in a variety of industries around the world and successfully working with dozens of companies in manufacturing, logistics & transportation, and consumer packaged goods, we have concluded that while there are many ways for companies to fail at deploying their AI pilots at scale, there are also some surprisingly common mindsets and behaviors among companies that succeed in deploying AI. Recent surveys indicate that only between 20% and 30% of AI pilots are progressing to at-scale deployment. This is a missed opportunity creating a massive waste of resources and lost momentum that prevents companies from gaining a competitive edge.
All of this can be avoided if executive management teams can consciously keep three guiding principles in mind about successfully deploying AI in their business operations.
- Take a long-term view on how AI can benefit your company and commit the necessary budget to support it. Clearly understand that it might take 6-12 months to start getting value from your first AI deployment. Integrating AI capabilities in your operations takes time to pick the right problems to solve, wrangle the data out of a wide array of IT and OT systems, clean it, and build, test and deploy the AI models into your existing work processes. The good news is that, after you have deployed 3-4 AI pilots into production, things start moving faster – data has mostly been aggregated, data pipelines have been built, and there is also greater organizational momentum and support after employees have seen a few AI successes.
- Start with an intent to deploy rather than to just experiment. This implies only picking problems that have business value, and defining what success looks like for the pilot instead of “just wanting to do something with AI” at your company. You should also pick problems that truly help take advantage of the unique capabilities that AI and ML techniques have to offer compared to other methods of solving the problem. You need to start thinking, at the outset, about how the AI solution will get integrated in your existing work processes and used by the people whose decision-making it is expected to augment. Lastly, and most importantly, investigate whether you have enough data in terms of quantity, history, and variety to increase your chances for a pilot that successfully transitions into deployment at scale.
- Understand the value of learning by doing in enterprise AI. In our experience, it is difficult for companies to just look at published case studies and replicate them. Success in AI depends on many of the contextual factors we have briefly alluded to above. Approaching pilots with an intent to deploy and going through that process in a systematic manner helps companies gain valuable insights regarding the state of their data, how to manage AI-related organizational changes effectively, etc. that they can continue to build on with each new AI pilot.
Remember, your competitors are also trying to figure out how to do these things effectively. The key to realizing value for your business KPIs from AI is to simply… start. Taking that first step is essential to integrating AI into your business and seeing returns. Remember, also, that there are some wrong ways to start. Choose a manageable problem for your first solution – a standalone problem, something you can measure and assess easily. This will become the building block for future AI deployments in your business. The challenge is not waiting too long to get going. Because of the time to on-ramp an initial deployment and bring AI to success, delay can create a situation where you fall behind trying to catch up to competitors – it’s better to lead AI implementations and create opportunities for advantages instead.
- Atlas, Noodle.ai’s Machine Learning (ML) Framework Part 2: Design Premise & Architecture
- Experiencing Supply Chain Blindspots?… Enter Athena Insights
- Atlas, Noodle.ai’s Machine Learning (ML) Framework Part 1: Challenges With Building AI Applications
- Selling Certainty in an Uncertain World: Why Noodle.ai is a “Top 15 Startup to Emerge Stronger from the Crisis”
- The Marketing of Industry 4.0: The Hype Cycle Ends Where Reality Begins