Chatbot pricing: calculate automated support costs

A customer service chatbot costs far more than its subscription. Between per-conversation billing, the channel's variable pricing, scenario maintenance and the human time spent on cases that escalate, the full cost can easily double your first estimate. This article gives you a method to price every line item, compare cost per conversation with cost per resolved case, and set a stop threshold before you commit.

Why the advertised price tells you almost nothing

Most chatbot offers are presented as a monthly licence, sometimes topped up with usage-based charges. That entry price is real, but it is only a fraction of the total. A chatbot that answers 80% of requests and hands the remaining 20% to an advisor also bills you, indirectly, for that 20%.

Two indicators help you reason correctly. Cost per conversation measures what you spend on every thread opened, resolved or not. Cost per resolved case divides total cost by the number of requests actually handled without human intervention. The second figure is the one that drives decisions, because it captures failed handovers, follow-ups and abandoned conversations.

If you have already budgeted a conversational agent, our guide on a realistic AI agent usage budget complements this method on the model side, while the full cost analysis of an AI agent covers the line items people forget. Here we stay focused on automated customer support and its channel.

The four components of a customer service chatbot's cost

A serious budget separates four lines.

• Licence or subscription: the platform's fixed cost, often indexed to the number of tool users or included conversations.
• Usage: what you pay beyond the plan, per extra conversation, per resolution or per message volume.
• Channel: on WhatsApp, the messaging platform's own billing, independent of your chatbot vendor.
• Maintenance and review: the time spent fixing scenarios, updating answers, reading conversations and adjusting handover rules.

The fourth line is the one most often missing from quotes. A chatbot is not software you install once: customer questions evolve, answers become outdated, scenarios contradict each other. Plan a regular review, even a light one, or the human handover rate will climb month after month.

The WhatsApp channel: understanding variable pricing

On WhatsApp Business Platform, the channel's pricing is set by Meta and billed per delivered message, by category and by market. A key point: not all messages are billable. Service messages — those exchanged within a conversation initiated by the user — fall under a different regime than business-initiated messages.

The 24-hour service window plays a central role here. When a customer writes to you, a window opens and you can exchange freely for 24 hours. Beyond that, you must use approved templates to re-engage. For a customer service chatbot, this means the structure of your scenarios — getting a quick user reply, following up within the window rather than after — directly affects your bill. The official platform pricing page details the categories by market, and the WhatsApp Business policy documentation governs what is allowed on this channel.

Practical consequence: never price the channel cost on a single average. Separate inbound conversations (often cheaper) from outbound campaigns or re-engagements after 24 hours, and check your templates for compliance before counting them in a volume scenario.

Cost per conversation or cost per resolved case?

The two indicators answer different questions and rarely mislead at the same time.

Cost per conversation is useful for comparing vendors or negotiating a plan. It is simple and legible, but blind to quality: a conversation ending in "I'll call an advisor" costs the same as one that solves the problem.

Cost per resolved case is the decision indicator. It is calculated as follows:

1. Add up over a month: licence, billed usage, channel cost, human cost of handovers, and the maintenance share.
2. Count the requests genuinely resolved without human intervention (verified as such, not just "closed by the bot").
3. Divide the total by that number.

A purely hypothetical example: a chatbot billed at €300 per month, €80 of usage beyond the plan, €120 of channel cost, €400 of advisor time for handovers and €100 of monthly maintenance. Total: €1,000. If 500 conversations open and 350 are resolved without a human, the cost per conversation is €2, but the cost per resolved case is about €2.86. The 43% gap between the two figures measures exactly what the chatbot does not do.

A hypothetical budget example on a mixed channel

The table below is a fictional example built to show a budget's structure, not an observed price. It assumes a WhatsApp chatbot with a small volume of inbound conversations and a few re-engagements after the 24-hour window.

Line item	Basis (hypothetical)	Monthly cost (hypothetical)
Chatbot platform licence	Plan including 400 conversations	€300
Usage beyond the plan	100 extra conversations	€80
WhatsApp channel cost	Delivered messages, inbound conversations and a few out-of-window re-engagements	€120
Advisor time on handovers	25 hours at a fictional loaded rate of €16/h	€400
Scenario maintenance and review	Half a day per month	€100
Total	500 conversations, 350 resolved	€1,000

This table makes three points. First, human time weighs as much as the licence: it is the line to watch first. Second, the channel cost is real but secondary at this volume; it would dominate with large outbound campaigns. Third, none of these values is a vendor price: every line must be recalculated with your own volumes and quotes.

Human handover costs: the most underestimated item

Every handover to an advisor costs twice: the human time consumed, and the cost of the bot that ran without resolving. A well-designed handover limits both — our guide on making human handover work covers the design rules, especially passing context so the customer does not repeat everything.

Three levers reduce this line item without hurting the experience:

• Detect early: some requests (disputes, urgency, sensitive topics) should go to a human from the first message. Routing them through the bot only adds cost and frustration.
• Pass the context: conversation summary, case ID, origin channel. A handover without context doubles handling time.
• Measure handover rate per intent: if one category of questions generates half the handovers, that is the one to rework or remove from the bot's scope.

Beware the opposite effect: a bot that refuses to hand over to protect its resolution rate manufactures furious customers and repeat conversations. Resolution rate is only a good indicator when paired with satisfaction measured at the end of the conversation.

