Most AI chatbot pricing pages quote $39. The real invoice lands at $1,800. Here is the per-resolution math, WhatsApp markups, and seat fees vendors bury in fine print.
A founder I will call Marta runs a 40-person logistics SaaS out of Rotterdam. Last March she signed up for what the pricing page told her would be a EUR 99/month chatbot. By July her invoice was EUR 1,847. She emailed me a screenshot with the subject line: "how is this legal."
The math was not illegal. It was just hidden. Her contract had a per-resolution fee of $0.99 that kicked in after the first 50 included resolutions. Her support team had grown from 3 to 6 agents, and each seat cost $74/month on top of the AI layer. WhatsApp conversations were billed at a 38% markup over Meta's published rates. The OpenAI token costs for the underlying model were a separate line item she had not noticed in the order form. And the "setup assistance" she had opted into during onboarding was a one-time $2,400 charge amortized over six months so it did not look like much on any single invoice.
None of that appeared on the marketing page. The marketing page said EUR 99.
I have now looked at 23 of these invoices from customers who switched to us in the last eight months. The pattern is the same every time. The headline price gets you in the door. The real cost shows up in five places the pricing page is engineered to hide. This piece walks through each one with real numbers from real published pricing pages, then runs the math for a representative B2B SaaS at three different volume tiers so you can stress-test your own vendor before the invoice surprises you.
The per-resolution fee is the biggest line item nobody talks about
Intercom rebranded its AI agent as Fin and made it the centerpiece of its pricing. Fin is a genuinely good product. It is also priced at $0.99 per resolution, and that single line item is where the bills get strange.
A "resolution" in Intercom's definition is any conversation Fin closes without escalating to a human. Sounds great until you do the math. A mid-sized B2B SaaS with 2,000 support conversations a month and a 70% deflection rate is paying Fin to resolve 1,400 of them. At $0.99 each that is $1,386/month just for the AI layer. On top of that you need Intercom Suite seats for your human team, which is where the rest of the bill lives.
Intercom Suite pricing as of their published page in 2026:
- Essential: $39/seat/month
- Advanced: $99/seat/month
- Expert: $139/seat/month
Most teams running Fin in production are on Advanced or Expert because the lower tiers do not include workflow automation. A 5-agent team on Expert is $695/month in seat fees. Add the Fin resolution charges at 2,000 conversations with 70% deflection and you are at $2,081/month. Their pricing page leads with $39.
The defense Intercom gives — and it is not unreasonable — is that you only pay when Fin actually works. If the AI does not resolve the ticket, you do not pay the resolution fee. That is true. But it also means your cost is directly tied to your conversation volume, which means a viral launch week or a Black Friday spike turns into a four-figure surprise on your next invoice. Founders running lean do not love variable line items they cannot cap.
Zendesk stacks the AI add-on on top of an already expensive seat
Zendesk's pricing approach is different but lands in the same place. The Suite Team plan starts at $55/agent/month. Suite Growth is $89/agent. Suite Professional is $115/agent. That is just for the helpdesk.
The AI Agents product (formerly Ultimate.ai, which Zendesk acquired) is licensed separately. Published rates fall in the $50-$80/agent/month range depending on volume and contract length, and that is on top of the seat fee. A 5-agent team on Suite Professional with AI Agents enabled is paying $115 + ~$65 = $180/agent, or $900/month base before any conversation volume considerations.
What Zendesk does that Intercom does not is bundle a fixed monthly volume into the AI add-on, which makes it easier to forecast. What they also do is require you to negotiate the AI Agents pricing through sales rather than self-serve, which means smaller teams often end up paying more than larger ones because they have less leverage. I have seen 8-person teams quoted $95/agent for AI Agents while 200-person teams in the same vertical were paying $52.
WhatsApp markup is the quietest fee in the industry
Meta charges for WhatsApp Business Platform conversations on a per-conversation basis, with rates that vary by country and conversation category. As of Meta's published 2026 rates, a business-initiated marketing conversation in Germany is about EUR 0.0876. A utility conversation is EUR 0.0421. A service conversation initiated by the customer within 24 hours of their last message is free.
Vendors that resell WhatsApp access — ManyChat, Respond.io, the WhatsApp tier of most chatbot platforms — add a markup on top of Meta's rates. ManyChat's published WhatsApp pricing works out to roughly 38-42% above Meta's direct rates depending on volume tier. Respond.io's markup floats around 25-35%. Some white-label vendors charge double Meta's published rate and call it a "managed WhatsApp service."
The markup is defensible when the vendor handles template approval, opt-in management, and Business Verification on your behalf. It is not defensible when they do not, and most of the cheaper tiers do not. If you are sending 50,000 WhatsApp conversations a month in Germany and your vendor is charging a 40% markup, you are paying roughly EUR 1,700/month extra for what is functionally a passthrough.
The fix is to either go direct through Meta's Business Solution Provider program or work with a vendor that bills WhatsApp at cost. SimplyBoost bills Meta's published rate plus a flat platform fee included in your plan, which is what we would recommend even if you do not use us — it is the only billing model where the WhatsApp line item does not drift upward with volume.
