Two-Thirds of Boards Demand AI ROI Proof as Vendors Rethink Pricing
Only 35% of finance teams can measure AI returns, two surveys show. Snowflake, Zendesk, Atomicwork and Rackspace are rewriting pricing to prove AI pays for itself.
By Amara Osei
5 min read

What's News
- CloudZero survey of 260 finance executives: two-thirds of boards require proof of returns before more AI spending; Protiviti finds only 35% of finance organizations effective at measuring AI returns.
- ServiceNow says internal AI generates over $1 billion in annualized value; Salesforce reports 7 billion agentic work units.
- Infosys AI revenue rose 29% quarter over quarter to 10.5% of sales, yet it cut its full-year growth outlook to 1%-2%.
- Snowflake cut one customer's AI workload costs by more than half by shifting routine tasks to smaller open-weight models.
- McKinsey survey: 80% report personal productivity gains from AI, but only 37% say AI contributed to EBIT.
Only 35% of finance organizations consider themselves effective at measuring AI returns, while two-thirds of surveyed executives say their boards require proof of returns before approving further AI investment, according to surveys by Protiviti and CloudZero. The gap between those numbers is now reshaping how AI vendors price, measure and renew their contracts.
The industry's preferred metrics offer buyers little certainty. Salesforce reports its technology has performed 7 billion agentic work units—activities such as invoking a tool or completing a reasoning chain—and a single business outcome can require many units. ServiceNow told investors on October 28 that its internal AI use generates more than $1 billion in annualized value, double its estimate at the end of 2025. More automated activity can mean useful work, or repeated attempts at the same task.
Where AI efficiency cuts the vendor's own revenue
The tension is already hitting sales. At Infosys, AI revenue rose 29% quarter over quarter and reached 10.5% of sales, yet the company cut its full-year growth outlook to 1%-2% amid pricing pressure and demands for greater productivity. When AI lets a client buy the same service for less, the client's case for investing strengthens while the supplier's contract value shrinks. New AI business must replace the revenue lost as existing work gets cheaper.
Why is Snowflake telling customers to spend less?
Snowflake CEO Sridhar Ramaswamy argues rising AI consumption is worthwhile only when it replaces existing costs. Snowflake reported $1.49 billion in product revenue for its second fiscal quarter of 2027, up 37% year over year, with 126% net revenue retention.
"Rather than celebrating consumption, I look for shrinking backlogs, faster ship cycles, and teams deploying previously impossible capabilities," Ramaswamy said. He points to a renewal with an enterprise customer where Snowflake's agents cut a two-week inventory reconciliation process handled by six analysts down to two hours—and the customer's finance team still resisted renewing.
"Our team was celebrating query volume and developer throughput, but the CFO did not care about consumption metrics," he said. The deal closed only after Snowflake showed the technology could prevent millions in obsolete product markdowns, retire two legacy BI tools, and eliminate store-manager escalations. "Finance buys hard cost displacement and operational certainty."
At one financial services company, Snowflake recommended routing routine support tickets and document extraction away from expensive frontier models to smaller open-weight models via its Cortex AI Gateway. The switch cut workload costs by more than half without sacrificing accuracy. "If a cheaper model gets the exact same job done with zero loss in fidelity, running the most expensive model on the market is just bad business," Ramaswamy said.
What counts as a paid outcome?
Zendesk CEO Tom Eggemeier sees outcome-based pricing as a way to tie spending to completed work. The company reports nearly 20,000 AI customers, AI annual recurring revenue above $200 million in 2025, and expects that to exceed $400 million in 2026. Zendesk uses a separate AI model to evaluate automated resolutions, and a closed ticket alone doesn't qualify for payment.
"Service has outgrown the ticket. A customer's issue can span channels and multiple conversations, so an outcome should be tied to eligible work on the issue, not counted once for every message or contact," Eggemeier said. The contracted price stays the same whether AI resolves the issue or a person helps; an AI-assisted handoff doesn't trigger a second charge. Zendesk's Outcome AI Platform Plans remain in early access.
Atomicwork takes a stricter line. "A closed ticket is an administrative event. An employee getting back to work is the outcome. If AI answers in 10 seconds but leaves someone with 30 minutes of troubleshooting, we've moved the work onto the employee," said cofounder and CEO Vijay Rayapati. Atomicwork charges $1 to $3 per completed outcome with annual minimums, plus a $25,000 annual platform tier, and guarantees ServiceNow migrants a 50% reduction in total cost of ownership. "If we're going to guarantee savings, customers should be able to hold us to the numbers," Rayapati said.
What about the humans keeping AI running?
Rackspace CEO Gajen Kandiah splits engineering time into "build" and "babysit," aiming to shrink the latter. "An agent that looks brilliant while an engineer quietly fixes its output isn't doing the work. The engineer is," he said. At a large U.S. healthcare provider, engineers spent most early deployment time improving source data; as data improved, each update cycle needed less support. Rackspace sometimes recommends less AI than customers request—one company wanted LLMs to run an entire master data management system, and Rackspace steered it to traditional software for routine processing.
"We have to be willing to help a customer spend less with us. It affects our revenue, and I do not assume every dollar comes back through a new AI engagement," Kandiah said. In July, Rackspace cut its annual revenue outlook by $150 million; Q3 results then prompted a $35 million increase in the full-year midpoint. It announced AI infrastructure agreements with Together AI and Blacksmith worth roughly $610 million in total contract value. Quarterly revenue totaled $609 million, down 9%, with a $54 million net loss and capital spending up to $104 million from $34 million a year earlier.
Does productivity actually reach the P&L?
McKinsey's 2026 State of AI survey of 1,719 respondents across 97 countries found 80% said AI improved their personal productivity, but only 37% reported AI contributing to EBIT—essentially unchanged from 2025—and about 6% attributed at least 5% of EBIT to AI. Reltio CPO Ansh Kanwar recommends measuring identical AI tasks on fragmented versus unified data to isolate what the AI itself delivers. Anthropic's Kate Jensen, whose S-1 reports $17 billion in Q3 revenue and about 900% net revenue retention excluding consumer plans, says companies should fix a baseline before deploying.
"As Claude takes on bigger projects, every leader should be able to say what got done, how well it was done, and where they would have been without the help of Claude," Jensen said. If an unmet business target gets replaced by a flattering usage metric at renewal, each contract risks becoming another paid experiment.
Original: salesforce.com
More from Amara Osei
Show full bio
Senior reporter covering companies, deals and strategy in entrepreneurship, startups and venture funding at Business Bearings.
2,422 articles