Blog
Field notes from the AI audit
Practical, evidence-first writing on where AI spend leaks, why the vendor dashboards won't show it, and how to get the dollars back.
By Spendassay Research · RSS
Jul 25, 2026 · 6 min read · Cost and benchmarks
The metered traps hiding in AI tool pricing
Four ways AI tool pricing bills past the sticker — allowances that run out, shared limits, uncapped tokens and plan changes mid-contract, July 2026 prices.
Read →Jul 23, 2026 · 7 min read · Finance operations
How to forecast next year's AI budget
AI spend is two lines that behave nothing alike. Forecast seats from the hiring plan, usage from run-rate, and name the assumptions that will break.
Read →Jul 17, 2026 · 6 min read · Buying and renewals
How to negotiate AI renewals from evidence, not vibes
Your vendor knows your usage; you have an invoice. How to close that gap before the renewal call — two separate asks, real benchmarks, and the three counters.
Read →Jul 15, 2026 · 5 min read · Waste and recovery
The hidden cost of idle AI seats — and how to find them in an afternoon
Idle AI seats run to roughly a tenth or a fifth of the bill. How to price them at the rate you really pay, and find them across every vendor in an afternoon.
Read →Jul 13, 2026 · 6 min read · Waste and recovery
How to find unused GitHub Copilot licenses
A by-hand method for checking Copilot seats: where the billed list lives, what use data GitHub shows, the cut-offs to use, and the traps that inflate it.
Read →Jul 11, 2026 · 7 min read · Cost and benchmarks
Cursor vs Copilot vs Claude Code: the real cost per engineer
How Cursor, GitHub Copilot and Claude Code charge — list prices read in July 2026, plus the usage charges that can flip the ranking on a heavy team.
Read →Jul 9, 2026 · 6 min read · Waste and recovery
Do you need both Copilot and Cursor?
A neutral way to decide whether to run GitHub Copilot and Cursor together, plus the seat-list join that shows which overlapping seats are dead weight.
Read →Jul 7, 2026 · 7 min read · Buying and renewals
What each AI vendor's admin console actually exposes
A working matrix of what the Copilot, Cursor, Anthropic, OpenAI, Gemini, Bedrock and M365 Copilot admin consoles will tell you - who, how, how late.
Read →Jul 5, 2026 · 7 min read · Cost and benchmarks
How much should you spend on AI tools per engineer?
Public AI-spend benchmarks disagree, and every one describes other companies. DX puts total cost at $200–600 per engineer per month — here is why the figures scatter, and how to place yourself against your own books.
Read →Jul 3, 2026 · 8 min read · Finance operations
How to file AI costs in your books
AI spend arrives as five kinds of charge. Here is how to split seats from usage, cloud from apps, and the AI you sell from the AI your own staff uses.
Read →Jul 1, 2026 · 6 min read · Cost and benchmarks
What share of the engineering budget should AI be?
There is no correct percentage for AI's share of engineering spend. Here are the two comparisons that hold up, the one that does not, and how to set a band.
Read →Jun 22, 2026 · 7 min read · Finance operations
Who owns the AI budget?
Engineering, finance, purchasing and security all have a real claim on AI spend. Here is the split that works, the grid, and the line nobody owns.
Read →Jun 20, 2026 · 8 min read · Measuring the payoff
What the research actually says about AI coding ROI
A tour of the AI coding research with the method attached to every claim, including what GitHub's 55% and METR's 19% figures can and cannot support.
Read →Jun 18, 2026 · 7 min read · Finance operations
What an AI cost report is, and how to build one
An AI cost report is one page showing what AI costs you and what you get back, with a proof level on every line. Here is the structure, line by line.
Read →Jun 16, 2026 · 6 min read · Measuring the payoff
Honest AI ROI modeling: ranges, not fake precision
How to build an AI ROI figure a CFO can sign — show what the speed cost you, put a proof level on every number, and publish a range with its assumptions.
Read →Jun 14, 2026 · 7 min read · Buying and renewals
How to attribute OpenAI and Anthropic API costs to teams
Seat spend labels itself. Token spend does not. The ladder of ways to match OpenAI and Anthropic API costs to teams, from vendor projects to gateway tags.
Read →Jun 12, 2026 · 6 min read · Waste and recovery
What good AI seat utilization actually looks like
Where the 85% and 60% AI seat use cut-offs come from, the gap between touched and really used, and what the finding becomes once yours is healthy.
Read →Jun 10, 2026 · 7 min read · Measuring the payoff
Why your AI ROI number is wrong
Eight named errors push AI return numbers too high — double counting, no cost of speed, teams that picked themselves. Here is what to report instead.
Read →Jun 8, 2026 · 7 min read · Buying and renewals
The evidence pack that wins an AI renewal
The seven-section document that turns an AI renewal into an audit. What to pull, when to pull it, how to state the two asks, and what quietly weakens it.
Read →Jun 6, 2026 · 7 min read · Waste and recovery
Finding shadow AI spend on the company card
Shadow AI is a spending problem before it is a security problem. How to find it in card data, sign-in logs and expense lines, and what to do with each hit.
Read →