How this differs from pricing a general AI agent

A customer service chatbot and a general-purpose AI agent are not priced the same way. The general agent is billed mainly on model usage — tokens, tool calls, task volume — and its budget varies with usage intensity. The customer service chatbot adds a dimension the agent lacks: the channel cost, here WhatsApp's per-delivered-message billing, which applies even when the bot barely "thinks".

Conversely, a fixed-scenario chatbot consumes very few model resources: its cost is mostly structural (licence, maintenance, channel). That is why two chatbots with the same licence price can have very different total costs, depending on whether they call a generative model on every message or rely on pre-written decision trees. When comparing offers, ask explicitly what is billed per message: licence only, licence plus model, licence plus channel, or all three.

Contract, invoicing and taxes: what to check before signing

Before comparing two quotes, put them in the same format. Points to clarify:

• Billing unit: conversation opened, message, or resolution? A "conversation" can mean a ten-minute session or three messages spread over two hours.
• What is included: does the plan cover conversations abandoned after one message? Are human handovers counted as conversations?
• Price review: on WhatsApp, channel rates are set by Meta and can change by category and market. Your contract should state who absorbs a channel cost increase.
• Taxes: depending on the vendor's billing country and yours, VAT may or may not apply to the listed price. A €300 quote excluding tax can become €360 depending on your situation. Also check how channel costs are treated, sometimes billed in a foreign currency and excluding local taxes.
• Commitment: minimum term, notice period, and above all ownership of your scenarios and history if you leave.

A contract clear on these five points beats a price 10% lower with grey areas, because those grey areas are what produce billing surprises.

Procedure: price your chatbot in one hour

Here is an actionable procedure, to repeat every quarter.

1. Export your volumes: monthly inbound conversations, breakdown by intent, current human resolution rate. Without this baseline, everything else is speculation.
2. Estimate the target resolution rate: err on the side of caution and take a low assumption. A bot that resolves 70% of simple requests in a fictional sector might resolve only 40% for you if your requests are more complex.
3. Calculate the human cost of handovers: average handling time multiplied by the expected number of handovers and your loaded hourly rate.
4. Add channel and usage: based on message categories and your inbound/outbound mix, keeping channel rates as a variable to confirm with quotes.
5. Add maintenance: even half a day a month. If no one is assigned, this item is not zero — it is simply invisible until something breaks.
6. Divide by resolved cases: that is your forecast cost per resolved case. Compare it with your current cost per manually handled case.
7. Set a stop threshold: for example, if after three months the cost per resolved case exceeds your manual cost by 20%, or the handover rate exceeds a level your team cannot absorb, you stop or change configuration. Writing this threshold before launch avoids negotiating it under pressure, swayed by money already spent.

Common mistakes and limits of the method

Three pitfalls come up again and again. The first: pricing on the average and forgetting the distribution's tail. A few very long conversations or poorly calibrated outbound campaigns can push the channel cost far beyond the average. The second: confusing "conversation closed" with "case resolved". A sample check — rereading a portion of the conversations closed by the bot, at a size suited to your volume — helps validate the indicator, complemented by the reopening rate and satisfaction measured at the end of the conversation. The third: ignoring seasonality. A volume spike multiplies usage, channel and handovers at the same time; your stop threshold should be set on a busy month, not an annual average.

Finally, the method has limits. It does not measure the reputational value of an instant answer at 10 p.m., nor the time saved by your teams on repetitive tasks — our article on measuring the time actually saved by automation offers ways to quantify that other side. And it assumes your volumes stay stable; a change of offer or channel means recalculating. For a WhatsApp chatbot, also think about compliance: consent, approved templates and channel exit are prerequisites covered in our guide on WhatsApp Business automation, consent and messages, and our article on the WhatsApp AI assistant for customer support covers the experience side.

Frequently asked questions

Is cost per conversation enough to compare two vendors?

No. Two offers with the same cost per conversation can have very different resolution rates, and therefore very different costs per resolved case. Use cost per conversation to compare price grids, and cost per resolved case to decide. Always ask how each vendor defines a "conversation".

Are all WhatsApp messages billed?

No. On WhatsApp Business Platform, pricing applies per delivered message by category and market, and not all messages are billable: conversations initiated by the user within the 24-hour service window fall under a different regime than business-initiated messages. Check the categories that apply to your message mix on the official pricing page.

How do I estimate human handover costs before launch?

Start from your current volume: number of requests per month, the share you plan to give the bot, and average human handling time for the rest. Multiply that time by your loaded hourly rate. Add a 20–30% margin for unexpected handovers, then correct after the first month with real figures.

Should I choose a flat plan or usage-based billing?

It depends on how steady your volumes are. A flat plan protects against spikes but makes you pay in quiet months; usage-based billing follows activity but exposes you to drifts during a spike or a broken scenario loop. Many organisations start with usage billing to learn their volumes, then negotiate a plan once the curve is known.

When should you abandon a customer service chatbot?

When the stop threshold defined before launch is reached: cost per resolved case above your manual cost, a handover rate your team can no longer absorb, or falling satisfaction despite the investment. Before abandoning it, check whether the problem is the bot itself or a poorly chosen scope: removing from the bot the intents that generate handovers costs less than replacing everything.