The Daily AI Executive
By Stephen Adegasoye
Executive Summary
- OpenAI's CFO has proposed retiring cost-per-token as the AI budgeting metric in favour of "useful intelligence per dollar" — a genuinely useful discipline for finance teams, but one built by the industry's most expensive model vendor, so treat the scorecard as a starting template, not gospel.
- Gartner now puts a number on the enterprise software disruption thesis: up to $234 billion of SaaS spending is exposed to "agentic arbitrage" by 2030, even as spending on AI models and platforms itself grows 63% this year — meaning the tech stack renewal conversation has fundamentally changed shape.
- A major streaming platform disclosed — inside a shareholder letter, not a press release — that generative AI touched roughly 300 titles in production this year, turning what was a content-ops pilot into a material, board-visible line item with real disclosure and rights implications.
By the Numbers
$64B (+63.4% YoY) Worldwide AI platforms & models spend, 2026 | $234B (~20% of SaaS spend) Enterprise SaaS spend at risk from agentic AI (by 2030) | ~300 Streaming titles using generative AI in production, 2026 YTD | ~$3B (roughly doubling YoY) Streaming platform ad revenue target, 2026 |
Vendor EconomicsOpenAI CFO Unveils "Useful Intelligence Per Dollar" Framework
What happened: OpenAI's CFO published a framework arguing that businesses should stop grading AI spend on cost-per-token and instead measure
the basic economic question facing CFOs and other business leaders is whether the value of the work AI completes grows faster than the cost of producing it, which requires looking more deeply than a metric such as cost per token
. The proposed scorecard rests on four questions:
how many customer issues did AI help resolve, how many code changes did it help ship, how many contracts did it review
, what each successful task actually costs once
employee time, human review, retries, and rework
are included, whether results are dependable, and whether value per dollar improves over time.
Why it matters: It reframes AI ROI around outcomes rather than usage volume, which is exactly the shift FP&A teams have been asking vendors for. But as Axios noted,
OpenAI has an incentive to steer customers toward measuring outcomes instead of sticker prices, since its models are among the most expensive
, and
OpenAI argues its most capable models will prove cheapest given the efficiencies associated with using top-tier intelligence
.
Who wins: Finance functions that build their own version of this scorecard, tailored to their own units of work, rather than adopting a vendor's definition of "success."
Who loses: Organisations still procuring AI purely on per-token pricing, and vendors whose value proposition depends on cheap tokens rather than reliable task completion.
Commercial implications: Procurement conversations shift from "what's the API rate card" to "what's the fully loaded cost per completed workflow, including rework."
Finance implications: FP&A needs a new unit of account for AI spend — cost per successful task, not cost per seat or per million tokens — and it needs to be independently defined, not vendor-supplied.
Media implications: For content and rights-driven businesses, "successful task" could mean cost per finished VFX shot, cost per localized episode, or cost per compliant ad creative — each with different quality bars and rework risk.
Long-term impact: Expect outcome-based AI pricing to become the default contract structure within 18–24 months, forcing renegotiation of existing seat- and token-based agreements.
Confidence: Medium **
Sources: OpenAI, Axios, Fortune
RegulationEU Orders Google to Open Android and Search Data to Rival AI Assistants
What happened: The European Commission adopted two binding Digital Markets Act decisions requiring Google to open Android integration points and share search data with competitors.
Currently, on Android phones, competitors' AI assistants only have restricted access to key functionalities of the Google Android operating system, without which alternative AI assistants are not competing on an equal footing with Google's own AI services that have full access.
The remedy means
users can activate their preferred AI assistant via voice commands, similar to the "Hey Google" command, and use third-party AI assistants to perform actions in apps on their behalf, such as booking a taxi or receiving suggestions for relevant replies in chat apps.
Search data sharing begins in January 2027, with Android changes required in the next major release by mid-2027.
Why it matters: This is the most direct regulatory intervention yet into how AI assistants reach consumers on the world's dominant mobile platform, and it previews how Brussels will treat other gatekeepers.
Who wins: Smaller AI assistant and search challengers gain a genuine distribution foothold in Europe for the first time.
Who loses: Google's default-placement advantage on Android narrows, and the case sets precedent that could reach other gatekeeper platforms, including Apple.
Commercial implications: Any content or advertising business that depends on default-assistant or search-driven discovery in Europe should plan for a more fragmented, multi-assistant distribution landscape from 2027.
Finance implications: Media planning and ad-tech budgets built around a single dominant discovery channel need contingency modelling for a more competitive, commission-fragmented environment.
Media implications: Content discoverability, GEO (generative engine optimisation) strategy, and audience-data partnerships all become more complex as multiple AI assistants compete for the same attention layer.
Long-term impact: Expect a multi-year compliance and appeals process, but the direction of travel — mandated interoperability for AI assistants — is now established policy in the EU's largest digital market.
