DMSi, a 350-person ERP vendor, is still working out agentic pricing. It’s building one controlled access point, planning usage thresholds, and using Workato to create a path to monetization.
Over three days in Las Vegas, World of Workato (WOW) 2026 delivered 50+ sessions and 100+ speakers. One fireside discussion, Monetizing the Agentic Tier: Cost, Pricing, and Going to Market with AI, drew question after question from a builder audience.
Alexandra LaPorta, Senior Embedded Account Executive at Workato, sat down with Jason Niemi, Chief Product Officer at DMSi, and Preston Badeer, Chief Strategy Officer at Contemporary Analysis and DMSi’s longtime AI partner. They didn’t offer a pricing formula. Instead, they described a sequence: control the plumbing, drive down the cost, then monetize. Here’s how it works.
Building AI agents for every customer isn’t realistic for a 350-person ERP vendor
DMSi builds ERP software for the building materials industry and recently marked 50 years as an independent software company. It serves roughly 650 customers and 40,000 users, from five-user lumber yards to companies doing up to $30 billion a year with 7,000 users on the system.
Across that range, every customer runs differently, down to details as specific as the color of a door.
“If we want to build an agent for somebody that’s going to make their day better, it’s going to be different for every single one of them.”
— Jason Niemi, Chief Product Officer, DMSi
Niemi’s approach is to focus on what DMSi does best. “We can’t do it all,” he said. “And we’re really good at what we do in the ERP and that’s where we need to live.” DMSi builds the foundation and uses Workato to give customers the surrounding ecosystem. One example of this approach in practice is DMSi’s rebuild of an existing integration on Workato. The payoff was clear and concrete: the rebuild cost less than the previous version while giving the company a better foundation.
Pricing isn’t solved, but DMSi won’t get burned
Niemi was direct about where the industry stands: “I don’t think anybody has really figured out the pricing.” Most DMSi applications aren’t priced on usage today, and he says that model won’t be right in 2027 and 2028, when every one of these features carries real usage. His concern is control:
“How do we put up thresholds? How do we make sure that we really understand what’s going on in our entire ecosystem?”
— Jason Niemi, Chief Product Officer, DMSi
For DMSi, working out pricing starts with understanding usage across the ecosystem. Its answer is architecture, not a price sheet: a middleware layer, or AIM, from Badeer’s team, plus a plan in three parts.
- One access point: Everything that touches DMSi’s systems, whether an API call or an agent, runs through a single access point, so DMSi can see model consumption at any moment.
- Thresholds: Niemi expects to introduce included usage with a monthly allowance and the option to buy more.
- Flexible pricing: Niemi expects hybrid, possibly outcome-based models. He’s tried to redo DMSi’s pricing about ten times, and each attempt runs into the range of customer sizes: “This customer is going to go up by 300x. And then our biggest customer went down by half.” Enterprise pricing across the board may be where it lands.
Start on the expensive side, then adjust
Understanding usage is one part of the cost picture. Badeer’s advice on model costs follows a different sequence: first learn whether an agent is useful, then optimize it.
“If you don’t know if the agent’s useful yet… then don’t worry about cost yet. It doesn’t matter. Use a frontier model and get something into production and learn from it.”
— Preston Badeer, Chief Strategy Officer, Contemporary Analysis
Niemi said DMSi starts each project on a frontier model like Opus, then works down through Sonnet, Haiku, and open-source Qwen, looking for the smallest model that still works. Once an agent is in production, Badeer recommends:
- Catch doom loops. “Doom loops are when the agent gets stuck and can’t proceed.” A hard cap on retries is simple to build.
- Try open models. “We commonly see a 10x cost reduction just moving from commercial to open even if you’re not shrinking the model size.”
- Right-size with evals. Smaller models need more framework and prompt work. Find the sweet spot.
- Treat tool definitions as prompt engineering. “There’s a lot of juice to squeeze there.”
For some customers, choosing a model also means deciding where their data can go. For customers who can’t send data to commercial providers, Badeer pointed to Workhorse, his company’s private-cloud offering. By his most recent check, he estimated Workhorse was roughly 87% cheaper than per-token pricing for comparable models.
Model size isn’t the only lever Badeer recommends considering. He also made the case for older tools. A new decision model called Jev can only respond with a choice from a set you give it, like tagging an email high or low priority. To him, that’s classic machine learning, and “random forests are still underrated.”
An agent in production, with early signs of trust
An order-entry agent built for how orders really arrive
The cost and model choices matter because agents need to work in real customer workflows. DMSi’s clearest example so far is its AI sales agent, which went live the week of the conference. It accepts orders in whatever form customers send them, whether text, photo, PDF, or email. Then it does the work a person would: checks past orders, relationships, and stock; drafts the order; and holds it for review.
“It’s not that they can’t just type it out, it’s not that we’re just scanning the image. It’s that the agent’s able to take a bunch of actions to go research all these additional pieces of context, draft the order, and allow them to safely review it, edit it, and approve it when they’re happy with it.”
— Preston Badeer, Chief Strategy Officer, Contemporary Analysis
That flexibility matters because real orders can arrive in unexpected forms. In one example DMSi collected, a customer had written an order on a board with a Sharpie, photographed it, saved the photo as a PDF, and emailed it in.
Those unusual inputs also give DMSi real cases to learn from. The company wants to hear when things fail: when customer service passed along complaints, Niemi asked for the documents and use cases behind them. He sees a customer saying “this didn’t work” as valuable feedback, calling it “a successful closed beta.”
Trust is another early signal. Badeer said customers who had started building their own agents told his team they’d rather DMSi own it: “We would much rather have DMSi own and run and build that agent. We trust them.”
What’s next for DMSi and Workato
The order-entry agent is one part of DMSi’s broader product roadmap. Workato is also being implemented across CRM, DocStorage, and Shopify integrations, with delivery partner DynPro. DMSi is exploring ITSM automation, recipe lifecycle management (RLCM), AIRO, XChange, and the multi-tenant structure. A DocStorage release is planned this year, and an EDI offering is on the longer-term horizon.
That roadmap includes features DMSi may treat as part of the platform, as well as add-ons it could monetize. Some of the document management work will be table stakes. “We’re planning to just eat the consumption of those,” Niemi said, with thresholds on top. The upside comes from add-ons built on workflows like accounts payable invoice validation.
“We’ll have Workato in every user’s hands, whether they know it or not, they’ll be using that and we want to monetize it.”
— Jason Niemi, Chief Product Officer, DMSi
The balance between moving quickly for one customer and building for many came up again at an AI roundtable during the conference. Some of DMSi’s more forward-leaning customers shared apps they’d built on the side and encouraged the company to move faster. Niemi’s takeaway: building something for one customer is different from building, securing, and deploying it for hundreds. LaPorta noted that Workato could have built all of those apps, and Niemi agreed.
The takeaway
Nobody has agentic pricing figured out. But DMSi’s experience suggests teams can still prepare: control access, watch consumption, and right-size models while they learn what customers need and what it costs to deliver. That groundwork can help product leaders decide what to include, what to charge for, and how an orchestration layer can support monetization.
Become the platform your customers, and their agents, can’t live without.
Explore Workato ONE for SaaS platforms.
