AI in the Enterprise: Hype vs. Reality

I am speaking at the AI Innovation Summit in San Fransisco later this month (https://analyticsevent.com/) on this topic. Here is the outline of what I am planning to cover. Would be great to have your feedback – nothing like crowdsourcing!

AI in the Enterprise

Step out of the giddy enthusiasm about all things AI in the FAAMG and their extended orbit of Tech companies into the wider world of physical goods and services – and one thing you will often notice when the topic of AI comes up: there is fatigue, even cynicism, on how ‘AI can transform your business’. And not without reason – and we will begin by exploring why that is so; and what an Organization’s manifesto to truly leverage AI might look like.

To be sure, AI in the Enterprise is overwhelmingly Machine Learning – for the rest of this discussion, I have used AI to encompass all of its offspring, including its most famous child, ML.

From what I have seen over the years, the gap between hype and reality with AI continues to widen because of these main reasons:

1. AI for the sake of AI – do the projects create true business impact? AI projects continue to be treated as sporadic experiments without the right focus that comes from Programs that are linked to business objectives
2. Lack of Organizational capability – where are the career paths? Hiring is often in pockets and there is no Enterprise wide strategy of building this talent base. No wonder AI/ML engineers don’t find rewarding careers
3. Data may be the new oil – but are the pipelines working? The traditional mindset of EDW and Data Models have made access to data not just cumbersome, but also slows down AI projects
4. The bogey of Determinism – does Bayes have a place in the Enterprise? Transactional Systems of Record (e.g. ERP) are built on enabling decisions through precise rules. In reality, decisions are almost always probabilistic.
5. Crossing the Scale chasm – are AI projects engineered for Scale? As AI projects continue to be executed in siloed projects, there has been no investment in an Enterprise level AI platform infrastructure

How to get AI front and center of the Enterprise

So, what can we do about it? Like most things that are transformational in nature, you need to start with a few North Stars – big goals that can serve as the guiding light. And here is my 5-point Manifesto:

1. Be ruthless about Outcomes:
   1. Quantify clear outcomes from AI projects that can be measured in business KPI terms.
   2. Measure the impact of AI projects as the rate of change in the business KPIs – remember that AI systems are designed to learn and get better over time
2. Invest in building Organizational Capability:
   1. Build a core team of AI/ML Data Scientists and Engineers to solve some of the hardest, cross-functional problems
   2. Enable a collaborative eco-system of AI capacity across business functions
3. Elevate Data to be a first-class citizen:
   1. In an AI world, elevate data to be a core strategic asset – governance and investments should reflect that. Do not let this be relegated to a BI/DW initiative with a few data architects.
   2. Make ‘Data as a Service’ real – democratize data by enabling wide access, going beyond structured data to the ‘dark web’ of unstructured data. Do not let the traditional ETL mindset be the bottleneck for getting access to data.
4. Integrate Probabilistic systems into the fabric of decisions:
   1. Organizational learning is in its essence, probabilistic inference. Embrace the notion that “all models are wrong, but some are useful”. This needs to be made real in two ways:
      1. Allow for processes to accept probabilistic decision points (across ERP, RPA solutions)
      2. Deploy AI models with the clear expectation that they need to show continuous improvement like any learning system: get better as they learn from data
   2. Learning is an active process through Experimentation. Exploit the power of experiments to continuously improve AI systems by getting a second level of feedback from field trials
5. Invest in ‘AI Platform as a Service’ –
   1. Reduce the friction of deploying AI at scale by investing in a Platform that enables a standardized process of deploying AI systems
   2. Recognize that the best Platforms are continuously evolving – and instead of monolithic systems, create an Engineering led eco-system of plug-and-play modules that can be assembled to create AI/ML systems

So where do we start?

To be sure, it would be hard to argue against this manifesto. The question is as always, that of execution. I am not saying that these initiatives are MECE or should be attempted in a sequential fashion: at the same time, attempting all of them together is how corporate disasters start. So where do we start?

The truth is that your typical Enterprise is a complex system – and taking inspiration from the Theory of Constraints, you might want to start with the one constraint that would be the ‘weakest link in your system’. What would be that ONE initiative from the manifesto that would be your starting point?

In my experience, it is the Platform question (#5) – this is one area that has the potential to significantly move the needle on other initiatives. More on the Platform question in a later post – and I also plan to cover this in my talk at the AI Summit. What would be the one constraint that you would want to solve?