AI can speed up procurement, but it doesn’t get the final say

Think about all the products you have in your home, from the food in your fridge and pantry to your shampoo and soaps. 

Ever wondered how all the ingredients come together to really make these? 

For example, where did the eggs in your mayo come from? Where do the surfactants in your body wash come from? (These make your soap get all sudsy, FYI.)

It’s a thread that might seem a little strange to tug, but, well, everything has to come from somewhere.

And it can get a little complicated, explained Jennifer MacLean, AI partnerships lead at Unilever’s Horizon3 AI Lab, speaking at ALL IN on Thursday in Montreal. 

Especially when you have products in the homes of 3.7 billion consumers, across 190 countries, and spend more than $33 billion buying materials across global markets to make those products.

A little complicated, indeed.

What happened and who was there

Unilever’s AI lab was established in Toronto in November 2023, to find business problems where AI could help, build, and test solutions, moving the ones that work into the wider organization.

According to MacLean, who shared the stage with Gary Bogdani, head of the lab, there are over 1,500 tools across the the organization’s computing suites.

“It’s really challenging to build something new that just slots in nicely,” she said. “So one of the things that we needed to do in order to do that effectively was create a lab that could actually embed, understand, and really accelerate what’s possible with AI.“

In the case of procurement and spending that $33 billion, buyers had plenty of dashboards, market reports, and data, but they also had to balance cost, suppliers, and sustainability goals, explained MacLean.

“The challenge we had was, really, not that our buyers didn’t have access to information,” she said. “If anything, they have access to too much information, and so really being able to parse through and understand what they needed in order to make effective decisions.“

So the lab built Competitive Buying AI, a system that pulls forecasting, predictions, supplier information, and other purchasing data into a single interface.

Buyers query the system instead of gathering information across dozens of sources, but “the ultimate decision stands with the user,” said Bogdani. “They assess risks, they make decisions, they accept recommendations.”

MacLean said the system has reached 95% accuracy in forecasting material prices, and covers 90% of the materials Unilever sources to make their products.

Three takeaways

The AI they built may be one product, but the company has had to decide what to own in the process, what to outsource, and what needs to stay flexible.

Ownership

• Supplier information, supplier relationships, and the way Unilever makes decisions are what Bogdani called its ‘sovereign knowledge,’ and he said keeping that in-house is non-negotiable. Some of the models and the software connecting them can come from partners, letting Unilever bring in outside expertise without giving up what it considers proprietary.

Build for change

• Bogdani said technology has “evolved so much in the last nine months.” Horizon3 now designs systems so models and components can be replaced or reused rather than tying the company to one setup. A better model shouldn’t mean rebuilding the whole system around it.

Adoption

• What makes people start and then keep using the system? Buyers need to understand how the system reached a recommendation, and be able to flag what isn’t working, while remaining responsible for the final purchasing decision. Bogdani said the tool also has to make the job faster and easier. If it doesn’t, employees won’t keep using it.

What to think about if you’re a tech leader

What are you willing to let a vendor own?

Bogdani said Unilever wants to keep its supplier information, supplier relationships, and procurement decision-making in-house.

“There is a commercial decision that’s being made, and that puts the user at the center,” he said. 

“AI can help with acceleration. It can help with getting the data faster in the hands of users, surfacing risks, creating more visibility in terms of what opportunities the user has.”

Bogdani said Unilever doesn’t need to build all of its foundational models itself, which gives Horizon3 room to change pieces of the system as the technology moves.

“The bottom line is proprietary judgment,” he said. “You own acceleration and expertise where needed. You delegate to partners.”

If your people already have more than 90 places to look for an answer, another AI tool isn’t much of a win. The goal is to cut down the hunting and get them to a decision faster.

What to share with your C-suite

A system can be accurate and widely used and still not answer the question of what it’s doing for the business.

MacLean asked how Horizon3 balances what it can deliver now with what it wants to build over time.

“We need to show value near term, but we shouldn’t lose track of the long term value that we can create,” Bogdani said.

He said part of that means building solutions that are modular and can be reused over time.

“Foundations are really important, and that’s how you build not just scale but also trust. “

Better forecasts are useful. The next question is what decisions they changed, and what those decisions were worth.

Digital Journal is at ALL IN in Montreal this week. Follow our coverage here.

Final shots

• An AI tool earns its keep when it cuts down the hunt for information and gets people to a decision faster.
• Decide what knowledge and judgment stays in-house before choosing the model or partner.
• Better forecasts are only useful if they lead to better purchasing decisions. You need to follow up by asking what changed in the purchasing decision, and what that was worth.