Confidence: High **
Sources: European Commission (Digital Markets Act), Unite.AI, TechTimes
Media EconomicsStreaming Platform Discloses Generative AI Use Across ~300 Titles
What happened: A major streaming service told shareholders in its Q2 earnings letter that
the technology's usage expands across every level of a program's production process, from its concept and pre-visualization to post-production and release
, naming specific titles where
the technology helped create "highly complex sequences" that included enhanced crowd sizes and battle sequences
. The company stated:
"We are increasingly leveraging these tools to deliver higher-quality output more quickly and at a lower cost than traditional methods. In some cases, productions would have had to leave out key shots and sequences in the absence of GenAI technology."
The disclosure landed alongside
Q2 2026 revenue of $12.56 billion, up 13% year over year, with EPS of $0.80
, and the company confirmed it
is on track to roughly double advertising revenue to about $3 billion in 2026
.
Why it matters: This is one of the first times a major content owner has quantified generative AI's footprint across its production slate to shareholders rather than in a press release, signalling that AI production cost savings are now material enough to disclose.
Who wins: Content businesses that get ahead of disclosure norms and build cost-savings evidence into investor communications; production vendors offering traceable, rights-compliant AI tooling.
Who loses: Productions and vendors unable to demonstrate provenance and consent for AI-generated assets face rising legal, union, and reputational exposure.
Commercial implications: AI-assisted production is shifting from an R&D pilot line to a recurring COGS reduction lever that finance teams should be forecasting explicitly, title by title.
Finance implications: Production budgeting and greenlighting processes need a standard way to quantify AI-driven cost avoidance (e.g., shots that would otherwise have been cut) alongside traditional above/below-the-line cost categories.
Media implications: Disclosure practices for AI use in content are becoming a de facto industry standard, ahead of formal regulation, raising the bar for consent, credit, and rights tracking across every production.
Long-term impact: Expect investor and analyst questions about "AI-assisted content ratio" to become as routine as questions about content spend efficiency within two to three reporting cycles.
Confidence: High **
Sources: Variety, AI Weekly
Deep Dive: Why "Cost Per Successful Task" Is About to Replace Cost-Per-Token
For two years, the default way to budget for AI has been simple and wrong: multiply expected token volume by the vendor's price per million tokens. That metric feels precise because it's quantifiable, but it answers the wrong question. A cheap model that needs three retries and a human rewrite to get a usable answer can cost more, in fully loaded terms, than an expensive model that gets it right first time.
OpenAI's CFO put a name to the fix this week:
what matters is the full cost of producing a successful outcome, measured against the value that outcome creates
. The logic, stripped of vendor self-interest, is genuinely first-principles:
| Question | What it replaces | What finance should track |
|---|
| Is the AI completing work that matters? | Usage/adoption metrics (logins, queries) | Number of tasks actually completed to spec | | What does each successful task cost? | Cost-per-token | Full cost ÷ number of successful outcomes, including |
employee time, human review, retries, and rework
|
| Can people depend on the result? | Model benchmark scores | Real-world error/rework rate by workflow |
|---|
| Does value grow faster than cost over time? | Flat renewal budgeting | Trend line of cost-per-successful-task quarter over quarter |
The reason this matters beyond one company's blog post is that Gartner's own research, independently, is describing the same shift from the buyer side. Gartner's forecast of $64 billion in 2026 AI platform and model spending noted that
enterprise AI budgets are coming under greater scrutiny, with increased focus on usage efficiency, cost control and measurable outcomes, and spending is shifting toward providers who can demonstrate clear value across cost, latency, performance and reliability
. And Gartner's separate $234 billion "agentic arbitrage" report makes the mechanism explicit:
agentic systems deliver outcomes directly, bypassing traditional user experience-heavy applications and making the software invisible, breaking the link between user growth and revenue growth for many enterprise software vendors
.
Put simply: buyer-side analysts and seller-side vendors are converging on the same conclusion from opposite directions. That convergence is the strongest signal yet that outcome-based AI budgeting is not a fad — it's about to become the default unit of account for AI spend everywhere, including inside media and content organisations that have historically budgeted technology spend by seat count and licence tier.
The catch every finance leader should hold onto: whoever defines "successful task" controls the scorecard. If that definition comes only from the vendor, the vendor wins the argument before the negotiation starts.
Commercial Finance Implications
Three opportunities
1. Renegotiate SaaS and media-tech vendor contracts now, ahead of 2027 renewal cycles, toward outcome-based or consumption pricing rather than seat-based licences — Gartner's own analysis frames this as a multi-year structural shift, not a one-off discount.
2. Build an internal "cost per completed unit" scorecard for AI-touched production and post-production workflows (localization, VFX, ad creative variants) to quantify genuine cost avoidance, not just headline AI adoption.
3. Use the current SaaS disruption window to consolidate fragmented point-solution licences into fewer, outcome-accountable platform contracts before vendors reprice around agentic capability.
Three risks
1. Vendor-defined ROI metrics (including scorecards published by model providers themselves) are structurally biased toward their own pricing model — adopt the framework, but define "success" independently.
