The AI We Didn't Build Matters Most

There's a question I keep getting asked: "How is Pivot implementing AI?"

‍

I sat down with Nicolas Marchais, Co-founder and CEO of Phacet, to share what agentic AI we’re building, along with what we’ve deliberately chosen not to build. With software vendors racing to put AI everywhere, the ability to practice restraint is becoming just as important as investment decisions.

‍

We were born in the AI wave

Pivot launched in 2023, right as the generative AI wave was breaking. We built the AI operating system for procurement to give teams the structure, clarity, and velocity they need to operate under increasing pressure. 

‍

We could’ve made a lot of AI noise just as the industry and headlines were pushing for. But since the beginning, my co-founders and I asked ourselves a simple question: what does AI need to truly add value for our clients? The answer was simple: clean data.

‍

We recognized that you can add as much AI as you want on top of unstructured, incomplete data, but it will never reconcile nonexistent financial information, nor will it fill in the gaps your system of record left open. And if procurement teams, who manage a lot of pressure and responsibility, can't trust the output, they won't adopt it. 

So, before thinking about how AI would be fundamental in Pivot, we focused on building a sturdy foundation that would make AI earn its place. 

‍

Pivot’s three-layer architecture 

We think about Pivot in three architectural layers:

‍

Layer 1: System of Record

The system of record (SoR) refers to the data layer where every request, contract, and invoice gets properly recorded and centralized across the enterprise. You can think of the SoR as a database that isn't necessarily visible to end-users. For most companies, that foundation typically lives in an ERP.

‍

It’s for this very reason that we've invested heavily in making our ERP integrations as deep and native as possible. We believe the success of the procurement operating system is dependent on maintaining the integrity of existing systems.

‍

Layer 2: System of Engagement 

A system of record is only as good as the data flowing into it, which is where the system of engagement (SoE) comes into play. Procurement operations are traditionally full of bureaucracy, which causes procurement process adoption to take a hit. By integrating with communication tools that teams already know (e.g. Slack and Teams), the SoE removes adoption challenges that have traditionally compromised enterprise procurement software. 

‍

Layer 3: System of Agentic Action and Automation

The system of record and system of engagement were built with a third layer in mind: The system of agentic action and automation. Agents built on incomplete or unstructured data are susceptible to producing outputs that procurement teams cannot trust or act on. A system of record enriched with clean data is what renders AI agents reliable and quick to deploy. Without a solid foundation, agentic AI in procurement is equivalent to automated risk. With a sound architecture, you turn your agentic capabilities into trusted, measurable levers of value.

‍

Why we built AI agents around our product, not inside it

Since the beginning, we knew where we didn’t want to put AI agents: in Pivot’s core. We keep agents around the product, meaning they’re pluggable, modular, and deployable on top of the system of record. Three reasons drove this decision:

‍

First: our clients need deep customization

We serve enterprise clients, and when you get into the details of their use cases, no two are exactly alike. If we were to put AI agents into Pivot’s core, we'd be endlessly building customizations that create technical debt and user complexity. This is precisely why our agents are modular, meaning that we can build what each client needs without compromising Pivot’s–and our clients'–core needs.

‍

Second: enterprise clients want control

Enterprise companies want to keep total control over their security, data, and governance. That means being able to choose which AI model is used, how it's hosted, and what data it can access. For example, a company might be comfortable routing purchase orders through an AI model, but draw a hard line at contracts. By keeping AI agents modular and pluggable into Pivot rather than baked into the product itself, each client can define exactly where AI operates and where it doesn't. Building this into the product off-the-shelf would mean endless customization, and a one-size-fits-all approach that ultimately only fits all but one.

‍

Third: the technology is moving fast

Lastly, technology is moving quite fast. We need to be conscious of a tech trend’s expiration date and accept that a new trend could emerge in 6 months. From a product standpoint, if we had deeply integrated the AI landscape 18 months ago, we would’ve incurred significant technical debt, rendering the operating system rigid and difficult for our customers to get the benefits of the latest AI models.  

‍

The solution? The AI Operating System for Procurement

The answer to all three constraints above is the same: build the AI operating system for procurement. 

‍

Pivot functions as the foundational layer, and AI agents live around it, plugged in on top of a clean, structured system of record. Agents are modular and deployable per client, per use case, per data boundary. To make that real, we built an in-house AI studio, a small team that combines product and no-code build capabilities with enough business context to translate what they build into measurable ROI. They go deep with each client, understand the use case, build or configure the agent, and demonstrate that it works.

‍

The studio can build agents for clients who want a turnkey solution, or integrate agents that clients bring themselves. Either way, they plug into the same foundation, which means the operating system gets more capable over time, without getting more fragile.

‍

Use case: the Negotiation Agent

To share a concrete example, we built a custom Negotiation Agent for our client, Lemonade, to address their unnegotiated contracts. In other words, untapped value hidden in plain sight. 

‍

Procurement teams have a fundamental problem: they're often understaffed relative to the volume of purchasing activity in their company. So when it comes to negotiating contracts, they focus on the top 10%, the big ones, the strategic vendors, meaning that the remaining 90% gets minimal review. Yet finding savings is one of the top three missions of any procurement team.

‍

We built a Negotiation Agent to equip procurement teams with the means to negotiate the often forgotten 90% of contracts. It does so by:

• Analyzing all the data on the request
• Benchmarking the pricing based on the company's size, the supplier, and the specific line items
• Uses web data to find what standard pricing looks like for companies like yours
• Sends a notification back to the requester

‍

The added value? That can translate to up to 10% in additional savings, because negotiating is no longer limited to procurement teams.

‍

Why we intentionally chose not to build some AI capabilities 

We were just as intentional about what we chose not to build as what we did build. We built the AI operating system for procurement with the idea to accelerate value through a foundation strong enough to support it. From a product perspective, this meant that we did not build the agentic layer prematurely in order to build a robust core. 

‍

We focused heavily on the system of record and the system of engagement because clean data is the prerequisite for all subsequent AI investments to pay off. What we have now is the foundation to welcome AI agents in a fashion that gives enterprise procurement the long-term ROI they need without getting overly attached to building in-app capabilities that may become obsolete.

‍

Curious to learn more about Pivot’s agentic AI capabilities? Feel free to reach out.