Every company is spending on AI.Almost none can say where the next dollar should go.
Today Garth tells heads of engineering what their AI spend returned, with the proof attached. In December 2026 the same record reaches the CEO and CFO: what came back, how sure you can be, and which dollar to scale, hold or stop.
KePSLA
Companies know what AI costs. They don't know what it earned.
Nearly every budget now carries an AI line, and it's usually the fastest-growing one. Ask what it returned and you get a different answer from every team, each worked out its own way.
A company pays $3M a year for AI tools. Is it getting $1M back, or $10M? Nobody in that company can say.
Sources: CloudZero, Finding the ROI of AI: The Finance Perspective (June 2026), a vendor-sponsored survey of 260 finance leaders. IBM Institute for Business Value, CEO Study 2025, 2,000 CEOs: only 25% of AI initiatives delivered the return expected. The $3M example is ours, not a customer's.
One record of what every AI dollar returned, read from the work itself.
This isn't a survey or a self-reported estimate. Garth sits inside the systems where the work is done, the tickets, the repositories, the release records, the tool bills, and reads what was asked, what was delivered, what it cost and what came back. That's what we mean by Enterprise AI Decision Intelligence.
Each team's own estimate, in a spreadsheet nobody checks.
One record for every team, worked out the same way, and a grade on how sure you can be.
- 01
Know what it cost
What each piece of work cost, people and AI together, before AI and after, not the AI bill on its own.
- 02
Prove what it returned
Every claim is graded: seen in the work, inferred, or proven in the accounts. So you know how far to trust a number before you act on it.
- 03
Decide where the next dollar goes
For each piece of work the answer is one of four: scale it, fix it, hold it or stop it. The head of engineering and the CFO decide it from one record, together.
Everything else is an estimate. Surveys give you hopes, login counts give you usage, and a number typed into a planning tool is still somebody's guess. The record of the work is the one source nobody gets to round up.
What "proven in the accounts" means. The saving or the revenue shows up in a line your finance team can point to: a cloud bill that fell, licences that were cancelled, revenue against a feature that shipped. "Inferred" means the work points to it and the ledger doesn't yet. The grade tells you which is which.
It starts in engineering because that's where the money is, and where Garth already runs.
Across enterprises, engineering is the biggest line in AI application spend and the fastest growing. It's also where Garth already runs, with five enterprise customers in the US and India. The leadership view reads that same record. We're not starting a second one.
Menlo Ventures, 2026: The State of Enterprise AI (November 2026), annualised spend measured to September 2026: $34.0B of $77.3B on coding tools, up 7.7x in a year against 5.4x for all enterprise AI; customer success is the next fastest application line at 4.3x. The agents engineering teams build and the models they call directly sit in Menlo's infrastructure line, so we read this as the floor of engineering's AI spend, not the whole.
G360Know the ROI. AI adoption and spend against what shipped, as a number a CFO can defend.→
GReviewGovern your AI code. What AI writes is checked against what was asked and your rules before it ships, so the spend doesn't buy rework.→
GReleaseAnticipate what won't ship. Promised dates against what has actually been delivered, every slip on record.→
GKSThe shared brain under all of it: one living map of your teams, tools, work and spend that every product reads from.→
Engineering stays the core. The same question reaches the rest of the AI budget, one business function at a time.
We ask every function that spends on AI the same question, and a function joins only once Garth can read its work from the systems it already runs.
- Now
Engineering
The full picture for the head of engineering, live today. G360, GReview and GRelease keep their own roadmaps. The leadership view reads their record and doesn't replace them.
Live · still the core The CEO and CFO view
The same evidence, carried to the people who sign the AI budget.
Design partner seats openOther business functions
Which come first will be settled with the design partners. Each adds its own work, its own costs and its own measures of value to the one record.
What stays the same, whichever function it is: what was asked, what was done, what it cost in people and AI, what it was worth to the function and to the company, and how strong the proof is. That's how a CEO gets to compare one function's AI with another's on one page.
Already a Garth customer? Nothing changes in what you use today. The numbers your engineering team already reports are the record the leadership view will read, and you'll see that view before your leadership does.
A few companies will shape the CEO and CFO view with us.
We're opening it to a small number of design partners: companies whose leaders want this answer on their own AI budget, and whose finance team will check our numbers. Tell us about your AI spend and who from engineering and finance you'd want in the room. We'll come back within one business day.
How we count. Every figure on this page that isn't ours carries its source and, where it matters, its caveat. Where something is an estimate or an example, we say so. The same rule applies to the leadership view: no number without saying how it was proved. Garth is made by Darner Technology, Inc., a US company with its product team in India.