QoE for AI Revenue: The Diligence Checks Buyers Miss in US Tech Deals
Venture funding to AI reached USD 212 billion in 2025, up 85% year over year, and many AI companies are posting triple-digit growth on "AI ARR" that may not behave like recurring revenue at all. The headline number and the durable number are increasingly two different things — and QoE for AI revenue is where that gap gets tested.
Hewlett-Packard paid over USD 11 billion for Autonomy in 2011. A year later, it wrote down USD 8.8 billion, much of it tied to how Autonomy had recognised and presented its revenue. The deal failed on revenue quality, not strategy. That lesson is resurfacing in a new form. Venture funding to AI reached USD 212 billion in 2025, up 85% year over year, according to Crunchbase, and many AI companies are posting triple-digit growth on "AI ARR" that may not behave like recurring revenue at all. This is where QoE for AI revenue earns its keep, because the headline number and the durable number are increasingly two different things.
Why AI Revenue Breaks the Old ARR Playbook
For two decades, SaaS valuation rested on a simple assumption: annual recurring revenue was contracted, sticky, and predictable, so a multiple on ARR was a reasonable proxy for enterprise value. AI-native businesses are dismantling that assumption. Usage-based pricing has moved from niche to default, with 85% of surveyed SaaS companies already using or adopting it and 77% of the largest software firms embedding consumption pricing, per Metronome and Greyhound Capital's 2025 survey. When revenue is metered, credit-funded, and tied to experimental projects, the line between recurring and one-time income blurs. A buyer who underwrites the reported ARR multiple without testing what sits beneath it is buying HP-Autonomy risk in a new wrapper.
Check 1: Separate Experimental Revenue from Recurring Revenue
The most expensive mistake in AI deals is treating experimental spend as recurring income. Investor Jamin Ball labelled this "experimental run-rate revenue" in 2024, and Primary Venture Partners' Cassie Young has warned of an impending "gross retention apocalypse" as AI pilots fail to renew. The data supports the concern: AI-native companies showed median gross retention of 40% and net retention of 48% in 2025, against a B2B SaaS median net retention of 82%, according to ChartMogul's SaaS Retention Report (2025). Much of that gap is experimental budget. Customers run a model on proprietary data for a quarter, then stop. RAND Corporation (2024) found that more than 80% of AI projects fail to reach meaningful production, twice the rate of non-AI IT projects, which means a large share of early AI revenue never converts to a production workload. What this means: reclassify revenue by commitment type, multi-year contracted, budgeted-and-renewing, or pilot, before applying any multiple.
Check 2: Stress-Test the Revenue Recognition Policy
Usage-based pricing complicates revenue recognition in ways a standard audit can wave through. Under ASC 606, variable consideration and performance obligations require judgment, and judgment is where presentation drifts from economics. KPMG's Software and SaaS Revenue Handbook (December 2025) flags contract modifications and evolving consumption models as a persistent source of recognition questions for the sector. In AI deals, the risks are concentrated: overage and consumption fees booked as committed ARR, multi-year deals recognised on optimistic ramp assumptions, and proof-of-concept consumption recognised before a customer is contractually committed. What this means: reconcile the revenue recognition memo against actual contract terms and billing data, not the policy statement at face value.
Check 3: Strip Out Model-Led Discounts and Free Credits
AI go-to-market is built on free credits and aggressive promotional pricing, which inflates both usage and reported revenue. As Oxx general partner Mikael Johnsson told PitchBook (December 2025), AI-native ARR is frequently a mix of one-off, credits-based, performance-based, and outcome-based contracts rather than clean subscriptions. FPV Ventures' Nikunj Kothari titled his December 2025 critique of these practices "Liar's Valuation." The QoE task is to rebuild revenue on a gross-to-net basis, stripping promotional credits, discount ramps, and outcome-based clawbacks to reveal the true run-rate at standard pricing. What this means: the number that matters is net revenue at sustainable pricing, not gross consumption propped up by free credits that expire after the next funding round.
