Enterprise AI: Hype vs. Reality

During the early days of the IT hype cycle in the late 1980s, the economist Robert Solow famously said, “You can see the computer age everywhere except in the productivity statistics”. This came to be called the ‘Solow computer paradox’ and spun a body of research on the true impact of IT spends on productivity, both at an enterprise level as well as more broadly, economies. Anyone who has worked in the corporate world in the last 20 years is fully aware of the fact that the productivity gains really came through Process Re-engineering, with IT acting as a key catalyst and often, the primary enabler.

Data and AI is going through a similar hype cycle – while that is driving the investment boom, the returns are not so obvious. While AI is clearly making a big difference in the B2C space (think FANG), this is not so apparent in the Enterprise world. And as it was it IT, Enterprise AI has some ways to go – and requires deliberate thought and approach.

What are the barriers?

1. Data: AI, more than anything else, relies heavily on data. In addition to the initial training of the models, there is the continuous learning and improvement. And this is a big gap in most enterprises, across use-cases. Enterprise data is, to put it mildly, messy – datasets are disparate and not easily stitched together; data pipes are unreliable. For instance, turns out that creating a 360 view of the customer journey across all touch points – from Sales, Product Engineering, Customer Support et al is a devilish task. Given that, looking for an end-to-end AI solution that spans across an entire value chain usually turns out to be an exercise in frustration.
2. Point solutions: AI systems more often than not, cater to very specific use-cases. The Customer support team might invest in a text mining solution that can extract valuable insights from call-logs. However, these insights are only as useful as the processes that they could impact. Can the call-log insights be used to trigger the ‘Next Best Action’ trigger by the Account Managers for the named Accounts? Can the insights be fed back into Product Engineering for feature prioritization? This is an example of an AI solution triggering a tweak in the business process. This is especially true in the enterprise context where value usually gets created at a process level and not at an individual task level. And then there are opportunities to re-engineer entire processes using AI as the primary catalyst. The most obvious example are chatbots – the increasing sophistication and accuracy of the chatbots has provided an opportunity to fundamentally re-engineer the entire Customer Contact Center.

What does this mean?

There are three immediate prescriptions:

1. Do NOT get sucked into a pure Data Quality Management cycle: While it is tempting to embark on a fixing the data quality, it is bound to be a futile exercise, especially if it is treated as an end in itself.
2. Identify Outcome led process re-engineering opportunities: Process re-engineering is bound to come back, albeit with a twist. Start with a measurable business outcome (e.g. improve First Touch Resolution – FTR – by x%, Reduce No. of customer contact points by y% etc.). Take a look at some of the AI led capabilities (there are several available off-the-shelf) and marry them up with the expected outcomes. Then comes the hard journey of making the AI capabilities real.
3. Execution, execution, execution: Gone are the days of the multi-year process re-engineering programs that sucked millions of dollars over the last 3 decades. In an environment where change is accelerating and increasing in complexity, there is the need for a solid execution mindset with a strong feedback loop. This warrants a whole post.