The Practical Uses of AI for Procurement

One of the discussions at eWorld recently came from Julien Nadaud, Chief Product Officer at Determine; he talked about the practical implications of AI in procurement and contract management, putting its use into real contexts. It was an interesting session, attracting a full room of delegates, in which he squarely layed out the real use cases for AI we can expect to see and how they are impacting related tasks.

He’s written quite a bit on that subject by the way – some of which can be found here.

Procurement is becoming more 'intelligent' -- this we know. AI learns continuously (with machine learning at its core) using big data coupled with historical knowledge of successful outcomes. The power for procurement, says Julien, lies in the recommendations AI can make for new situations, and in replacing time-consuming jobs (like spend classification, supplier matching and contract review) and the time it frees up for tasks that add value to the process.

The use of AI in procurement really comes down to helping us make the right decisions. But there’s a lot of hype around AI generally, so to prevent us from ‘not seeing the wood for the trees’ he maps out four headings that apply to real use cases:

Classification – We know that AI can save us money, basically because it can replace certain human tasks. In terms of classification, AI can classify items into categories in a way that wouldn’t be possible without it.

Recommendation – is about anticipation, giving users the best options to choose from, be that of products, suppliers or contract types. And it does it in a speeded-up, optimised way.

Prediction – AI helps with spend prediction, price prediction, risk prediction, payment delays and so on – leading to savings, better visibility and lower risk

Conversation – is about normal discourse, the ability to have natural conversations through the likes of chatbots, which, the more you use, the more they understand.

So some ideas within these 'buckets' are:

While we can ‘Google’ supply chain risk, hurricane activity for example, it will not be 100% accurate and takes a lot of work. But with machine learning, those findings can be analysed, compared, and tell you if your supplier is at risk, or moving in the right direction.

When buying online, AI can tell us what we will buy next based on what other users bought together. This used to be too labour-intensive, admin, working out the relationship between products, had to configure catalogues to link items together. But AI enables this, not replacing us but making us more efficient.

Then, following a purchase, it can make suggestions and even place orders, or put together a kit ready to be added to the basket. Recurrent purchases can also be detected along with the lifecycle of the product and when it needs to be replaced.

Coming soon, we will be able to analyse every vendor. Supplier information, their activities, online catalogues, websites, their PR and social media networks, will be understood through AI, analysing our requirements and matching potential suppliers for us to source.

Data from different vendors, can be used to make correlations between price and context. For anything from paper to rubber tyres, making the right product decision at the right time with the right vendor, through AI prediction, can save time and money. If prices are going up, it can advise purchasers to wait, or  buy quickly before a seasonal event – like schools buying up paper for the start of the academic year for example.

Contract data can be easily extracted – requirements, dates, specs, clauses, obligations etc. But just loading contracts is only part of the process; if the user adds valuable data, like associated suppliers, the more it can analyse, the better the results. Once all the metadata is there, AI will improve it, bringing more value. We have already learnt that in legal document review, AI outpaces lawyers.

Then there are the conversational agents, like Alexa or Cortana, capable of discussions that are far more natural than clicking on a series of apps and icons, navigating screens, searching for criteria. A smart agent can help with buying, and confirm the purchase. Could we replace the full web front-end software (and all its complexities) with a simple user interface, the smart agent – maybe?

Takeaways:

We can have more accuracy, data consistency, analysis that covers 100% of data, continuous quick learning and more detailed results.

We can have faster, almost real-time processing.

We can augment human capabilities, getting the right information to make right decisions – and share them.

We can make new savings through efficiency, processing huge amounts of data and continuously learning from it

AI is inevitable – we are generating more and more data and the only way to process it is through AI, and that will create more value. The need for data scientists – in most use cases – will diminish; teams that were expensive to use and manage will disappear. AI doesn’t need the developer or the analyst. There is no need to build and maintain complex algorithms, or expert systems; the hardware to run neural networks is available now.

So will AI use AI to analyse the best model to run AI? making AI even cheaper to run and maintain.

We do note however, that as we become more dependent on data, the cleaning and processing of it is critical, or the system will learn the wrong things. And we met someone at eWorld who maintains that it is the people who are behind the classification of data that makes for real AI success – and we’ll be hearing from her soon.

First Voice

  1. Secret Squirrel:

    Looking at these in turn……

    Classification – It can classify only to what we teach it to classify. You will be able to classify some automatically but it shouldn’t come up with new insights into how classification can happen. But a rules engine could do much the same.

    Recommendation – That’s possibly fair but only to what we’v taught it to recommend.

    Prediction – This is a touch spurious, IMO. Econometrics shows exchange rates are unpredictable, so a lot of price prediction goes out of the window. You might be able to put in some externalities into an analysis (like weather for crop yield estimation) but such things aren’t beyond a well formed rules engine using ‘reference classes’. It might also help spotting payment delays but nothing that a well designed data warehouse analysis wouldn’t spot (IMO).

    Conversation – That’s fair.

    To my mind, outside of natural language processing, AI is the new data warehouse and its prime purpose is a new ‘tool’ for software suppliers to sell and the same investment in your existing systems will be less disruptive and as effective.

Discuss this:

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.