Check 4: Reconcile Cash Against Reported Growth
The oldest QoE discipline is the most useful here. Operating cash flow should track reported earnings over time, and a persistent gap is a warning. Many AI companies present optimistic, back-loaded contracts as today's recurring revenue, making growth look stronger than the cash collected. What this means: build a bridge from reported ARR to billed revenue to cash collected. Where the three diverge sharply, the recurring story is weaker than the deck claims.
Four Red Flags in AI Revenue Quality
• Short opt-out windows with high exit rates. Contracts with three-month opt-outs where most customers leave are momentum, not recurring revenue. Case in point: Jasper, valued at USD 1.5 billion, cut its 2023 ARR forecast by at least 30% and lowered its internal valuation by about 20% once ChatGPT gave customers a low-cost substitute (The Information, 2023). • ARR built on credits and pilots. If a material share of revenue comes from promotional credits or proof-of-concept budgets, the run-rate is overstated. Case in point: Builder.ai, once valued at USD 1.5 billion and backed by Microsoft, was alleged to have inflated revenue through improper billing practices, including invoice exchanges with a partner firm, and filed for bankruptcy in May 2025 (Bloomberg; Financial Times, 2025). • Recognition policy changes timed to a raise. A policy that shifted shortly before a fundraise deserves forensic attention. Case in point: Outcome Health recognised revenue on under-delivered advertising campaigns, overstated its 2015 and 2016 revenue, and used the inflated audited financials to raise USD 487.5 million in early 2017. Three executives were convicted of fraud in 2023 (US Department of Justice, 2023). • High ARR, low cash conversion. Reported growth that does not convert to cash signals back-loaded or experimental revenue dressed as recurring income. Case in point: WeWork reported rapid revenue growth while burning billions in cash, leaning on a self-defined "community-adjusted EBITDA" metric. Its 2019 IPO collapsed once investors focused on the cash reality.
How MARC Adds Value
MARC applies institutional-grade Quality of Earnings analysis to the specific economics of AI and usage-based businesses, separating durable revenue from experimental spend before a buyer commits capital. • QoE-level depth on revenue quality. We reclassify revenue by commitment type and rebuild it on a gross-to-net, sustainable-pricing basis. • Independent validation. A third-party view that tests the recognition policy against contracts and billing data, free of management bias. • Working capital and cash flow diagnostics. We reconcile reported ARR to billed revenue and cash collected to expose the growth that is real. • Fast turnaround without sacrificing depth. Built for deal timelines where AI valuations move quickly and rigour cannot be the bottleneck. In a market where AI multiples are set on metrics that are getting harder to trust, that discipline is the difference between a defensible price and a write-down.
FAQs
What is QoE for AI revenue? A Quality of Earnings analysis adapted to AI and usage-based businesses. It tests whether reported ARR reflects durable, recurring income or a mix of experimental spend, credits, and one-time consumption, establishing the true run-rate revenue a buyer is actually acquiring. Why is AI ARR considered unreliable? AI ARR often blends pilots, free credits, and usage-based consumption that does not renew. AI-native companies showed median gross retention of 40% in 2025 (ChartMogul), well below the SaaS norm, so reported ARR can overstate committed, recurring revenue. How does usage-based pricing complicate due diligence? It introduces variable consideration and performance-obligation judgment under ASC 606. Overage fees, credits, and back-loaded ramps can be recognised in ways that flatter growth, so diligence teams must reconcile the recognition policy against contracts, billing data, and cash collections. What are the biggest red flags in AI revenue quality? Short opt-out contracts with high exit rates, revenue built on promotional credits and pilots, customer concentration above 40%, recognition policy changes timed to a fundraise, and high reported ARR with weak cash conversion.
Key Takeaway
The future of tech M&A belongs to buyers who underwrite revenue durability, not revenue velocity. As consumption pricing becomes universal, the firms that interrogate what sits beneath reported ARR will price deals correctly and protect their downside. Those that keep paying a clean SaaS multiple for experimental revenue will repeat the Autonomy lesson at AI speed. Revenue quality, not growth speed, is now the deciding variable. Let MARC power your next diligence move.
About MARC Analytics Team
Our research team comprises experienced financial analysts and consultants with over 50+ years of combined experience.
View all articles →Test What Sits Beneath the ARR
Let MARC power your next diligence move in AI and usage-based tech deals.
