Why chatbot pricing feels confusing (and the four models that explain it)
AI chatbot pricing in 2026 is the most fragmented category in SaaS. Two products that solve the same job for the same buyer can quote you $29 a month and $1,500 a month, and both prices are internally consistent with how each vendor thinks about value. The confusion comes from four distinct pricing philosophies that have evolved over the last three years, each with a real argument behind it.
The four models, in the order they emerged:
- Per-message pricing. You buy a monthly message quota and the vendor caps you there. Used by ChatRaj, Chatbase, SiteGPT.
- Per-resolution pricing. You only pay when the bot actually resolves a conversation. Intercom Fin at $0.99 per resolution, HubSpot Breeze at $1 per resolved conversation.
- Per-seat plus AI add-on. You pay for human agent seats and an AI tier on top. HubSpot Service Hub, Tidio plus Lyro, and (until recently) Drift.
- Flat enterprise contracts. A negotiated annual contract bundling unlimited usage with SSO, dedicated support, and procurement-grade terms. CustomGPT Enterprise and every large vendor's custom tier.
The right model depends on your traffic shape, team structure, and how predictable you need your bill to be.
The 4 pricing models in detail
Model 1: Per-message pricing
You buy a plan covering N messages per month and at the cap the bot either stops answering, charges overage, or downgrades the response. Every "message" usually counts both the user question and the bot reply as one unit.
This model exists because it is the closest analog to legacy SaaS metered billing. The unit is visible and easy to audit. The downside: quotas force you to either over-provision or under-provision.
ChatRaj uses a flat monthly quota with no overage. Pro is $29 a month for 10,000 messages. Growth is $99 a month for 50,000 messages. Chatbase uses message credits where heavier models consume credits faster. Standard is $150 a month for 10,000 credits ($0.015 per credit); Pro is $500 a month for 40,000. SiteGPT uses plain per-message: Starter at $39 a month for 4,000 messages, with extra messages at $39 per 5,000 (about $0.0078 each).
Model 2: Per-resolution (outcome) pricing
You pay nothing for messages that did not solve anything. You pay a fixed price every time the bot resolves a conversation. The vendor's incentive aligns with yours.
Intercom Fin AI Agent defined this model at $0.99 per resolution. A resolution is counted when the bot's final reply ends the conversation either because the customer confirmed satisfaction or left without escalating. There is a 50-resolution-per-month minimum ($49.50 floor). Intercom's published average resolution rate is about 67%, so model expected cost as conversation volume times 0.67 times $0.99.
HubSpot Breeze Customer Agent uses the same pattern at $1 per resolved conversation, but it requires a Service Hub Professional seat at $100 per seat per month plus a one-time $1,500 onboarding fee. Each seat includes 3,000 Breeze credits (enough for roughly 30 conversations); additional resolutions are $1 each.
The outcome model is genuinely cheaper than per-message when your bot deflects above 60% and your conversation volume is moderate. It gets expensive at scale: 10,000 resolved conversations a month at $0.99 each is $9,900.
Model 3: Per-seat plus AI add-on
This is the model legacy customer service platforms picked when they bolted AI onto an existing seat-based product. You keep paying for human agent seats and pay an extra fee for the AI on top.
HubSpot Service Hub Professional is $100 per seat per month with a $1,500 onboarding fee. Breeze AI sits on top at $1 per resolved conversation; seats include 3,000 credits each. Tidio starts around $59 a month for the chat platform, with Lyro AI as a $39-a-month add-on for 50 AI conversations (about $0.78 each). Standalone Lyro pricing without the human-chat plan starts at $0.58 per AI conversation.
Drift was the third major vendor in this category and is no longer a live option. Clari plus Salesloft announced the Drift sunset on March 6, 2026 and named 1mind as the exclusive AI successor. Drift had previously gone offline in September 2025 after an OAuth security breach that affected over 700 organizations. Existing Drift customers are migrating to 1mind, which focuses on inbound qualification and AI-powered live demos rather than general support chat. If you are evaluating Drift today, do not.
Model 4: Flat enterprise contracts
Above roughly 50,000 conversations per month, metered models stop being competitive against a negotiated annual contract. The contract usually bundles high or unlimited usage, SSO, SOC 2 Type II evidence, a signed DPA or BAA, dedicated account management, and custom SLAs.
CustomGPT.ai Enterprise is one example: unlimited data capacity, custom SSO, dedicated engineering support, 99.9% uptime SLA, pricing on request. ChatRaj Enterprise covers the same ground at lower base pricing because the cost structure is leaner.
The argument for flat contracts is predictability. Finance budgets the line item once a year, security reviews the vendor once a year. The argument against is that you usually pay a premium over what a heavy-user metered plan would cost.
The philosophical case for each model
Per-message optimizes for transparency. You can audit the unit. The cost: you pay for bad answers the same as good ones.
Per-resolution optimizes for alignment. The vendor only earns when the bot works. The cost: "resolution" becomes a measurable thing the vendor controls, and the definition can drift.
Per-seat plus AI add-on optimizes for continuity. Teams add AI on top of an existing help desk without rebuilding their workflow. The cost: you keep paying seat fees even after the AI deflects most of the queue.
Flat contracts optimize for budgeting. Procurement gets one number. The cost: opacity. You rarely know the effective per-conversation rate.
None of these is wrong. The model that fits depends on which tradeoff hurts least.
Per-message in practice: ChatRaj, Chatbase, SiteGPT
Per-message works best for SMB and mid-market sites between 1,000 and 50,000 conversations a month. Below 1,000 you overpay for unused quota; above 50,000 you want a flat contract.
