Why cost per customer is the question that matters
A total monthly cloud-plus-AI-plus-SaaS bill tells you almost nothing about whether the business is getting more efficient or less. It just grows with usage, and a growing number in isolation isn't a signal either way. Divide that same spend by active customers and you get a number you can compare over time. Does it cost more or less to serve the same customer as the product, its usage patterns, and the infrastructure change?
This is also where cloud spend and AI/LLM spend converge into one question. A per-request AI cost and a per-customer infrastructure cost are the same idea applied to two different meters. As more products wrap an LLM around a core workflow, both meters increasingly bill the same customer.
How the calculator works
Cost to serve, per customer, is total monthly cloud, AI/LLM, and SaaS spend divided by active customers. Gross margin percentage is (MRR minus that same total spend) divided by MRR. Gross margin dollars per customer is revenue per customer (MRR divided by customers) minus cost per customer. All three update live as you move the sliders. Nothing you enter is sent anywhere.
The verdict band is CostMon's own framing. It builds on Benchmarkit's 2025 SaaS Performance Metrics report, which puts median SaaS gross margin at 77%. Subscription-specific margin runs 75–80%+ for the same cohort. Below 0% is underwater, below 60% is thin, 60–80% is healthy and roughly brackets that median, and 80%+ is strong, top-quartile territory.
What moves your cost per customer
Three things move this number, and only one of them is really about infrastructure spend. Customer count moving up while spend stays flat improves it automatically: economies of scale, not optimization. Spend moving down while customers hold steady is the classic FinOps lever. It covers rightsizing idle cloud resources and cutting SaaS tools nobody uses. It also covers catching AI cost regressions, such as a prompt that grew unchecked or a model swap nobody tested for cost.
The third lever is architectural, not operational. A feature that adds an LLM call to every customer request adds a per-customer cost. That cost scales with usage, which flat infrastructure spend does not. That's worth watching on its own, separate from the blended total. A rising AI share of the mix, even with a flat or improving total, is worth understanding before it compounds.
A single estimate isn't the point: the trend is
This calculator is a live, in-browser estimate for a single month's numbers. It is useful for a gut check. It does not replace tracking the real trend. The number that matters is whether cost per customer improves or worsens quarter over quarter. To answer that, you need the real spend and the real customer count, not a guess you re-enter each time someone asks.
CostMon's Unit Economics feature tracks that trend. You define a business metric (active customers, requests, transactions) and record its value for each period. CostMon divides your connected cloud, AI, and SaaS spend by that value. It splits the result by allocation dimension and charts it over time.