Would you pay someone to tell you what your AI actually costs per customer?
Quick gut-check from the people actually living this.
Most teams I talk to have no idea what their AI costs per customer, per feature, or per month. The ChatGPT seats, the Claude bill, three API keys, a couple of tools someone expensed last quarter. It creeps up quietly, and when I ask "what's your AI cost per active user?" the answer is almost always a shrug.
I'm building a service around that. Not another dashboard. The free cloud tools and the $29/mo SaaS already do "here are your charges." What people keep telling me they actually want is someone to set up the visibility, then sit down and say where the money is going and what to cut. Human interpretation and a prioritized cut list, not more charts.
The ladder I'm testing:
- Free AI spend calculator: enter your spend, get a savings range
- $1,500 one-day "vitals scan": connect billing + one usage source, top 3 findings
- $9,000 fixed-scope audit: per-feature and per-model unit economics + a prioritized list of what to cut
- $3-4k/mo retainer: ongoing monitoring as models and prices shift
Honest questions, brutal answers welcome:
- At what monthly AI spend would that $1,500 scan feel like an obvious yes vs a waste of money?
- Is $9K for a fixed-scope audit reasonable, too high, or a joke?
- Would you actually pay a person for this, or would you never pay for a service and just want a tool?
That third one is the one I most need the truth on. Thanks in advance.
Monthly spend is the wrong threshold; gross margin at risk and recoverable waste matter more. A $5K bill with 40% avoidable usage may justify the scan, while a $50K optimized bill may not. Price one pilot against verified 90-day savings: baseline, recommended changes, and realized savings. If you cannot attribute roughly three times the fee, $1,500 will not feel obvious.
This makes sense, thanks. Bill size is a weak proxy and the 3x-recoverable test is going straight into how I qualify this, which also puts the weight back on attribution.
And since the scan connects to billing anyway, measuring realized savings against a baseline is the easy part. The real variable is who owns implementation, which probably decides where savings-based pricing fits: the one-time scan flags the waste, but verified 90-day savings really lives in an ongoing engagement where I can see it through.
Appreciate you sharpening this.
That's two products, not one pricing ladder. The scan can charge for diagnosis because it doesn't control implementation. Savings-based pricing only fits the managed engagement, with an agreed baseline window and exclusions for growth, model changes, and seasonality; otherwise every win becomes an attribution argument.
I think the interesting question isn't whether companies need another AI cost dashboard—they probably don't. I'd keep validating whether buyers are really paying for cost visibility or for the confidence to scale AI usage without unpleasant surprises. If your recommendations consistently unlock more AI adoption rather than just lower spend, the value becomes much easier to justify.
Thanks for the reply. Your take on focusing less on the number but more on what they can or can't do with that number is something I didn't think about but makes a lot of sense. I saw the same thing with cloud costs: the bill was hard to predict, which made it hard to greenlight new capabilities that needed more spend.
AI is a little different in that the capabilities are clearer than cloud infra, but the questions are still there. What could we turn on if we trusted the cost estimates?
That's exactly the interesting question.
Reading your reply gave me one thought about the difference between making costs visible and making expansion decisions feel safe. I don't think I could explain that properly in a thread without oversimplifying it.
I'd rather discuss it in the context of what you're building.
If you're interested, what's the best email to reach you on?
I am definitely interested. seth@sslaboratory.com is the best email. Async is great, or a quick call if that's easier to talk through.
Thanks for putting this much thought into it, it has been a really valuable thread so far.
Thanks! I’ve just sent it over.
Looking forward to hearing your thoughts whenever you have a chance.
Yes, if the answer connects model, retrieval, tooling, retries, support, and infrastructure costs to a customer or workflow. Founders need unit economics, cost drivers, and scenarios they can act on, not another cloud bill summary.