Who Owns This AI Spend? The Attribution Gap Hiding Inside Every AI Strategy
Why the next AI conversation in the boardroom will be about accountability, not capability, and why governance and cost can no longer live in separate tools.
Picture a quarterly business review at a large Gulf enterprise. The Group CFO has a single question for the technology team: "We spent materially more on AI this quarter than we budgeted. Which business units drove it, which systems, and what did we get for it?"
The CIO has the provider invoices. The Chief AI Officer has a register of approved use cases and their risk ratings. Neither can join the two, and a large slice of the bill sits on shared API keys that nobody owns.
The scene is a composite; the numbers behind it are not.
The numbers behind the discomfort
AI spend has moved from innovation budgets into the operating cost base in two years. In the FinOps Foundation's State of FinOps 2026, 98% of respondents now manage AI spend, up from 31% two years earlier, and managing AI value is the top priority and the most sought-after skill for FinOps teams [1].
Managing is not the same as understanding. Gartner warns that organisations that don't understand how their generative AI costs scale could make a 500% to 1,000% error in their cost calculations. It also reported that 72% of CIOs say their organisations are breaking even or losing money on AI investments [2]. CloudZero's State of AI Costs found that only 51% of organisations can confidently evaluate the ROI of their AI spend [3].
The consequences show up as abandoned programmes. Gartner predicted that at least 30% of GenAI projects would be abandoned after proof of concept by the end of 2025, citing escalating costs and unclear business value alongside data quality and risk controls [4]. It also expects over 40% of agentic AI projects to be cancelled by the end of 2027 for similar reasons [5]. MIT's Project NANDA study went further: despite US$30–40 billion of enterprise investment, 95% of organisations were seeing no measurable P&L return from generative AI [6].
Behind many of these failures sits a simpler problem than model quality. Nobody can say, with evidence, who spent what, on which AI system, for which outcome.
Why AI spend resists attribution
Cloud FinOps took a decade to reach decent allocation through tags and account ownership. AI breaks those habits in four ways.
- Consumption is shared by design. Model access is centralised behind a platform team, a gateway or a few enterprise API keys. That is good architecture, but the invoice shows one line owned by IT while demand comes from dozens of business units.
- The unit of cost is invisible to the business. Input, cached, output and reasoning tokens are each priced differently, and the same question can cost ten times more depending on prompt design or model choice. Finance sees a total; the business sees an answer.
- Agents multiply cost silently. One agentic request can call several models, tools and other agents. Logged per API call rather than per business transaction, the true cost of a process disappears.
- AI hides inside other budgets. Copilot-style seats, AI features bundled into SaaS contracts, capacity commitments and vendor credits sit in different ledgers.
- The result is a growing "unallocated" pool. Finance either spreads it across departments, which destroys accountability, or leaves it with IT, which makes AI look like overhead rather than investment.
The Middle East lens: bigger ambition, higher stakes
In the Gulf, this problem is amplified by scale and speed. PwC estimates AI could contribute US$320 billion to the Middle East economy by 2030, equal to about 14% of GDP in the UAE and 12.4% in Saudi Arabia [7]. IDC projects AI spending in the Middle East, Türkiye and Africa to rise from US$4.5 billion in 2024 to US$14.6 billion by 2028, a 34% compound annual growth rate. It notes that organisations are moving "from proof of concept to proof of value" [8].
Three regional traits make attribution more urgent here than in most markets.
- Centralised, sovereign-first delivery. Governments and conglomerates increasingly deliver AI through shared national or group platforms on in-country infrastructure. Efficient, but without attribution the cost to each ministry, subsidiary or business line is invisible.
- Governance is institutionalised early. The UAE's Charter for the Development and Use of AI (2024) sets 12 principles, including governance and accountability [9]. Saudi Arabia's SDAIA published a national AI Adoption Framework in 2024 and activated AI offices in government entities [10].
- Visible national KPIs. With AI tied to UAE AI Strategy 2031 and Saudi Vision 2030, executives answer for measurable outcomes, not just adoption.
Globally, the drivers differ: in Europe, regulation such as the EU AI Act leads; in North America, cost pressure and shareholder scrutiny lead. The Gulf faces both at once, at unusual speed.
Two tool markets, one blind spot
Enterprises have responded by buying tools, and the market has split in two.
