A Dollar Per Resolution

Intercom charges $0.99. Zendesk charges $1.50 per committed resolution ($2.00 pay-as-you-go). Gorgias charges $0.90 on annual billing. For each customer question their AI resolves, you pay roughly a dollar or more.

If your AI handles 500 questions a month, that's $450 to $750 just for the AI resolutions — on top of your monthly subscription.

A dollar doesn't sound like much until you multiply it.

Where That Dollar Goes

The big tools use large language models (GPT-4, Claude, or fine-tuned variants) to power their AI agents. Here's the cost breakdown:

LLM inference costs: A typical multi-turn support conversation costs $0.03 to $0.15 in API fees to the LLM provider. Longer conversations with more back-and-forth cost more. Let's call the average $0.08.

Knowledge base retrieval: These tools search your help center to find relevant articles, then feed them to the LLM as context. Vector search and retrieval-augmented generation add $0.01 to $0.03 per query.

Infrastructure and compute: Servers, databases, queue systems, monitoring. Maybe $0.02 to $0.05 per resolution at scale.

Total actual cost per resolution: roughly $0.10 to $0.25.

The rest — the $0.75 to $0.90 difference between their cost and your price — is margin. And platform premium. And because they can.

Why They Can Charge That Much

Lock-in is the biggest factor. If you've built your support operation on Intercom or Zendesk — with workflows, macros, integrations, and historical data — switching is painful. They know this.

Bundling helps too. The per-resolution fee sits "on top of" your subscription. It feels incremental even though it adds up fast.

And honestly, the market just hasn't pushed back yet. Per-resolution pricing is new enough that when Intercom set $0.99 and Zendesk followed at $1.50, everyone accepted it as what AI support costs.

What a Fair Price Looks Like

If the actual cost to process a resolution is $0.10 to $0.25, a fair margin would put the customer price at an example $0.20 to $0.40. That gives the provider healthy margins while keeping the cost reasonable for customers.

At an example rate of $0.20 per resolution, 500 messages costs $100/month instead of $500. That's $4,800/year in savings. For a startup watching every dollar, that's real money.

The Classification Cost Advantage

Here's why some tools can charge less: they don't use an LLM for every interaction.

A purpose-built classification model costs a fraction of what an LLM costs to run. The model is smaller (hundreds of millions of parameters vs hundreds of billions), the inference is faster (100ms vs 2 seconds), and it doesn't need to process a knowledge base with each request.

The trade-off is that a classifier doesn't generate freeform answers. It identifies the customer's intent and triggers a pre-configured response. For the 60 to 70% of support messages that are predictable (billing, passwords, order status, refunds), this works just as well. For the 30% that need nuanced, contextual responses, you still need a human.

But here's the thing: at $1/resolution with Intercom, those 30% of complex messages are going to a human anyway. Fin doesn't handle them. So you're paying $0.99 for the same outcome that classification achieves for well under $1.

The Volume Comparison

*Zendesk offers volume discounts at higher tiers (down to $1.00/resolution at 5,000+). At 5,000 resolutions/month, the difference vs classification is $4,000/month. $48,000/year. That's a salary.

When $1/Resolution Is Worth It

There are cases where paying more per resolution makes sense:

• You need multi-turn conversations handled by AI (not just classification)
• Your questions are highly varied and don't fit a fixed intent taxonomy
• You've already invested heavily in a knowledge base that Fin or Zendesk AI uses well
• Your volume is very low (under 50/month) so the per-resolution cost barely matters

When It's Not

For everyone else — teams with predictable support patterns, price-sensitive startups, companies with moderate to high volume — paying 5x more for the same outcome doesn't make sense.

Prices are already coming down. As more alternatives offer classification at an example $0.20 to $0.30, the $1/resolution price will look increasingly hard to justify.

Check the math for your own volume. If the difference pays for a new hire, better marketing, or another month of runway, you know what to do.