DPW New York 2025: Put AI to work

Last week, we attended the DPW (Digital Procurement World) conference event in New York City. It was DPW’s second annual event in North America and was modeled on its signature Amsterdam event. The Brooklyn event was much larger than last year’s conference, and it better resembled the breadth and vibe of the Amsterdam event. There were multiple stages for presentations, dedicated areas for vendor booths and two days on the agenda instead of one.

The theme of this year’s conference was ‘Put AI to work,’ which was apparent not just in DPW’s messaging but also in many, if not most, of the sessions on the presentation stages. The most discussed topic at the conference was undoubtedly agentic AI, its present state and its future potential.

Many of the vendors in attendance (several of which were also presenters) already offer AI agents to customers with varying levels of sophistication. We were also happy to moderate several sessions on this topic and present our own session on Thursday morning about practical next steps to prepare for enterprise-wide adoption of AI agents, especially with regards to data and knowledge management.

Because we were preoccupied with our own sessions (e.g., we presented one titled ‘Agentic AI Meets SaaS: Teaching Old Elephants New Tricks’ that we received great feedback on) and vendor research duties, we do not have a recap of every presentation and all the content shared at DPW. Instead, below are some key takeaways drawn from both the various presentations and the conversations we had with procurement practitioners and vendors.

Agents, agents and more agents

As mentioned, AI agents were by far the most popular topic at the conference, and even though presentations were full of castigations about hype and buzzwords, there was a lot of terminology about ‘real’ agentic AI, ‘AI native,’ ‘agent first,’ etc. being thrown around without much precision beyond saying that they were not about bolting simple agents onto traditional SaaS apps. 

This is a false choice, and it is why we focused our presentation on how agentic AI systems are actually very complementary with next-generation enterprise SaaS architectures (e.g., building knowledge models that tie the language models that power the agents to the structured/narrow data models of fragmented SaaS apps) — if they are framed and adopted properly to work with each other and elevate each other.

In fact, the buzzword wars surrounding ‘agentic’ are so heated that we felt the need to cite Andrew Ng, the founder of the Google Brain and Coursera, who recently rightly stated that in today’s agentic hype cycle:  

“We would succeed better as a community if we just say that there are degrees to which something is agentic – so then if we just say if you want to build an agentic system with a little bit of autonomy or a lot of autonomy is all fine… No need to spend time arguing “is this truly an agent”. Let’s just call all of these things agentic systems with different degrees of autonomy and I think that actually hopefully reduce the amount of time people waste spent arguing… and then get on with it.”

Deja vu all over again, but also a bigger fork in the road to take

In the middle of the spectrum of upstart agentic AI startups and larger SaaS players, established providers such as ORO Labs and Zip took the stage to discuss their visions for the future of agent-enabled procurement software. But there was also a chance for practitioners and vendors to kick the tires on roughly a dozen present agentic AI startups. These start-ups, many of which we plan to cover more extensively in the coming months as they gain traction, use agents to address a variety of S2P and S2P-adjacent use cases.

Interestingly, actually building an AI agent for more narrow use cases and specific agent types, e.g., building an agent with a corporate Microsoft Copilot Studio license, is not extremely difficult. Any relatively computer-savvy person (no coding background required) can do so relatively quickly, even for strictly personal reasons (one presentation showed the n8n environment). What is more difficult is developing an agent that both addresses an impactful business use case and is extensible (able to scale to large environments, able to communicate with other applications and/or agents), configurable, scalable, secure, cost effective and future proofed.

As a result, the surge in agents on the market will only increase the importance of thorough evaluation processes by interested organizations. An agent may look capable in a demo, but there are several questions that still need to be asked based on the criteria cited above. We are currently developing our version of these for our clients beyond the current requirements we have on AI, Gen AI and agents in our current SolutionMap, so please reach out to us if you would like to compare notes. 

A somewhat similar hype phenomenon happened in the intake and orchestration market over the past year or so (albeit of a smaller scope, as agentic AI touches more than just the I&O segment). The popularity of the market has led to an influx of no-code, natural-language-oriented intake software with visually appealing user interfaces, easy configurability and little or no user training required. We have been stressing to clients, however, that this is only part of what makes intake and orchestration software valuable; what you do not always actually see with the product (the underlying tech platform/ecosystem, data modeling/management, integrations, etc.) and the company is even more important.

Seeing beyond agents to the broader tech stack and digital adoption

When it comes to agents, a similar mindset must be used during tech selection. Beyond the agents themselves, an open environment will also be attractive to organizations (especially large/global organizations). Many providers have or will have support for bring-your-own-agent, agent libraries (like a template library) and other features to ensure that customers do not have to solely rely on the agents the software provider developed. All of this is important for organizations to consider now. Even if some of these features seem unimportant, they could become essential in the future as their tech architectures and requirements evolve.

The hype around agents at DPW was understandable, which is why we want to emphasize the importance of initial investments (beyond just piloting/experimenting with free or cheap tools) as well as the work that needs to be done to ensure organizational unity with agent-enabled software. This is a team sport, and cross-functional collaboration is critical. While one keynote speaker unfortunately recommended throwing IT out of the room, this area is decidedly one that requires procurement and IT to collaborate with regarding architecture, investment, project alignment and governance. IT and Procurement must at least agree that ‘shadow procurement’ and ‘shadow IT’ are both bad.

Alignment with other departments as well as process and data standardization are crucial steps to ensuring successful adoption of any enterprise software adoption, and agents are no exception. As discussed on several panels at the conference, though, while all of this is essential, there still will never be a ‘perfect’ time for an organization to take the plunge and begin using the latest technology.

While organizations must do what they can with their data and processes prior to beginning massive implementations, they also must experiment in the near-term with said technology. Piloting and testing different agents also increases the chances of successful broader adoption in the future, due not just to familiarity with the tools but also to the ideas that can be generated by employees when using the tools for specific processes.

Besides a few issues related to the venue itself (namely the acoustics), we believe the DPW conference was quite a success. The event was very well-run and packed with presentations, and there were countless opportunities to share ideas with peers. It was also nice seeing senior practitioners mingling freely during the main event with each other and vendors rather than being cloistered away in secret executive conferences. Moreover, the focus on AI agents was very relevant to where the S2P market is already. We look forward to continued support of DPW and would highly recommend that firms of all stripes have some type of presence at future DPW events.

If you’d like to receive a copy of our presentation and discuss its insights, please don’t hesitate to contact us!