3 Misconceptions About AI In Procurement


Being in the AI procurement business means that our team gets an inside view into how business leaders across the industry are thinking about implementing AI in their enterprise. Across the board, we see people caught between curiosity about what’s possible and caution against snake oil. Embedded within this crosscurrent of excitement and skepticism about AI, we hear the same misconceptions over and over again, many taken from the popular media. Here are the top three:

1.    AI is better at X business process than the best human. This is likely the misconception that we come across the most. It springs from the tendency to conflate AI circa 2019 with futurists’ popular science hypotheses about what a “strong” or “general” AI would be like (which does not yet exist (and which would, theoretically, replicates human intelligence). Let’s think about this in the context of autonomous driving. Do we believe that AI driving cars on the street today is better than the best race car driver? Heck no. However, if it’s better than the average driver and doesn’t need to eat, sleep, use the toilet (or get a paycheck) then there remains a pretty obvious business case around autonomous vehicles. The same holds true for business process AI.

2.    AI only makes sense after we get our company’s data in pristine condition. This one gets a lot of airtime on the Internet blogs, so I’d like to unpack carefully. While true that “garbage in, garbage out” can apply to some AI analysis of *incomplete* or *inaccurate* internal datasets, there are now a ton of great platforms that will help cleanse and format data that may have duplicates or be missing rows. I’d actually say that it’s reasonable to expect any leading provider of AI services to offer data cleansing, aggregation and re-structuring as an enabling service for whatever their core value prop is. If you’re contemplating a “data lake” strategy as your company’s endgame, I’d suggest checking out Tamr CEO Michael Stonebreaker’s excellent talk about why “data lakes” are just the beginning. Cleaning and integrating datasets can produce enormous value, but much of what AI can help with today is data labeling and pattern recognition to assist with that taxonomy. Ultimately, this is an iterative, nonlinear process rather than a “first clean data, second AI” process, and will run concurrently in most organizations that successfully implement these platforms.

Do we believe that AI driving cars on the street today is better than the best race car driver? Heck no.

3.    AI only works in specific domains, such as computer vision, image recognition, etc. While the state of the AI market itself remains a subject of controversy, there is no doubt that the advances in computer vision are certainly impressive (and will likely be responsible for disruptive innovation in manufacturing over the next decade). However, these are not the only places that AI applications can cause substantial improvements in process efficiency and outcome optimization. For example, many AI theorists point to the moment that Google’s DeepMind won at Go as a kind of “Sputnik moment” for renewed excitement about AI. Consider that this breakthrough moment occurred around a wave of so-called “gameificiation” where tasks in business, education and tradecraft professions are being recast as games with optimal pathways and rewards and punishments. How unreasonable is it to imagine that if AI can win a game as complex and nonlinear as Go, that AI might be able to win a game as simple as, say a price negotiation?

The truth is that AI in the workplace will inevitably be caught between the overpromises of visionary optimists and the bitterness of skeptics who see the rise of AI thought leadership as snake oil. Our approach to these hot and cold reactions is simple: what are your savings goals, and would they become more achievable if your team could negotiate with twice as many counter-parties over the next year? What about 10x many counter-parties? If having a procurement team that could work 10x tempo is an exciting proposition, then there’s probably a low-commitment way to run an experiment in your own organization and see the value for yourself.


Bid Ops on LinkedIn

Bid Ops on Twitter

Bid Ops on Facebook

Bid Ops on Instagram