Per-message tiers create cliff edges that hit at the worst time
Chatbase is one of the more popular site-embed chatbot builders in the market, and their pricing illustrates the tier cliff problem better than anyone else. As of their published pricing page in 2026:
- Free: 20 message credits/month
- Hobby: $40/month, 2,000 message credits
- Standard: $150/month, 12,000 message credits
- Pro: $500/month, 40,000 message credits
- Enterprise: custom
The jump from Standard to Pro is $350/month — more than triple your bill — to get from 12,000 to 40,000 credits. If your site does 15,000 monthly conversations you cannot stay on Standard. You would be forced onto Pro and you would be paying $500 to use 37% of the credits in your tier. The effective rate per conversation on Pro at 15,000 actual usage is $0.033, versus the $0.0125 it would be at full Pro utilization.
Tidio's Lyro pricing has the same structure with a sharper edge. The free tier caps at 50 Lyro conversations per month. Fifty. A small e-commerce store doing 30 conversations a day blows through that in two days. The paid tiers start at $39/month for 200 conversations and scale up from there. The math at any meaningful volume is that you are paying $0.20-$0.40 per conversation, which is fine for low-volume use cases and terrible for anyone doing real support work.
Model token costs as a separate line item is the new gotcha
This one is recent and it is getting worse. Several vendors — I will not name them because the practice is changing fast — bill the underlying LLM costs from OpenAI or Anthropic as a separate passthrough line item. You pay the platform fee, and then you also pay the GPT-4 or Claude API costs at whatever the model provider charges, sometimes with a markup.
The argument vendors make is that token costs are variable and depend on conversation length, so they should not be bundled. Fair. But the consequence is that your monthly bill has two unpredictable components — conversation volume and per-conversation token usage — and the second one is invisible until you get the invoice. A long, complex support conversation can burn 8,000-15,000 tokens. At GPT-4 Turbo's published rates that is $0.08-$0.15 per conversation in raw model cost. Multiply by your monthly volume.
What to ask any vendor before signing: "Are LLM API costs included in the plan price, or billed separately?" If the answer is separate, get the projected token cost in writing based on your expected conversation count and average length.
Frequently asked questions
Why do vendors hide pricing this way?
Because the landing page conversion rate is driven by the headline number, and the headline number is the cheapest possible tier. Sales teams know the real bill will be 3-5x that. Once you are integrated, switching cost is high enough that most teams renew rather than migrate.
Is per-resolution pricing always bad?
No. If your conversation volume is low and predictable, per-resolution can be cheaper than per-seat. The break-even is roughly 800-1,200 resolutions/month at $0.99 — below that, Fin is competitive. Above it, flat pricing wins.
How much does Meta actually charge for WhatsApp conversations in 2026?
Meta's published rates vary by country. Service conversations initiated by the customer within 24 hours of their last message are free. Business-initiated conversations are categorized as marketing, utility, or authentication, with rates ranging from roughly $0.005 to $0.15 per conversation depending on country and category. Germany marketing conversations are about $0.094 USD equivalent.
What should I ask a vendor before signing the contract?
Five questions: 1) Are LLM API costs included or billed separately? 2) What is the per-resolution or per-conversation rate above the included volume? 3) Is the WhatsApp rate at Meta's published cost or marked up? 4) Are seat fees mandatory or can the AI handle volume without additional seats? 5) Is there a setup fee, and if so, when does it appear on the invoice?
Why does SimplyBoost price flat?
Because variable pricing punishes growth and we would rather have customers who scale with us than customers who delay scaling because the chatbot bill is the bottleneck. Our $39-$169/month plans include conversations, AI model costs, and WhatsApp at Meta's direct rates. If you outgrow Scale we have a custom enterprise tier, but most teams do not.
Methodology
All vendor pricing in this article was checked against the vendor's published pricing page as of May 2026. Specific sources:
- Intercom pricing: intercom.com/pricing — verified Fin at $0.99/resolution, Suite tiers at $39/$99/$139 per seat/month
- Zendesk pricing: zendesk.com/pricing — Suite tiers verified at $55/$89/$115 per agent/month. AI Agents pricing aggregated from publicly shared Zendesk quotes, May 2026
- Chatbase pricing: chatbase.co/pricing — tier structure and credit allocations verified May 2026
- Tidio Lyro pricing: tidio.com/pricing — free tier limit of 50 conversations verified May 2026
- ManyChat WhatsApp markup: calculated by comparing ManyChat's published per-conversation WhatsApp rate against Meta's published WhatsApp Business Platform rates
- Meta WhatsApp Business Platform rates: developers.facebook.com/docs/whatsapp/pricing — May 2026 rate card
Where vendor pricing is custom or negotiated, estimates are based on quotes shared by founders in customer Slack groups and Reddit threads in the last six months. Actual quoted pricing may vary.
---
I am one of the founders of [SimplyBoost](https://simplyboost.io). We build flat-priced AI agents for sales and support across web, WhatsApp, Instagram, and Facebook Messenger — EUR 39 to EUR 169/month, no per-resolution fees, no per-seat fees, WhatsApp at Meta's direct rates. We are registered in the Netherlands (KVK 87456346) and host on Frankfurt EU infrastructure. If you have been quoted something that looks like Marta's invoice and you want a second opinion before you renew, email me directly. I will do the math with you on a call, no pitch, even if you stay with your current vendor.