The AI Tools Your Team Bought Without Telling You
Phil Bolton · April 27, 2026 · 2 min read
A founder I work with runs a 60-person services firm. Last quarter, she asked me to review software spend. We pulled twelve months of credit card statements, expense reports, and the GL. Total came to $612,000. Her budget for the year was $440,000.
Not one big line item. Forty-seven small ones. AI writing tools, AI meeting note tools, AI research tools, AI image tools. Most under $40 a month, charged to personal cards and reimbursed.
Nobody was hiding anything. The spend never crossed the threshold where it triggered a review.
The math at $5M to $15M revenue
Industry data puts average software spend at roughly $10,800 per employee for 2026. For a 60-person company, that's $648,000. AI-native applications more than doubled in spend over the past year. Most of that growth happened underneath the procurement radar.
Mechanically, it's simple. A product manager subscribes to an AI research tool for $29 a seat. Marketing signs up for an AI copy tool at $59. An engineer adds an AI coding assistant at $40. Each one slides under approval thresholds. Annual contract reviews don't catch them either.
By the time finance sees the aggregate, it's a budget overage with no obvious cause.
Software spend used to be a procurement problem. The AI wave turned it into an expense report problem.
What gets missed
Three things tend to fall through the cracks in this environment.
First, duplication. Two teams using two different AI writing tools stays invisible until someone runs a category audit. Recent benchmarks show 65% of enterprise SaaS sits in shadow IT, and a meaningful share is functionally redundant.
Second, per-seat creep. AI tools price aggressively per seat to maximize land-and-expand. A subscription that started at $200 a month for four people becomes $1,400 when usage spreads to 28. Same vendor, no renegotiation, no review.
Third, renewal drift. Most AI products auto-renew on cards. Cancellation friction is real. Tools that stopped being useful six months ago still bill because nobody owns the cleanup.
A simple fix
A monthly software audit takes a competent finance person about two hours. Pull every recurring charge from cards and AP. Add expense report items tagged software or platform. Group by category. Two questions per line: who uses it, and what does it replace?
For most companies, that exercise recovers 15-25% of total software spend in the first pass. The bigger value is the discipline it builds. Once a category is visible, the next AI tool that gets added shows up in the right place.
Software spend grew quietly because AI tools moved faster than finance processes. Make the spend visible monthly, and October's budget conversation stops being a surprise.
Phil Bolton
Founder & Principal at Manitou Advisory
More from the blog
Your Negotiation Agent Optimizes the Wrong Number
Autonomous procurement agents are scored on price. So they'll trade away your payment terms to win one, and hand you a working capital bill nobody approved.
You Priced on Outcomes. You Can't Book the Revenue Yet.
Per-outcome pricing closes deals faster because the buyer only pays when the agent works. Under ASC 606 that same structure is variable consideration, and it pushes recognition out past the quarter you thought you earned it in.
Your Collections Agent Can't Tell a Slow Payer From a Strategic One
An autonomous AR agent optimizes days-outstanding, but some of your DSO is a decision you made on purpose. The agent reads it as delinquency and dunning-emails your best account.
Want to talk about your finance setup?
We help growing companies build the right finance function.
Book a Call →