How Much Does an AI Customer Service Chatbot Cost?

An AI customer service chatbot typically costs somewhere between $50 and $2,000 per month for a small or mid-sized business, but the headline subscription is rarely the number you actually pay. What determines your real bill is the pricing model underneath it: flat subscription, per resolution, per session, or per contact. Two vendors with identical sticker prices can produce invoices that differ by a factor of five once your conversation volume is real.

This is a guide to working out your true cost per conversation before you sign, rather than discovering it on the third invoice.

Last updated July 2026.

The four pricing models, and what each one really means

Almost every vendor in this market uses one of four models. Learning to recognize them on sight is most of the battle.

Notice what three of those five have in common. Per resolution, per session, and per contact all charge you more precisely when the product works better. That is a strange thing to buy. You are paying for deflection, then paying again each time deflection happens.

Working out your real cost per conversation

Here is the arithmetic worth doing on a napkin before any demo. Take a mid-sized team: eight support people, about 3,000 customer conversations a month, of which an AI can plausibly resolve 70 percent, so around 2,100 AI resolutions.

• Per-resolution vendor at $0.99: 2,100 times $0.99 is $2,079 in AI charges, on top of eight seats at roughly $85 each ($680). Total in the region of $2,759 per month.
• Per-session vendor at $49 per 100: 2,100 sessions is 21 blocks, or about $1,029, on top of agent seats at roughly $55 each ($440). Total around $1,469, though sessions and resolutions are not defined identically, so this comparison is approximate by nature.
• Flat subscription: the tier price. It does not move.

Now double the volume for a peak month, because December, a product launch, or a shipping problem will do exactly that. The per-resolution bill goes to roughly $4,158 in AI charges. The flat bill goes nowhere. That gap, not the sticker price, is what you are really choosing between.

The fees hiding underneath the subscription

Even a flat AI subscription sits on top of channel costs that belong to someone else, and this is where quotes quietly diverge.

• WhatsApp: Meta charges per conversation, commonly a few cents depending on category and country. Many tools resell this with a markup of around 20 percent. Ask directly whether the vendor passes Meta's fee through at cost or adds a margin. We break the mechanics down in WhatsApp Business API pricing explained.
• SMS: carrier fees per message, plus 10DLC registration in the US.
• AI assistant add-ons: a copilot for your human agents is usually billed separately, often around $29 per seat per month.
• Onboarding and implementation: enterprise contracts frequently carry a one-time fee that never appears on the pricing page.

A single WhatsApp conversation can therefore carry three separate variable costs: Meta's fee, a vendor markup on that fee, and a per-resolution AI charge. Each is individually defensible. Stacked, they make your cost per conversation nearly impossible to forecast.

Is an AI chatbot cheaper than hiring another agent?

Usually, yes, and it is worth being precise about why rather than hand-waving at it. A US support representative costs somewhere in the region of $45,000 to $60,000 a year in salary, and the loaded cost with benefits, payroll taxes, software, and management overhead runs meaningfully higher. Divide by twelve and one additional rep is roughly $5,000 to $7,000 a month, all in, before you count the weeks of ramp time.

Against that, even an expensive per-resolution AI bill of $2,000 to $3,000 a month looks reasonable, and a flat subscription looks cheap. The comparison is not quite apples to apples, because a human handles the hard, ambiguous, emotionally charged conversations that AI should not touch. But that is the point: the AI is not replacing the rep, it is clearing the routine tail so the rep spends their day on work that actually needs a person. The real question is not whether to bring another support hire onto the team or buy AI. It is how much of the routine volume you can responsibly automate before that next hire becomes necessary at all.

What should a small business expect to pay?

For a small business with modest volume, budget $50 to $300 per month for a capable AI chatbot on a flat plan, plus channel fees if you use SMS or WhatsApp. Mid-market teams with real volume and multiple channels land in the $300 to $1,500 range on flat pricing, and considerably higher on metered pricing once resolutions add up. Enterprise deals with custom SLAs, SSO, and onboarding run well beyond that.

Be wary of anything advertised as free. A free tier is generally either sharply limited on conversations, missing the AI entirely, or a funnel toward a metered plan where the real money is collected. Free is a customer acquisition tactic, not a pricing model.

How do I compare chatbot pricing fairly?

Do not compare sticker prices. Build a single spreadsheet, put your own numbers in it, and force every vendor into the same shape:

1. Your peak-month conversation volume, not your average. Price the worst month, because that is the one that hurts.
2. The share the AI will plausibly resolve. Be conservative. Ask the vendor for their observed rate on businesses like yours.
3. Every per-unit charge, and the vendor's exact definition of the unit. A resolution, a session, and a conversation are three different things, and each vendor defines its own.
4. Full seats versus limited seats. Bundled Lite or collaborator seats cannot work a queue.
5. Channel fees, and whether they are marked up.
6. The contract floor: minimum commitments, annual lock-in, and what happens if you shrink.

Then compute a single number for each vendor: total dollars divided by total conversations in a peak month. That is your true cost per conversation, and it is the only figure that compares honestly across pricing models.

Why we price flat

We will be direct about our own bias here, because you should weigh it. MessageAgent charges a flat subscription and passes carrier and Meta message fees through at cost with no markup. There is no per-resolution meter and no per-contact ladder.

The reason is the incentive problem. When a vendor charges per resolution, a better AI means a bigger invoice, and you end up quietly hesitating to route more volume to the product you already bought. We would rather the AI resolve everything it possibly can and have the bill stay where it was. That means your cost per conversation falls as the agent gets better, which is the direction it ought to move.

If you want to see what the agent actually handles, the customer service chatbot page walks through it, and the conversational AI platform page covers how one brain works across six channels. For a worked example of the metered model on a real vendor's numbers, see Intercom pricing in 2026.

The short version

Budget $50 to $300 a month if you are small, $300 to $1,500 if you are mid-market on a flat plan, and considerably more if you pick a metered vendor and your volume is healthy. Whatever you do, price your peak month rather than your average one, get the vendor's exact definition of a billable unit in writing, and check whether channel fees carry a markup. The businesses that get burned are almost never the ones that paid too much per seat. They are the ones that never asked what happens to the invoice when the product works.