AI is now becoming table stakes for enterprise software. But AI solutions are different. Several strategies for driving successful digital transformation with AI are completely opposite of what has been conventional wisdom for decades.
For over 30 years, I have been privileged to help scores of global manufacturers and retailers transform their operations. In doing so, we have gone from leveraging rules based expert systems back to supply chain optimization systems and then adding in AI based execution orchestration and recommendation systems during the last seven years. Here are four strategies that may seem counter-intuitive at first, but I have found in my experience to be highly effective for driving successful transformation with AI.
Design for Garbage In – Good Stuff Out
Of-course, “garbage in; garbage out” is still true. However, we all know that enterprise data is never clean. With big data the problem is only amplified further. Data is indeed the new oil; but we need refineries. With AI you can build a data cleansing, harmonization and enrichment pipeline as a part of your solution using machine learning models. At LevaData, we process millions of data points this way daily to refresh content that serves as a foundation for our direct materials sourcing solutions. I recommend you design your digital transformation strategy to take garbage in and pump good stuff out at scale into your enterprise applications. You can leverage content from solution providers like us to get a jump start.
Start with the trivial many
Conventional wisdom says use the pareto principle to focus on the significant few first. Companies look to get maximize returns from their IT investments. But, as we all also know, along with data veracity the other biggest challenge to successful digital transformation is change management. In the early stages of a transformation journey there is a lot of distrust in AI based recommendations – particularly for decisions that impact the most consequential segments of your business. This mistrust is well founded. It takes time for the data pipeline, the business processes and machine learning models to mature. For these reasons, I frequently recommend my customers to start with the tail of the pareto rather than the top business segments. We should pick segments that will deliver significant enough impact to the business to demonstrate noticeable value but avoid the core segments where people will be understandably the most protective. Think of the disruptive innovator curve in planning your change management journey. At one manufacturer, we deployed AI assisted processes to reveal opportunities in certain business segments that had been left largely to the control of outsourced parties. Soon, the question naturally came up as to why we could not apply the same ideas to the core business and drive even higher value. With AI solutions, you can start at lower end of your business. But, as the business processes mature and trust in the AI solution takes hold, they will displace legacy solutions in increasingly more consequential segments of the business.
Optimize the learning not the math
Traditional decision support systems were all about developing solutions using the most sophisticated optimization technology. The focus was on getting the optimization model right and precise. However, business conditions change faster than traditional optimization models can keep up. We still use sophisticated math with AI. However, now the math is applied to “learn” new features and continuously train and adapt the AI solution to the changing business environment. In working with global manufacturers, I have found that simple solutions that learn are far more effective in realizing business value than highly sophisticated optimization models that are brittle and fail to adapt.
Bootstrap, test, learn and iterate
You cannot use traditional approach of with minimum hurdle rates for evaluating investments in AI. AI systems are different than traditional enterprise solutions. You have a dilemma akin to the chicken or egg first problem. AI solutions must be used for them to become useful. This is another reason why I advise my customers to start out by deploying AI to segments of the business that would have gone mostly neglected otherwise. We must set management expectations that the bigger returns on investment will come over the long haul. The ROI hurdle rates are met only after sustained use and ongoing refinements. It is critical to identify use cases that drive initial value through automation and agility just enough to fund the initiatives. You must expect that the systems will be error prone to begin with and you must build safeguards for manual curation and oversight to bootstrap the usage. You must test, fail fast, learn and iterate. AI solutions require continuous retraining. The trick is to bake in the retraining as part of the process and system itself. Once you overcome a critical mass of maturity, a virtuous cycle takes over to make the solutions increasingly useful with greater adoption.
I refer to AI as “Augmented Intelligence” for the business. AI solutions can augment and upskill your business processes at scale. But you must embark on your transformation journey with your eyes open and expect it to be very different than the norms that we have come to accept for traditional information technology initiatives. Just as other disruptive technologies, the initial outcomes may be modest but they put you on a path that will help you scale far beyond the limits of traditional technologies.