ChatRaj is the lowest published per-message rate. Pro at $29 covers 10,000 messages ($0.0029 each). Growth at $99 covers 50,000 ($0.00198 each). Free is 100 messages a month. Chatbase Hobby is $40 for 1,500 messages ($0.0266 each); Standard $150 for 10,000 ($0.015); Pro $500 for 40,000 ($0.0125). Chatbase ships WhatsApp and Slack channels that ChatRaj does not have yet. SiteGPT Starter is $39 for 4,000 messages ($0.0098 each); Growth on annual is about $79 for 10,000.
The pure per-message vendors fit when your bot's job is content-grounded Q&A on a marketing or documentation site with reasonably predictable traffic.
Per-resolution in practice: Intercom Fin
Intercom Fin is the cleanest example of the per-resolution model. $0.99 per resolution, no seat fee for Fin itself, 50 resolutions per month minimum.
At 1,000 conversations a month with a 67% resolution rate, Fin costs about $663 (670 resolutions times $0.99). At 5,000 conversations, $3,316. Fin reads your existing Intercom help articles and conversation history, which gives it strong grounding. The catch: Fin is most cost-effective when you also pay for Intercom's seat plans (around $39 per seat per month), so $0.99 per resolution is rarely the only line item on the bill.
Per-seat plus AI: HubSpot, Tidio plus Lyro, Drift sunset
HubSpot Service Hub Professional is $100 per seat per month with the mandatory $1,500 onboarding fee. Breeze AI is $1 per resolved customer conversation. Each seat includes 3,000 credits per month (about 30 conversations). Adding seats does not add more included credits; the 3,000-credit pool is the same regardless of seat count. Above 30 conversations per seat per month, top-ups are $10 per 1,000 credits.
Tidio with Lyro starts at $59 a month for the chat plan plus $39 a month for Lyro's 50-conversation tier. Higher Lyro tiers run roughly $79 for 500 conversations and $149 for 1,000. Standalone Lyro starts at $0.58 per AI conversation.
Drift is no longer a live option. The Clari plus Salesloft sunset announcement was March 6, 2026 with 1mind as the named successor. 1mind focuses on inbound qualification and live AI demos rather than general support chat, so the migration is not feature-for-feature. If Drift is on your shortlist, replace that row.
Flat enterprise contracts: when they make sense
Above roughly 50,000 conversations a month, every metered model becomes worse than a negotiated annual contract. Per-unit pricing optimizes for the median customer and overcharges the upper decile.
CustomGPT.ai Enterprise: unlimited data capacity, custom SSO, dedicated engineering support, 99.9% uptime SLA, pricing on request. ChatRaj Enterprise covers the same ground (SSO, SOC 2 Type II evidence, signed DPA, dedicated success manager) at lower base pricing.
Negotiation levers that work on flat contracts: annual prepay (10 to 20% off), multi-year commit (15 to 25%), case-study rights (5 to 10%), reference-customer participation (5 to 10%). The published "enterprise pricing on request" is almost always 20 to 40% above the floor a procurement team can hit.
Hidden costs every buyer should price in
The advertised plan price is rarely the total. Six line items show up most often after signing.
Overage fees. If your plan caps at 10,000 messages and you hit 12,000, what happens? Some vendors block the next message (ChatRaj). Some auto-bill overage credits (Chatbase, HubSpot Breeze). Some downgrade quality. Ask before signing.
Model upcharges. "GPT-4 access" sometimes means GPT-4.1-mini in practice; full GPT-4.1 is a paid upgrade. "Claude" might exclude Sonnet and only give you Haiku. Read the footnote.
Integration tiers. WhatsApp Business, Slack, Zendesk, Salesforce are typically gated behind higher tiers regardless of message volume. Your plan tier is sometimes set by the integration list, not by your message count.
Onboarding fees. HubSpot Service Hub Professional has a mandatory $1,500 onboarding fee. Intercom has a similar implementation fee at higher tiers. ChatRaj, Chatbase, SiteGPT do not. On a sub-$5,000-a-year budget, an onboarding fee can be a third of your spend before you send a message.
Annual prepay commitment. Advertised monthly prices are often the annual-billed price divided by 12. Pay-monthly is typically 15 to 25% more.
Volume cliffs. Some vendors price the next tier at 5x to 10x with no middle option. Tidio's pricing jumps from $59 to $749 with nothing in between.
Building your own: when in-house RAG actually saves money
The threshold where building in-house beats paying a vendor is higher than most engineering teams estimate.
You need at least one full-time engineer who knows retrieval, embeddings, and vector databases. Fully loaded cost in a US market is roughly $200,000 a year. You need infrastructure: a vector DB ($200 to $2,000 a month), an LLM provider bill ($500 to $20,000 a month depending on traffic), and an observability stack ($300 to $1,000 a month). You need 30 to 50% of one engineer's time on steady-state maintenance.
The lowest believable annual cost for an in-house build is about $250,000 in year one and $150,000 a year after. Break-even against ChatRaj Enterprise is north of 5 million conversations a year. Against Intercom Fin variable cost alone (ignoring engineering salary), break-even is roughly 350,000 resolutions a year.
Build in-house when you have unique data the vendors cannot ingest, regulatory requirements that prevent third-party processing, or genuine top-decile scale. Otherwise the vendor option is cheaper and ships faster.
Putting it together
The framework in the install steps section below maps traffic, team structure, and budget to one of the four models. Most SMB and mid-market buyers land on per-message; most support-led teams land on per-resolution; most enterprise buyers land on flat contracts.