The Procurement Interview: Zycus Founder & CEO Aatish Dedhia

AI has become the focal point and dominant force of today’s business landscape.

This includes procurement, where the technology is poised to power fundamental transformation. 

Aatish Dedhia, Founder and CEO of Zycus, has been at the forefront of this revolution since 2001 – long before AI went “mainstream”. His journey from solving data classification challenges to pioneering agentic AI in procurement offers valuable insights for leaders navigating the technological shift.

Aatish spoke to Procurement Magazine during Zycus Horizon at Dana Point, California.

Solving real problems with machine learning

When Zycus began its journey more than two decades ago, the company faced a specific challenge that would shape its future direction. 

"When we started Zycus in 2001, at that point we had a problem where our first customer – one of the largest companies in the world – had huge amounts of data," Aatish recalls. "At that point, they had people really sifting through the data, classifying the data and we said, ‘there has to be a better solution’.”

Coming from an engineering background, Aatish approached the problem methodically. The solution lay in machine learning – then far from the force it is today.

Partnering with a university professor who had previously worked with Google's co-founders, Zycus developed an initial AI solution called AutoClass, which proved transformative, helping to classify trillions of dollars-worth of spend.

"It was not a hammer in search of a nail," Aatish emphasises. "We had a problem, we wanted to solve the problem and we said, ‘why not look at some innovative ways of doing it?’ That's how we looked at AI and procurement."

Understanding agentic AI

While generative AI has captured global attention, Aatish advocates for a more structured approach through agentic AI: "Everything is a layer on top of each other. So, you have LLM, which is ChatGPT, which is generative AI. Agentic AI is nothing but the structure which is put on top of generative AI."

The distinction becomes clearer through his comparison to the industrial revolution. 

"You could think of LLM as the first industrial revolution,” Aatish continues. “To convert that power, you really had to create processes and factories and assembly lines in a similar way. Agentic AI is harnessing the power of generative AI, but then putting it in structured agents."

These structured agents can work in sequences or parallel, collaborating to achieve specific business objectives. Just as steam engines required careful process design to produce chemicals or manufacture goods, agentic AI transforms raw computational power into practical business solutions. 

"It's built on top of generative AI, but it harnesses it for business use," Aatish says.

Rethinking the procurement framework

Traditional procurement has long operated within the source-to-pay (S2P) framework, encompassing strategic sourcing through to tactical procurement processes. 

However, Aatish argues this model overlooks a crucial element: "One of the things people miss out on the left side is the interaction with the business users, your stakeholders in the company, which is intake because a lot of communication happens through the business users through emails, calls and so on. It was messy."

This oversight has significant consequences. Even exemplary procurement work can result in dissatisfied users due to poor interaction experiences and extended cycle times. 

"You might do a good procurement job, but your users are unhappy because of the interaction and the cycle time," he explains.

The intake-to-outcome model addresses this gap: "Intake is your front door to procurement and it goes through source-to-pay and AI to give you outcomes.”

These outcomes must be tangible – increased productivity, reduced cycle times and enhanced savings.

The built-in advantage

In implementing intake solutions, organisations face a critical choice between bolt-on and built-in approaches. Aatish firmly advocates for the latter, citing fundamental integration challenges with bolt-on systems.

"In intake, you have some solutions which are just the layer on top, where they really integrate with third-party S2P systems,” he says. “The challenge with that is integration because no application exposes the innards; no application exposes all the deep APIs and data which is available.”

This limitation creates two problems. First, bolt-on intake systems cannot access all necessary information to serve end users effectively. Second, they introduce fragility through multiple integration touchpoints. 

"The moment any of those systems change, things break," Aatish warns.

Built-in systems offer seamless integration with S2P platforms, accessing required data while maintaining the flexibility to connect with third-party systems where necessary to complete workflows.

The tail spend challenge

One of the most compelling applications of agentic AI emerged from discussions with Zycus' AI council, a panel of chief procurement officers managing billions of dollars in spend. Their message was clear: buyers should focus on strategic work rather than tactical purchasing activities like obtaining three bids for small-value items.

"That's a perfect case for AI because it's a lower-risk, smaller-value item,” says Aatish. “People don’t really have time to handle those thousands and thousands of tactical purchases.”

Autonomous negotiation delivers value through two mechanisms – first, by capturing savings that would otherwise go unrealised, Aatish says: "Normally, people don’t have time to negotiate on a US$10,000 purchase. But AI can work tirelessly and get you that saving.”

Second, it liberates human resources for strategic initiatives. The considerable effort required to manage tail spend can instead be redirected towards activities that drive greater organisational value.

Orchestrating end-to-end processes

While individual AI agents handle specific tasks, agentic flows orchestrate multiple agents to complete entire processes. 

Aatish illustrates this with a practical example, saying: "Say I need to order certain items for an event. Agentic AI would automatically identify the suppliers, create an RFQ, float it to the suppliers and then it’ll negotiate."

The key differentiator is comprehensiveness. Rather than isolated task completion, agentic flows execute end-to-end processes. Moreover, each flow should represent a significant portion of a person's role.

"If you consider a tactical buyer,” Aatish goes on, “they spend maybe 50% of the time really doing tactical buying. Autonomous negotiation can take that 50% of effort away.”

The human-in-the-loop approach

Despite AI's capabilities, Aatish emphasises the importance of maintaining human oversight, particularly during initial implementation. 

"Companies would want to ensure they do not take any extra risk,” he says. “So, initially, we always have a process where there's a human in the loop."

In autonomous negotiation scenarios, for instance, humans review AI-suggested final negotiated prices from multiple vendors before making decisions. As organisations gain confidence, they can progressively reduce human intervention. 

One Zycus customer began with full human oversight but eventually authorised autonomous decisions for items below US$20,000, within established guardrails.

"It's very important for us to have customers build trust in AI,” adds Aatish. “It just requires one wrong thing for AI to get a bad name.”

The system supports human-in-the-loop and fully autonomous operations, allowing organisations to progress at their own pace.

The strategic elevation of procurement

Looking ahead, Aatish envisages a fundamental shift in procurement's organisational role.

He asserts: “Procurement is really going to evolve from a more tactical function to a strategic function. A lot of the tactical stuff will now be done by AI."

This transformation will free procurement professionals to focus on strategic activities: building supplier relationships, innovating and identifying new vendors for product introductions. 

Additionally, AI promises to transform procurement's internal reputation. 

Business stakeholders often view procurement as an obstacle, rather than an enabler. However, AI-powered instantaneous responses and streamlined flows stand to enhance satisfaction.

"It should elevate the image of procurement within the organisation," Aatish predicts.

The quick wins strategy

For procurement leaders uncertain about AI adoption, Aatish recommends a pragmatic approach focused on quick wins: "It's best to start off with some quick wins. Take something which is lower risk, which AI can do comfortably well, and start doing that. Then you can go to higher and higher value-adding cases."

Examples of suitable starting points include autonomous negotiation for tail spend and AI-powered evaluation of vendor RFP responses.

Aatish concludes: "Once you prove the value, then everybody's a convert. And once you convert, then you can start taking other agentic flows, which can deliver more and more value."

More advanced applications – autonomous supplier onboarding or autonomous contract management – can follow once initial successes build organisational confidence.