The real 10-year cost of an AI chatbot subscription.

Most chatbot vendors quote you a monthly number. You compare it to your other SaaS, decide it is reasonable, and put a card down. Two years later you are paying four times what you signed up for and you cannot remember why.

The sticker price is the smallest line on the invoice. The interesting numbers are the ones that compound — quietly, monthly, for as long as the product lives in your stack. This is the math worth running before you commit.

The six lines that actually compound

Pull any chatbot vendor's pricing page. Run a highlighter over everything that grows when your business grows.

1. Base subscription. Predictable. The only line you actually budgeted for.

2. Message overage. Most plans cap monthly messages. Cross the cap and the per-message rate is two to ten times the in-plan rate. The successful month is the expensive month.

3. Model upgrade tax. GPT-4 costs more than GPT-3.5. The next model will cost more than GPT-4. Every model release becomes a vendor billing event, often disguised as a "premium" tier.

4. The watermark tax. "Powered by ___" stays on your widget unless you upgrade. The upgrade is rarely a small one — usually a tier jump or a flat $200–500/mo "white-label" line.

5. Seat creep. Customer support adds an agent. Marketing wants to see the dashboard. Each chair is $30–80/mo. Five seats over five years is not a rounding error.

6. The exit cost. Conversation history, embeddings, training documents, custom prompts — all of it lives in the vendor's database. Migrating to a competitor means rebuilding from zero, and most teams discover this the moment they try to leave.

Plug in two scenarios

Generic numbers are easy to dismiss. Real ones are not. Pick the scenario closer to you.

Scenario A — A SaaS growing from 5K to 50K messages a month

Year 1 you are inside the plan: $79/mo, $948/year. Comfortable.

Year 3 your usage hits the cap. You jump to the next tier: $299/mo. You also turn on GPT-4 because your competitor's bot is sharper: that is a +$200/mo line.

Year 5 you cross 50K messages, blow past the next cap, and start eating overage. Real bill: ~$900/mo. You pay $200/mo extra to remove the watermark because two enterprise prospects asked.

Year 10, run the cumulative number. You are at roughly $80,000–110,000 in subscription fees alone, and you still do not own the conversation logs.

Scenario B — An agency reselling chatbots to 20 clients

Most "agency" plans cap at 5–10 chatbots before forcing you to the enterprise tier. Either you split clients across multiple accounts (operational debt) or you go enterprise (pricing-on-call, usually $999–2,499/mo).

20 clients × ~$120/mo client-facing tier = $2,400/mo, of which ~$1,500/mo is your cost. Margin is fine until the vendor raises prices, which they will, and the increase eats into the margin you priced your retainers around.

Year 10, you have paid $180,000–300,000 to the vendor. You have also handed them a customer list — the vendor knows your every client by sub-domain.

The framework, in one paragraph

Take any chatbot vendor's pricing page. Identify the six lines above. Project your usage three, five, and ten years out. Multiply. Add a 5% annual price increase, because vendors raise prices and you will not have the leverage to negotiate it away. Compare the total to a one-time number. The gap is the real cost of the subscription model.

Where this lands

The arithmetic does not need a vendor's marketing copy to make a point. A subscription is a long-tail liability with a friendly first invoice. A one-time deploy on infra you already pay for is a fixed liability with no second invoice.

We built Chatmancer for the second model — a one-time deploy to your own AWS account, no per-message fees, no watermark tax, no seat creep, and the conversation data lives in a database you own. But the math holds even if you pick someone else: run the 10-year number before the 30-day trial ends. The invoice you do not budget for is the one that hurts.