Your browser doesn’t support HTML5 audio
Spend management vendor Coupa has laid out an ambitious plan to automate B2B commerce using AI agents. At its annual Inspire conference this year, the company shared a roadmap for its planned product investments over the next three years focused on this goal. Salvatore Lombardo, Chief Product and Technology Officer at Coupa, told attendees:
We are no longer just an app company. We are a data-driven agentic AI company, and we are just getting started. I'm excited and confident that together we will build the network that powers the future of trade.
Its challenge is to keep the faith of customers as it embarks on this journey with them. I had a chance to catch up with Lombardo at the company's London event and asked him how he plans to keep customers on board. He tells me that delivering new capabilities incrementally, and at pace, is a key part of it:
It's not a multi-year thing here that we are talking about. If you have seen even today, three of the agents are released. Three others are built and just training, in limited availability... It is a step-by-step, very near, one, two years, when you think about the finalization, maybe, of all the agentic skills, that's number one. And we have a lot of AI-native capabilities already built into the platform by itself.
The other consideration is that, while customers could turn to point solutions from AI startups, their preference is to have all the functionality within the existing Coupa platform and its Navi AI interface, provided it becomes available on a similar timescale. He goes on:
When you are on Coupa already, you are quite digitalized. You run one platform, you have one data set. You don't have all these integration problems, which maybe others have... [If] Navi can do the same as start-up solutions with 20-30 people... why [go elsewhere]?
The agent landscape evolves
Coupa's agent roadmap encompasses skills and activities on both the buy side and the sell side, from document discovery, analytics and supply chain modeling, to sourcing and supplier onboarding. More are being added to the list, along with others that partners and customers are looking to build on the platform. Lombardo tells me:
Customers are coming up with ideas, and this is what exactly I wanted to trigger. They're coming up and saying, 'Salva do you think you will ever build that skill? Or is this a skill which I can prepare my IT team whenever you open the platform to build?' That's exactly the discussion we need to have.
As time goes on, some early agents may become redundant as more sophisticated versions come on stream, while enterprises become more comfortable with giving them more autonomy. He goes on:
There was one aspect I tried to build into the keynote, which is, not only, okay, what is the value we deliver, and so on, but the evolution... I'm not believing that Navi will do everything at once and 'Boom!' it's in. It's an evolution there, right?
For example, customers today might find it very useful to give instructions to a workflow authoring agent, perhaps to change the threshold for approving office material purchases. But in the future, this may be redundant, because other agents will automatically adjust those thresholds. He comments:
The agent is here, but do I really need this agent? Maybe it is one of the first agents I'm switching off. Because the real thing should be, Navi says, 'Hey, dear buyer, I would recommend that I put the threshold to 500 because your spend is so high this month. Let's put it to 500.' That's the other way. But this will take time.
This is where Coupa's extensive community data comes into the picture, he adds, because it provides a broad foundation for the vendor to ensure that agents are making sound recommendations.
AI agent pricing
Another area where customers need to build up confidence is in the pricing structure for these agent capabilities, and how quickly to invest in rolling them out. He comments:
I think where they are still struggling, though, is the model of pricing and if they can afford to start with a pilot, and how to touch and believe that this is possible, what we are promising? Well, obviously the answer is, just get started and we will help you. But yeah, that's the only barrier at the moment. Pricing models, the AI credit model... and when to start, with which team, and do I have time for trial-and-error, and so on. This is a barrier which is normally in everything which is new.
Traditionally, SaaS vendors have priced their software based on the number of users — a seat-based pricing model. But AI is increasingly being priced on usage or value — anything from raw processing cost to metrics such as number of documents processed or answers delivered. Coupa has adopted an AI credit system, where customers consume credits based on a measure of the value delivered. Lombardo explains:
You only pay if I deliver value to you. That's the idea. This is how we have set up this model, the AI credit model, because you pay after Navi has given you a recommendation. A recommendation can also be, obviously an answer to a question you had, but it can also be a recommendation on a contract summary, or a recommendation on changing a clause in the contract, or a population of a field just because you entered something.
Obviously those will have different amounts of credits. Populating a field is surely not as expensive as giving you a contract summary of a 200-page contract, right? So this is the distinguish of credits — one costs two credits, the other 200. These credits, when you use them, will be subtracted from your credit account. This is how the model, in a simplified way, [is] working. With a very transparent way, with a flexibility to switch on, switch off, agents.
With Coupa's ultimate aim being to establish a network of buyer and seller agents, I wondered whether in some circumstances, the vendor might look at charging a percentage of the transaction value, rather than a set number of AI credits. For example, there might be an agent that automatically discovers potential new suppliers or buyers and suggests a match between them. Where Coupa already negotiates a discount rate for its marketplace, is there a case for holding onto a percentage of that discount where it has helped both sides get together? It's a question that needs thinking through before the AI network is fully operational. He comments:
Do you put Coupa in the middle and say, 'Hey, if you have savings, then the buyer has something, the supplier, something. What about the trader in the middle?' If you think that's the Uber principle — I don't know how much, but Uber gets some percentage [of its fares]. So what about a small percentage for when a deal happens, so to say? That's something I'm thinking through before maybe this big AI thing kicks in.
Moving forward with any new pricing model nevertheless requires care and sensitivity to how customers react. He comments:
That's where I will have my highest attention on, really listening to customers intensely and finding out how is the acceptance model, because that's what ultimately counts.
My take
This was a candid interview about some of the behind-the-scenes thinking that is going into Coupa's agentic roadmap. It highlights the many different factors that enterprise application vendors need to think through as they bring their customers along into the agentic AI era — how to introduce new agent capabilities, how those agents and customer usage patterns are likely to evolve as time goes on, and how to navigate big changes in how vendors price their offerings. Coupa's roadmap is unusually well defined, but even so there are many uncertainties still to navigate in the next few years.