Five things CPOs must consider before implementing AI

1) Which AI use cases are right for you?

AI has an almost limitless number of potential applications across the procurement function. During our roundtable, we heard from leaders who use it to pre-populate and validate RFPs, classify and track spend, and even continuously monitor Scope 3 emissions across complex supply chains.

But AI isn’t a magic wand that’s going to solve all your biggest challenges instantly. So, it’s important to consider what it really is you’d like to improve or streamline with the help of AI and select the most relevant and impactful use cases for your organisation.

Whatever use cases you adopt, it’s also important to ensure that you apply AI in ways that augment and extend the capabilities of your human experts within those areas. At the roundtable, we heard from numerous leaders who use AI co-pilot tools – like WNS’ own PIA+ – to help procurement professionals accelerate and automate numerous routine tasks and spend more time focusing on creating value for the business.

The ideal AI use case should:

• Help solve at least one specific challenge you’re facing within procurement

• Support your people in ways that they welcome

• Drive you towards achieving one (or more) of your top procurement and business goals

2) Is your data foundation AI-ready?

With AI, the quality of what you get out depends on the quality of what you put in. So, it’s essential to have the right data foundation in place before you get started.

Contributors at our roundtable discussed how AI demands significant volumes of data that many teams may not have access to today. While AI models can be trained using only your own data, it’s typically worth supplementing that data with external sets to bring additional context and perspective to your AI-generated outputs.

But data quality is an even bigger concern than quantity. If you can’t trust your data, then you won’t be able to trust your AI outputs. And when AI-generated insights and recommendations are informing high-value procurement decisions, that’s a serious problem.

CPOs at the discussion remarked on the need for a transparent AI core – one that helps you understand the quality and trustworthiness of your data and clearly see how it’s being processed by AI to generate outputs. This helps make the AI process traceable and transparent, so that if something goes wrong, you can quickly pinpoint exactly how and why it happened and take actions to fix the process going forward.

3) How will you validate the outputs from your AI models?

Today’s leading AI algorithms have taken years to develop and train. If you start training your own model today, don’t expect perfect results right away. And no AI model is 100% reliable, even after many years of training.

So, it’s essential that human teams validate everything you use AI to generate. Our CPOs discussed how, in the rush to rapidly scale AI, some of them hadn’t properly scaled their validation efforts – leading to poor decision-making or detrimental stakeholder conversations.

In addition to clear explainability and transparency of AI models, teams also need defined processes and accountability for validating AI outputs before they’re translated into actions.

4) How will you get your people on board and secure stakeholder buy-in?

Generative AI has dominated this year’s headlines – but it isn’t always painted in a positive light. CPOs at the roundtable remarked on rising concerns around security, privacy, and how public generative AI models like ChatGPT use the data provided to them as prompts.

All of these concerns are legitimate – especially for procurement teams that handle sensitive commercial and operational data. The last thing you want is for someone to leak sensitive contract or proposal details accidentally by asking a public AI model to rewrite an RFP, for example.

However, not all AI is generative. Our AI-enhanced Smart Risk platform, for example, currently doesn’t have a generative element, it simply makes it easier for teams to monitor and assess supplier risks, by combining human intelligence and artificial intelligence to help companies make sense of large volumes of data.

When you’re discussing AI with other senior stakeholders, draw the distinction between AI and generative AI to ensure that everybody understands the true level of risk involved with the use cases you propose.

5) How will you measure and control AI’s TCO?

Many AI models and capabilities are charged at a by-user or by-search price. That’s great when you’re just getting started. However, several of our contributors remarked on how they’ve seen AI costs rapidly spiral.

If your entire team starts using an AI model to perform a task like category monitoring – something each person does numerous times every day – or if you want to roll out an AI-powered buying desk across a large organisation, it’s easy to see how costs can get out of control.

As you start forecasting the TCO of your AI tools, it’s essential to look at how costs could potentially scale in the future. Sometimes, that will mean pay-per-search solutions aren’t viable. In areas like that, it may well be worth working with a partner to develop your own bespoke solutions that can scale more cost-effectively.