More Pilots Than an Airline: Scaling AI and Digital Transformation
Learn about common pitfalls companies struggle with as they move from PoC to actual value at scale from IoT and machine learning and what you can do about it.
Referring to their internal digital program, a customer once said: "On Artificial Intelligence (AI) we have more pilots than a major airline". This is a common pitfall companies struggle with as they move from proof of concepts (PoC) to actual value at scale from internet of things (IoT) and machine learning. Why? What can they do about it?
WHY? SOMEWHERE BETWEEN PILOTS AND HUGE DATA LAKE PROJECTS
In fact, we often find customers "stuck" in one of two situations:
- Stranded successes: They’re running an endless stream of PoCs and pilots, but they’re never scaling any of them properly.
- Never ending plumbing: They’re running endless data lake programs without ever arriving at real business value.
Often there are many technical factors and arguments justifying those situations, but I would convey that it’s mostly a mindset and attitude problem. These are also supported by faulty assumptions.
Also read: Why do so Many AI programs Fail?
ENDLESS POC AND PILOTS
In case of endless PoC and pilots, the problem often lies in how we handle risk. To move forward, we need to accept that we’re going to run a portfolio of initiatives to scale, and some will inevitably fail. The major problem in large industrial companies is that there’s (rightly so, for the most part) strong adversity to failure. The culture in these organizations is about excellence and risk mitigation. Failure for some of their products simply isn’t acceptable (who would like to fly on an airplane with a minimum viable product (MVP) jet engine?).
However, for the majority of IoT / machine learning / AI applications we’re talking about "open loop" recommendations. These can bring considerable value but aren’t "life-threatening" as the human is always in the loop (vs. "closed loop" where the machine decides). If a prediction system doesn’t work, the worst that could happen is that you didn’t save any money.
ENDLESS DATA LAKE PROJECTS
In the case of endless data lake projects, the faulty assumption is often one of the following:
- "We first need all the data, then we can start working on valuable use cases"
- "If we don’t create an all-purpose data lake we won’t be able to scale applications"
In principle, the idea of a data lake (or data platform or similar nomenclature) isn’t wrong. It’s the way you get there that’s often the problem. Decades of studies on large IT projects have so far proven that this strategy (first a data lake/platform, then the use cases) doesn’t pay off. Both need to go hand in hand.
WHAT CAN YOU DO ABOUT IT?
The approach that we have developed at Arundo, comes from learning from these pitfalls. Many of us, myself in particular, have been through those in our past work experiences and made our good deal of mistakes and learned from those.
Here’s what we learned:
- Courage: Start with a portfolio of PoC/pilots but dare to quickly choose a few to push to full scale while accepting that some will fail.
- Value: Relentlessly focus on value. Even when we’re working in standardizing some part of what may become the future data lake/platform, everything that we do needs to be linked to a valuable application scaled from point 1.
- Speed: Aim to develop, deploy and use IoT / machine learning applications within 90 days. This is a good rule of thumb used in other digital native industries (mostly B2C like media, digital services) which is a welcomed revolution in asset-heavy industries. Develop and launch within a quarter to ensure fast learning cycles.
- Embracing creative chaos: Innovation is often a non-linear process and it can happen anywhere. Be ready to embrace new ideas, prototypes and quickly bringing them to scale even if they don’t fit your original digital program slide.
Running a digital program and implementing novel solutions successfully in the IoT and machine learning space isn’t easy. We often try to "control" our fear of failure by not committing to scaling PoC, or by blocking innovation and application development until we have all the data available in a single, standardized universally available repository.
To win in this game, we need to shift our mindset towards attaining value, have the courage to take some bets, accept inevitable cases of failures, and create a culture capable of embracing creative chaos.
Want to learn more about the Industrial Internet of Things, how it will disrupt traditional thinking and its barriers? Have a look at our whitepaper: The Executive Guide to the Industrial Internet of Things.