Governing AI Inside Enterprise Pricing And Discount Workflows
Eshaan Jain is a Senior Product Manager at Mphasis, focused on governance, security, and AI-driven automation for CPQ and CLM on Salesforce.
gettyI spent three years building the clause-extraction models behind Amazon’s $40 billion transportation procurement portfolio. They pulled pricing terms and renewal dates from scanned agreements with a level of accuracy the business trusted without a second read. The hard part was never getting the model to read a contract correctly. It was building an approval trail that let a finance leader sign off on the output with confidence.
That lesson carried straight into my next role as Lead Product Owner for Salesforce and Vlocity CPQ at T-Mobile (employed through Mphasis), where the systems in question don’t just read contracts. They set prices and approve discounts.
Once AI moves from reading contracts to influencing prices and approving discounts, governance becomes a business problem rather than a purely technical one. When an AI system can recommend or make a pricing decision, the critical question becomes where to draw the line between what the system can decide on its own and what still requires human approval.
Quote-to-cash is where AI governance is first tested because the output has a dollar figure attached.
A pricing engine that recommends a 12% discount, or an agent that routes an exception around a human approver, produces a business decision made without a human in the room signing off on it.
That gap between what these systems can now do and what companies have in place to supervise them is what enterprise AI leaders need to close this year, before it shows up on a customer’s invoice instead of in an internal review.
Board and audit-committee data back this up. Audit committee members naming AI governance a priority jumped to 35% in Deloitte and the Center for Audit Quality’s 2025 survey, up from 20% the year before, and the share of committees taking primary AI oversight rose from 14% to 20%.
At the board level, only 31% of directors now say AI is absent from the agenda, down from 45% in the prior edition. Still, 66% describe their board’s AI knowledge as limited to none.
Awareness is rising faster than expertise, and pricing and approval workflows are exactly where that gap shows up in dollar terms rather than in a slide deck.
The agentic layer exacerbates the exposure. Only 21% of organizations have a mature governance model for agentic AI, even as 74% expect to be running AI agents at meaningful scale within two years, per Deloitte’s 2026 “State of AI in the Enterprise” survey of 3,235 leaders across 24 countries. Eighty percent said they lack mature capabilities for the basics: audit trails, real-time monitoring and human-approval boundaries for what an agent can and can’t finalize on its own.
Those are the exact controls a discount-approval workflow needs before an agent gets write access to pricing. In a quote-to-cash system, that boundary usually comes down to one variable: the size of the discount or price exception relative to a threshold the revenue and compliance teams agreed on before the agent went live.
The incident count is rising alongside the deployment count. Stanford’s 2026 AI Index reports 362 incidents in the AI Incident Database in 2025, up from 233 the year before, a roughly 55% jump spanning privacy failures, biased outcomes and plain algorithmic errors.
A pricing or discount system sits closer to several of those categories than most people assume, because pricing decisions made at scale can produce disparate outcomes across customer segments even when no one designed them to. A lot of those incidents trace back to systems making decisions at a speed and volume nobody was reviewing in real time, which describes exactly how an automated discount engine operates by default.
The EU AI Act provides one example of where regulators are already imposing additional governance requirements on certain automated decisions affecting people’s financial circumstances, and that’s worth watching even for companies with no EU exposure today. The EU AI Act’s Annex III classifies AI systems used for creditworthiness evaluation and life and health insurance risk pricing as high-risk, alongside systems that affect the terms of work-related contracts. Deployers of those systems must complete a Fundamental Rights Impact Assessment before deployment, and penalties for violations reach 35 million euros or 7% of global turnover for the most serious breaches.
Quote-to-cash pricing sits outside that annex today. The underlying logic behind it, that automated pricing decisions affecting a customer’s financial terms deserve a documented review before they go live, is a standard likely to spread well past its current scope.
The fix is to build the review layer into the pricing workflow itself as a working part of the system, not as paperwork attached after deployment.
The CPQ platform work I’ve led cut contract-creation time by half just by process and platform enhancements. The approval logic mattered as much as the speed gain: which discount tiers an agent can finalize on its own, which ones route to a human, and what gets logged either way for every price change the system makes. Building that boundary in from the start is generally easier than retrofitting it after an incident.
Discipline and results move together. Gartner’s November 2025 research, covering 360 organizations with more than 250 employees, found that companies running regular AI system audits were more than three times as likely to report high value from generative AI. That correlation makes sense once you’ve sat through the alternative, a post-incident review where nobody can say which version of a pricing rule was live when a specific quote went out.
A review process built into the pricing system from day one produces both the audit trail and the value the business is chasing, because the two depend on the same underlying discipline.
For any enterprise running AI inside pricing, discounting or contract terms, the practical starting point is narrow.
Know which decisions an agent can finalize alone. Log every one it makes. Set a real human checkpoint for anything above a defined dollar or percentage threshold. That’s a smaller lift than a full governance program, and it’s the piece regulators, auditors and customers will ask about first.
Companies that build it now, while agentic pricing tools are still new enough to design around, will spend a lot less time retrofitting it once a bad decision has already gone out the door with a customer’s name on it.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
