Twenty years ago, a CIO told a room of his peers that he felt like a magic orange. Every budget cycle his CFO asked for ten percent more for ten percent less, as if the IT budget could be squeezed forever and still keep its shape. Back then, the juice went back to finance. Today, whatever is freed from the run is often the only money there is to fund AI.
Capital committed to the wrong AI programs at scale does not come back. Cut too early and competitors pull ahead, and spend into something nobody is measuring and the P&L takes the hit. Either mistake can be serious enough for the board to lose its nerve halfway through the change.
Four figures show how little most enterprises can see on either side.
95%
of enterprise generative AI pilots delivered no measurable P&L impact.
Source: MIT Project NANDA, The GenAI Divide, 2025
65 to 75%
of technology spend goes on keeping the lights on in most enterprises that have not measured it, against a target of 50 percent run, 25 percent grow and 25 percent transform.
Source: Gartner target. The range is a practitioner observation, not survey data.
When two thirds of the technology budget goes on running what already exists, very little is left for what comes next, and that is a question of strategy before it is a question of efficiency.
Most cost tools assume that cost stays put, that spend stays inside IT and that planning happens once a year. AI changed all three at once. Cost now rises and falls with usage, and spend arrives through business units and the software they buy. Agents also do work whose cost sits right beside the payroll it replaces.
FinOps brought urgency to cloud cost and stops where the cloud invoice stops. IT financial management brought discipline to IT cost and works on a monthly or quarterly rhythm. FP&A brought rigor to planning and works from general ledger exports. A newer set of AI cost tools watches the token meter, and Excel still holds most of it together.
Moving money from the run to the grow needs both sides of the trade in one model. Working out what the run can spare is an allocation question across cloud, software, labor, the data center and every vendor contract. Working out what AI returns is a unit economics question across tokens, platform, data and people. A tool that holds only one side cannot make the trade.
AI development cost falls under the same accounting standards as any other software: IAS 38 under IFRS, and ASC 350-40 with ASU 2025-06 under US GAAP. The eligible share can be capitalized and amortized instead of being expensed in the year, which changes when the cost reaches the P&L and how much the run has to absorb. Eligibility depends on evidence of which project, which activity and which hours, and most enterprises expense all of it because that evidence was never captured.
Deloitte, Accounting for AI costs associated with internal-use software development, Technology Spotlight, 2026
In practice, moving money from the run to the grow means consolidating overlapping vendors, exiting contracts on time and renegotiating at renewal. Those decisions have to be read from the same model as the AI cost, or the reallocation and the proof that it worked end up in two spreadsheets that never meet.
MagicOrange holds both sides in one allocated model, across every category of technology spend: consumed, owned and human. MOTEF, the MagicOrange Technology Economics Framework, describes what the model does at board level as See, Price, Plan, Optimize, Control and Prove. The first five make reallocation possible, and the last one shows whether it worked.
Allocated and defensible by service, business unit and vendor.
Enterprise Allocation Modeling, IT Financial Management
Because money is freed at consolidation, exit and renewal.
IT Financial Management, with the vendor and contract optimization reports
On the same model as the run that pays for it.
AI Economics, FinOps
Run, grow and transform as rolling, consumption-based scenarios.
Budgeting and Forecasting
So the reallocation and its proof sit in one place and stand up to the CFO’s first question.
Every capability, through the allocation engine
Project and development-stage tagging, hours by project from Time Capture, AI usage as standard objects, an opex and capex split on each project and amortization once an asset exists. Your accounting policy stays your own, and the model holds the evidence and shows how it got there.
IT Financial Management, Time Capture and Resource Planning, AI Economics
Instead of a quarterly slide prepared by IT finance, the ratio is read live from the full cost model. The CIO and the CFO work from the same number, the gap to target is plain to see and the three largest contributors to the run are named.
Your team builds the model, and the information it holds is what runs the company.
Neither result is about AI. Both show the run side of the trade working, which is where the money for AI has to come from.
Fragmented Excel models consolidated into one automated system, run times cut to minutes, defensible chargebacks posted back into the general ledger. Major international bank.
VAT recovery improved by separating recoverable from non-recoverable IT spend, a further €5m of infrastructure savings identified, audit-ready across hybrid IT. Major European bank.