2. Regulatory shifts in platform interoperability change distribution and discovery economics for any content or advertising business reliant on default placements — budget scenarios should not assume today's channel mix persists past 2027.
3. Under-disclosed or inconsistently governed AI use in content production is now a material rights, consent, and disclosure risk — treat it as a financial control gap, not just a legal one.
Three ideas to explore
1. Stand up a joint Finance–Legal–Production governance forum to track where generative AI touches content, ensuring cost savings are captured and rights/consent obligations are met before the next reporting cycle.
2. Pilot a "useful work per dollar" dashboard on one high-volume workflow (subtitling, ad trafficking, financial close automation) this quarter, using the four-question framework as a template but your own unit definitions.
3. Insert a data/knowledge-ownership clause into every AI vendor renewal — Gartner frames this bluntly as the question of
"who owns what the system learns from you," warning that enterprises risk a new form of vendor lock-in if operational learning remains with software providers rather than the customer
.
Executive Talking Points
1. Cost-per-token is being retired as a budgeting metric; the industry, on both buy- and sell-side, is converging on cost-per-completed-task as the new standard.
2. Up to $234 billion of enterprise software spend is exposed to agentic disruption by 2030 — a meaningful share of every technology renewal negotiation is now in play.
3. Generative AI in content production has moved from pilot to shareholder-letter disclosure at material scale — this is now a COGS and governance conversation, not an innovation-lab one.
4. Regulatory intervention in AI distribution (EU Digital Markets Act) is reshaping how content and advertising reach audiences on the dominant mobile platform — plan distribution economics for a multi-assistant world.
5. The most valuable clause in the next generation of AI vendor contracts isn't price — it's who owns the operational knowledge the system accumulates about your business.
AI Tool of the Day
AskGartner AI — Gartner's proprietary AI research assistant, built into existing Gartner client subscriptions, letting enterprise leaders query Gartner's analyst research and benchmarking data directly rather than searching reports manually. Who it's for: CIOs, CFOs and strategy teams already subscribed to Gartner research who need fast, sourced answers on vendor benchmarking and AI ROI questions. Pricing: bundled into existing Gartner subscription tiers; no separate published price. Why it matters: as vendor-supplied ROI scorecards proliferate, having an independent, analyst-backed benchmarking source becomes more valuable, not less. Should a finance leader learn it: yes, if your organisation already holds a Gartner licence — it's a low-effort way to pressure-test vendor claims. Time required: roughly 30 minutes to explore core query patterns. ROI: potentially significant analyst-hours saved on vendor and market benchmarking work.
AI Paper / Report of the Day
Gartner: "Worldwide AI Platforms and Models Market to Grow 63% in 2026" — Problem: enterprises lack a clear read on where AI model and platform spend is actually heading amid volatile pricing and rapid vendor churn. Method: Gartner's worldwide end-user spending forecast across the AI models and platforms market segment. Findings:
worldwide end-user spending on AI models and platforms is projected to total $64 billion in 2026, up 63.4% from $39 billion in 2025, with spending on GenAI models forecast to grow 117%, while AI platform spending will rise 36.9%
. Why executives should care: the report explicitly links future vendor winners to
vendors that help enterprises manage where and how AI is used across the business, as more models enter the market and usage-based pricing becomes harder to predict
— a direct signal to consolidate spend with fewer, more transparent providers.
Build Something
Exercise (25 minutes): Draft a one-page "Useful Work Per Dollar" scorecard for a single AI-touched workflow your organisation already runs — subtitling, ad-creative production, financial close, or contract review. Using the four-question framework (task definition, full cost, dependability, value trend), write down your own definition of "successful task completion" for that workflow before you next speak to a vendor about it. Why it matters: whoever defines success controls the negotiation — this exercise ensures it's you, not the vendor.
Skill of the Day
Model routing — the discipline of directing different tasks to different models based on cost-per-successful-outcome rather than defaulting to the most capable (and expensive) model for everything. Difficulty: Medium. Time to learn: 2–3 hours for core concepts, ongoing refinement in practice. Best resource: start with OpenAI's own "Useful Intelligence per Dollar" framework and Gartner's AI Platforms and Models forecast as reference points, then map your own workflows against them.
Executive Quote
"The basic economic question facing CFOs and other business leaders is whether the value of the work AI completes grows faster than the cost of producing it."
— Sarah Friar, CFO, OpenAI
Sources
What You Should Do Today
1. (15 minutes) Pick one AI vendor renewal due in the next six months and check whether the contract is priced by seat/token or by outcome — flag it for renegotiation if it's the former.
2. (20 minutes) Draft the four questions from today's Deep Dive against one live AI workflow in your organisation, and identify who currently owns the definition of "success" for that workflow.
3. (25 minutes) Ask your production, marketing, or content operations lead for a rough tally of where generative AI is already used in your output — before an external disclosure requirement forces the conversation.
|