Multi-Tenant AI Cost Allocation for SaaS Companies

For B2B SaaS companies, the attribution problem is inverted. You're not allocating costs across internal business units; you're allocating costs across customers. If you run a customer service platform that uses AI for ticket categorization, summarization, and auto-resolution, and you have 1,200 customers on your platform, the critical question is: what does it cost to serve each customer with AI?

This is multi-tenant cost allocation, and it's a margin risk that most SaaS CFOs aren't measuring.

Why This Matters

A SaaS company with $10M in annual revenue and 78% gross margin has $7.8M in gross profit. If your per-customer AI cost is eating 8% of that margin for your largest customers (because they drive the most AI volume), you have a pricing problem, a product problem, or both.

Consider: a customer with 50,000 AI-assisted interactions per month costs more to serve than a customer with 500 interactions per month. If your pricing plan doesn't account for this, your largest customers might be unprofitable. You need to know the cost per customer so you can:

1. Set fair, profitable pricing that accounts for their AI consumption
2. Identify customers where AI is destroying margin
3. Decide whether to expand AI features, restrict them to higher-tier plans, or sunset them

The Math: A Worked Example

A mid-market customer service SaaS has:

• 1,200 active customers
• Total AI volume: 120,000 AI-assisted tickets per month
• Total AI cost: $50,000/month (API, embeddings, observability, human review)

Average cost per ticket: $50,000 / 120,000 = $0.417 per ticket.

Now allocate this cost to your customers. Rank them by volume:

| Customer Tier | Number of Customers | Tickets/Month | Cumulative % | AI Cost Allocated | % of Total Cost | | --- | --- | --- | --- | --- | --- | | Top 10 | 10 | 36,000 | 30% | $15,000 | 30% | | Next 40 | 40 | 48,000 | 70% | $20,000 | 40% | | Mid-tier | 200 | 24,000 | 90% | $10,000 | 20% | | Small | 950 | 12,000 | 100% | $5,000 | 10% |

This distribution is typical: the top 50 customers account for 70% of volume and 70% of AI cost. Your bottom 950 customers account for 10% of volume and 10% of cost.

Now ask: Are these customers profitable?

A top-10 customer paying $5,000/month in subscription revenue, consuming 3,000 tickets/month, is being allocated $1,251 in AI cost (3,000 * $0.417). If their subscription margin is 75%, that's $3,750 in gross profit against $1,251 in AI cost, leaving $2,499 in net margin. That's healthy.

But a customer paying $800/month in subscription, consuming 2,000 tickets/month, is allocated $834 in AI cost. If their subscription margin is 75%, that's $600 in gross profit against $834 in AI cost. They're unprofitable by $234/month.

This is the margin risk that multi-tenant AI cost allocation surfaces. Many SaaS companies don't know this math, so they keep unprofitable customers on the books, burning margin at scale.

Allocation Methods

There are three common methods for allocating multi-tenant AI cost:

Method 1: Proportional to volume (simplest)

Allocate all AI costs proportionally to each customer's share of total work items. A customer using 1% of tickets pays 1% of the total AI bill.

Cost per customer = (% of total tickets) × Total AI cost

Pros: Simple, fair, easy to automate. Cons: Doesn't account for complexity (a premium feature customer using advanced features might cost more to serve).

Method 2: Tiered allocation (most common)

Allocate based on customer tier or plan. A customer on the "Pro" plan pays more for AI than a customer on the "Starter" plan, regardless of actual usage.

• Starter plan: included 100 AI interactions/month, included in $29/month subscription
• Pro plan: included 1,000 AI interactions/month, included in $99/month subscription
• Enterprise plan: unlimited AI interactions, tiered pricing starts at $500/month

Pros: Aligns pricing with tier, encourages lower-tier adoption. Cons: Doesn't actually allocate real costs, might mismatch customer consumption.

Method 3: Usage-based allocation (most accurate)

Meter each customer's actual usage (API calls, embeddings, compute) and charge them directly. This is outcome-based pricing taken to the extreme.

• Base subscription: $29/month
• AI add-on: $0.001 per embedding lookup, $0.004 per API call, $0.05 per hour of human review

Pros: Perfectly fair, incentivizes customers to optimize their usage, aligns revenue with cost. Cons: Complex to implement, might surprise customers with overage bills, requires robust instrumentation.

Most SaaS companies start with Method 1 (simple allocation), move to Method 2 (tiered allocation) as they scale, and graduate to Method 3 (usage-based) if AI becomes a material part of their COGS.

Protecting Margin: A SaaS AI Pricing Framework

If AI is now part of your product, you need to think about pricing defensively. Here's a framework:

1. Measure actual cost per customer. For each customer, sum all AI costs (API, embeddings, human review) divided by their usage metrics (tickets, interactions, requests).

2. Identify margin-positive and margin-negative customer cohorts. Are your largest customers profitable? Are your smallest? Are there cohorts (by industry, by region, by feature tier) that are more or less profitable?

3. Set pricing to guarantee margin. If your largest customer cohort costs $0.40 per ticket in AI, don't price your product at $0.30 per ticket. Add a margin buffer.

4. Migrate unprofitable customers. Customers that are margin-negative after allocating AI costs should either move to a lower tier (restricting AI features), move to a higher tier (paying more), or be sunset.

5. Publish cost allocations transparently. If you're moving to usage-based pricing, show customers exactly what they're paying for and why. Transparency builds trust and incentivizes optimization.

The goal is not to maximize AI cost allocation (that's a cost center, not a revenue opportunity). The goal is to ensure that your AI features don't destroy margin and that you're pricing fairly for the cost you incur.

Example: The Math for One Customer

A mid-market customer service platform customer (let's call them Acme):

• Monthly subscription: $2,500
• Monthly AI volume: 8,000 AI-assisted tickets
• Their share of total AI cost (8,000 / 120,000) × $50,000 = $3,333

Acme's economics:

• Subscription revenue: $2,500
• Gross margin @ 78%: $1,950
• Allocated AI cost: $3,333
• Net margin: -$1,383 (unprofitable)

Options:

1. Move Acme to a higher-tier plan (charge them more for the AI they're using)
2. Restrict AI features to lower-volume features (reduce their AI usage)
3. Implement usage-based pricing where they pay a per-ticket fee on top of subscription
4. If none of the above works, sunset them

Without multi-tenant cost allocation, you'd never know Acme was unprofitable. This is how AI can silently erode SaaS margin.

For a detailed walkthrough of how to set up multi-tenant allocation in your own infrastructure, see the article on allocating AI costs to a customer, or return to the pillar article on AI cost attribution.