AI + Procurement: what are we missing?

With so much buzz in our space about AI and cognitive over the past couple of years, you’d be forgiven to start feeling inpatient about how exactly this technology will be advancing business performance. Having invested the last few years of my career in the emerging technology space and working with some great technologies like IBM Watson, I’ve seen some good first steps in the practical applications of AI. However, I also believe that we are missing some great opportunities to fully exploit the capability and I’m not convinced that some of our first steps have kept us on track in how we will fully unlock the future potential of this technology in helping us transform the future of procurement.

So, what’s the issue?

  1. Making sure we know what AI is, and what AI is not
    I don’t believe we’ve been clear on the main ingredients that makes something AI. We are running the risk of confusing or diluting the term and the value it delivers. For instance, making sure we don’t confuse AI with areas like analytics and being able to clearly validate the AI components and drive the value.
  2. Setting the expectations of AI today vs the future
    We have wonderful expectations of how AI will transform our personal and working lives. These expectations have tempted many to tackle more ambitious AI use cases first, but these larger use cases have been difficult to complete while missing out on a load of simpler quick wins. We must keep in mind that we are still in the very initial stages of developing AI engines, platforms and applications.
  3. Putting AI ahead of design
    Too often teams have been tasked with finding a way to use AI when we should be stepping back and exploring a wide range of solutions to drive performance, of which AI may be one component of a broader, more effective answer.

What is AI?

There are many definitions out there, some miss the mark, and some of the better definitions are complex. Many will refer to interpreting unstructured data such as free text, dialogue, or images. More technical definitions differentiate between deterministic vs probabilistic computing models. While all these, and many more, highlight important capabilities that can boost an AI application, alone I personally don’t class them as AI unless they have some key ingredients.

Three key ingredients of AI: Learning + Adaptation + Experimentation.

  • Learning
    The ability for a machine to consume various information and draw relationships between a range of factors in determining useful knowledge such as cause and effect.
  • Adaptation
    The ability for a machine to change its behaviour, its outputs, within context of what it has learnt.
  • Experimentation
    The ability for a machine to identify blind spots in knowledge, and be able to test a new area and learn from the outcome to enhance knowledge and drive further performance in the future.

While this initial definition may not be perfect for all, it is simple and gets the debate started with more stakeholders to accelerate our progress with what is possible with today’s technology.

The good news is that we don’t have to create entirely new models to use AI, many already exist, some of the more popular ones I’ve used in my previous projects include genetic algorithms and generative adversarial networks. A lot of ready to use APIs from a wide range of providers (see IBM, Google, Microsoft) are also available which are often free to try. These are robust starting points without recreating the wheel!

What are today’s practical use cases in procurement?

AI can be applied to procurement, supply chain and wider business operations in many ways. In procurement alone I’ve come close to documenting 40 use cases in recent years. These have ranged from:

  • Moving away from reporting dashboards and into smarter highly personalised delivery of spend and procurement insights with next best action recommendations.
  • Cognitive category research and planning tools to augment sourcing professionals with broader and better knowledge in forming their strategies.
  • Pricing intelligence platforms that use a wide range of sources to provide up to the minute price signals to improve the timing of contracts/negotiations.
  • Buying assistants for helping users requisition the most optimal items (or services) for their needs from a contracted supplier.
  • Quick and easy receipting tools that improve the timing and response rate of users while also capturing high quality user feedback about supplier/product performance.
  • Supplier performance and risk monitors that use a wide range of sources to provide early warnings of performance issues or risk events.
  • and the list goes on…

My most recent project was the forth on the above list, the Cognitive Buying Assistant while working at IBM and CognitiveScale. However, some of the other use cases are still waiting to be developed.

What is the value of AI?

Many link AI with automation benefits. I prefer to start the value conversation talking about scalability: the speed and the extent to which an AI system can scale knowledge in driving more effective outcomes. This encompasses automation but touches on some of the practicalities of where we are today.

If an AI system can learn from existing data and observations, adapt behaviours to drive better outcomes but also experiment where it doesn’t have knowledge to learn more, then it will build a valuable knowledge base in a specific domain that would surpass a single person, a team and eventually a whole organisation. The key benefit is faster and improved decision making. However, these types of AI systems require a lot of time to learn and build knowledge to become effective. Therefore the most impactful uses in most cases will be to augment professionals with better information, not replace them, for some time to come.

Key takeaways

  1. Clearly define AI and the value to your organisation, use other technologies as needed to enable your solutions but don’t blur the lines of what each component brings.
  2. AI systems need to learn before they become effective, machine learning can be accelerated by using historic data and assisted learning techniques. It is important an AI system reaches a minimum point of effectiveness before you unleash it on users otherwise it will stall.
  3. Carefully manage how an AI system experiments, boundaries need to be set up on the extent you’ll allow a system to learn in new areas, and further safeguards can be put in place by having a human review process in the initial stages.
  4. Use design before technology, use AI as one capability in the overall toolkit but always start with the design problem/ambition, come up with a range of solutions and include AI if it makes sense.
  5. Get the right experts in the room and start with small steps, the right combination of skills in procurement + design + emerging technologies (including but not limited to AI) will help you explore the best entry points for your organisation.
  6. Take risks. If you fail, make sure you fail fast, fail cheap and move forward.

I’m always open to a good conversation on topics like these so don’t hesitate to get in contact via Linkedin (click here) or via this form (click here).

Find out more about Pascal d’Arc (click here).


Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s