As spending on AI grows, it’s no secret that enterprises are anxious to see returns on investments in the technology. Partners are on the hook to deliver something more than more tools with distant promises of productivity gains.
“The market has moved from rewarding AI ambition to demanding evidence of AI productivity,” Jim Piazza, chief AI officer at managed services provider Ensono, told Channel Dive.
The spend cycle has entered a new phase.
“In the first phase, rising capital expenditure signaled that a company was positioned for the next technology cycle,” said Piazza, who served as a director of operations at Facebook parent company Meta and global head of services at Penguin Services prior to joining Ensono in 2024. “Now that spending is flowing through depreciation, operating costs and cash flow, investors want to see a credible connection between each additional dollar invested and revenue, margin or competitive advantage.”
The numbers that matter include utilization, revenue generated per unit of compute, AI-related revenue growth, customer adoption and retention, gross margin after infrastructure and inference costs, and return on invested capital.
Piazza said customers should be measuring cycle-time reductions, labor capacity released, additional revenue and errors or losses avoided.
“At the end of the day, we have to impact existing business metrics,” he added.
Spending calculus
While top-line margin concerns are a cause for partner concern, they don’t signal defeat. The key is to clear a path to elusive ROI.
“Temporary margin pressure can be acceptable when utilization, backlog and customer value are strengthening,” Piazza said.
Piazza pointed to Microsoft as an example of the dynamic. The company acknowledged that its gross margin percentage on Microsoft Cloud was down year over year, driven in part by massive capital investments in AI infrastructure, during a July earnings call for the fourth quarter of its 2026 fiscal year. However, the margin erosion was partially offset by “ongoing efficiency gains,” Microsoft CFO Amy Hood said.
At the enterprise scale, that calculus can shift the focus from AI spending to outcomes.
As AI token spending rises, partners should tie costs to value.
“Value per token means the business benefit created by AI relative to its fully loaded cost,” Piazza said. “That benefit might include revenue gained, time saved, faster service, reduced fraud or fewer errors.”
Tokens are only part of the equation.
“Enterprises should measure the whole workflow — not just the model’s token charge — including integration, retries, human review and monitoring,” he said. “In practice, cost and value per successful task are often more useful than cost per token. Consumption is healthy when business value grows faster than total cost.”
Nevertheless, partners need to help customers control AI spending and share the responsibility for investments in infrastructure, applications and agents.
“Finance should set economic guardrails, technology teams should manage architecture and model selection, business leaders should own the expected outcome, and partners should help integrate and optimize the environment,” Piazza said.
The goal should be to manage spending on pilot projects and apply greater discipline as projects move into production.
“Organizations can protect a defined experimentation budget while requiring production projects to have an owner, baseline and measurable objective,” Piazza said.
Model choice
Partners are the trusted advisors who can pave the way to ROI, but success has less to do with selecting the right LLM than using the technology wisely.
“As models become more widely available, advantage increasingly comes from the data and business context surrounding them — and from the permission to act inside real workflows,” said Piazza.
The shift pushes the managed services provider opportunity further up the stack.
Customers need help connecting agents to enterprise systems, preparing and governing data, putting identity and approval controls around AI, redesigning workflows and managing multiple models.
“The greatest value will come from delivering secure, measurable business outcomes, rather than simply reselling AI licenses or compute,” Piazza said.
