Enterprise generative AI spending hit $37 billion in 2025, up 3.2x from $11.5 billion in 2024, according to Menlo Ventures. Token prices dropped over 90% in under two years. Yet 80% of enterprises missed their AI cost forecasts by more than 25%, per Benchmarkit research. Prices went down. Spend went up. Your clients are living this right now, and most consultants are not advising on it.
The Unit Price Trap
OpenAI's flagship model pricing fell from $30 per million input tokens in 2023 to $2 per million with GPT-4.1 in 2025. That is a 93% reduction. Anthropic's Claude Opus dropped from $15 per million tokens to $5, a 67% cut, per their published pricing.
The natural assumption is that AI got cheaper. It did not. AI got cheaper per unit. Usage exploded.
I saw this exact pattern at Hartford Steam Boiler. Sensor costs dropped 80% over five years. The response was not to spend less on sensors. It was to instrument everything. Total monitoring spend quadrupled while per-sensor cost collapsed. The vendors won. The buyers spent more than they planned.
Usage-based pricing now represents 43% of SaaS companies, up from 35% in 2024, according to Metronome's State of Usage-Based Pricing report. When your clients adopt AI tools priced per token, per API call, per image generated, or per agent hour, their monthly cost becomes a function of consumption, not commitment.
The Three Spend Failure Modes
Mode 1: The Experiment Trap
A client starts with one AI tool at $50/month. It works. They add another. Then three more. Then an enterprise platform. Within six months, the median AI spend reaches $11.38 per employee per month, per Ramp's AI Index. For a 50-person company, that is $6,828 annually. For the top 1% of firms, spend hits $7,449 per employee per month. That is $4.5 million annually for a 50-person company.
The experiment never had a budget because nobody called it an experiment.
Mode 2: The Shadow AI Problem
Individual team members adopt AI tools with personal credit cards. Marketing signs up for Jasper. Sales starts using an AI dialer. The CEO pilots an AI writing assistant. None of these appear on the company's SaaS audit. The finance team discovers the spend at quarter-end when they reconcile credit card statements.
Mode 3: The Integration Multiplier
Every AI tool that connects to other systems generates API calls that generate costs on both ends. A CRM integration that triggers an AI summary after every customer call sounds efficient until the client realizes they are paying for the CRM seat, the AI summary API, the storage for the transcript, and the token cost for processing every conversation.
The Consultant's Advisory Framework
Your clients need a framework, not another tool recommendation. Here is the FOCUS Strategy applied to AI spend:
Find your actual AI spend. Audit every credit card, every department head, every software subscription. Most companies undercount AI spend by 40-60% because individual purchases are categorized as "software" or "marketing tools," not "AI."
Organize by function, not by tool. Map every AI tool to the business function it serves: content creation, customer service, sales enablement, data analysis, operations. If two tools serve the same function, one should go.
C , Calculate cost per output. What does it cost to generate a blog post? Handle a customer inquiry? Score a lead? Convert these into cost-per-output metrics so the client can compare AI-assisted costs against manual costs and competitor benchmarks.
U , Understand the consumption curve. Usage-based tools have nonlinear cost curves. The first 10,000 API calls might cost $0.01 each. The next 100,000 might cost $0.005 each. Or they might cost $0.02 each. Read the pricing page. Model the 12-month cost at projected usage, not current usage.
S , Set a spend ceiling. Every AI tool gets a monthly budget cap. When the cap is hit, usage pauses until the next billing cycle. This is the equivalent of a stop-loss. It prevents the experiment from becoming a budget crisis.
The Deliverable: AI Spend Audit Report
Consultants should package this as a $3,000-$5,000 engagement. The deliverable is a 10-page report covering:
- Current AI spend inventory. Every tool, every cost, every user.
- Function mapping. Which business function each tool serves, with overlap identified.
- Cost-per-output analysis. Benchmarked against manual costs and industry averages.
- 12-month projection. At current growth trajectory, what will AI cost in 12 months?
- Optimization recommendations. Consolidate overlap, negotiate volume pricing, set spend ceilings, identify tools to cut.
This engagement is repeatable across every client in your portfolio. It positions you as the advisor who saves money, not the one who adds costs.
The Doctrine Connection
Verification beats optimism. Your clients believe AI is saving them money because the per-unit cost is lower. Verification means tracking total spend, not unit price. The consultant who shows a client their actual AI cost trajectory, grounded in data rather than marketing copy, earns trust that no tool recommendation can match.
Frequently Asked Questions
Q: What is the average AI tool spend for a small business in 2026?
Ramp's AI Index reports a median of $11.38 per employee per month across all industries. However, this median obscures massive variance. The bottom quartile spends under $3 per employee per month. The top 1% spends $7,449 per employee per month. The question is not what the average is. It is where your client falls on that curve.
Q: How do I convince a client to conduct an AI spend audit?
Frame it as risk management, not cost-cutting. "80% of enterprises miss AI cost forecasts by more than 25%. Let us find out where your company falls before the next budget cycle." The data sells the engagement.
Q: Should consultants recommend specific AI tools to clients?
Recommend categories, not brands. "You need an AI content generation tool" is advice. "You need Jasper" is a vendor relationship. When you recommend a category, the client makes the choice. When you recommend a brand, you own the outcome if the tool underperforms.