AI governance platforms are growing fast; Gartner forecasts spending of US$492 million in 2026, passing US$1 billion by 2030 [11]. They answer what AI do we have, how risky is it, is it compliant? They rarely answer what does it cost, and is it worth it?
FinOps platforms have added AI dashboards. They answer how much did we spend, and where can we save? They rarely know which AI system a token belongs to, its risk tier, its owner, or whether it is a pilot or in production.
The executive question
The questions that matter most to a CFO and a Chief AI Officer sit in the gap between the two. Governing AI without its economics produces compliant waste. Managing AI cost without governance produces cheap risk.
What good looks like
Organisations that close the attribution gap follow five principles.
- The AI inventory is the cost object. Spend is attributed to the same AI systems, agents and vendors that governance tracks, not to a parallel list of cost centres.
- Attribute at the source. Capture usage from gateways, provider APIs and connectors with system and business-unit context, rather than reconciling invoices afterwards.
- Make the unallocated visible. Unattributed spend is a line to fix, never quietly redistributed.
- Chargeback finance can trust. Frozen statements, versioned pricing, flagged estimates, and disputes corrected through adjustments.
- Tie cost to the lifecycle. From demand to pilot, production and retirement, every stage gate asks "is this still worth it?" with real numbers.
How AnnexIQ closes the gap
We built AnnexIQ around a simple conviction: AI governance and AI economics are one discipline, and they belong on one platform.
On AnnexIQ, AI Governance and AI Cost & Chargeback share a single AI inventory: AI systems, agents, models, tools, MCP servers and vendors. The system your risk team assesses is the system your finance team charges back.
- Attribution by design. Usage from the AI gateway, an ingestion API and connectors is rated against date-versioned price lists and attributed to AI systems and business units, with token types kept separate so nothing is double counted.
- Visible unallocated spend and trusted chargeback. Unattributed spend is shown openly; statements are frozen when issued, estimates flagged, and disputes settled as next-period adjustments.
- The full cost of AI. Seats, capacity commitments, credits, agent budgets, experimentation budgets and reusable assets sit alongside token spend, with budget tracking, anomaly detection and rate-variance analysis.
- Governance and cost, joined up. The same platform runs demand intake, lifecycle gates, risk and compliance assessments, responsible-AI reviews and incident management, so a system's risk rating, compliance status and cost appear side by side.
- The right view for each leader. Business-unit heads see their own spend, executives the whole portfolio, under role-based access and a full audit trail.
The outcome is lifecycle visibility and automation for AI, from the first business request through risk gates, production and chargeback to retirement, on one evidence base.
Three questions to ask your team on Monday
- What percentage of last month's AI spend can we attribute to a named AI system and business owner?
- Are our three most expensive AI systems also rated as our highest-value, and how do we know?
- If a regulator or the board asked for the risk rating and the cost of each production AI system today, how long would it take to answer?
If the answers are "we're not sure", "we don't know" and "weeks", the problem isn't your AI strategy. It's the missing link between governing AI and paying for it.
References
1. FinOps Foundation, State of FinOps 2026 (1,192 respondents): 98% manage AI spend, up from 31% two years earlier. data.finops.org; Linux Foundation press release
2. Gartner, IT Symposium/Xpo 2024 keynote, via CIO Dive: 500–1,000% cost-calculation error; 72% of CIOs breaking even or losing money on AI. ciodive.com
3. CloudZero, The State of AI Costs in 2025: 51% can confidently evaluate AI ROI. cloudzero.com
4. Gartner press release, 29 July 2024: 30% of GenAI projects abandoned after proof of concept by end of 2025. gartner.com
5. Gartner press release, 25 June 2025: over 40% of agentic AI projects cancelled by end of 2027. gartner.com
6. MIT Project NANDA, The GenAI Divide: State of AI in Business 2025. Report PDF
7. PwC Middle East, The potential impact of AI in the Middle East. pwc.com
8. IDC, AI spending in the Middle East, Türkiye and Africa (2024–2028 forecast). emsnow.com
9. UAE Government, The UAE Charter for the Development and Use of Artificial Intelligence (2024). u.ae
10. SDAIA, AI Adoption Framework (2024); Saudi Press Agency. sdaia.gov.sa; spa.gov.sa
11. Gartner press release, 17 February 2026: AI governance platform spending US$492M in 2026, over US$1B by 2030. gartner.com