How much does AI customer service cost?

Published 23 August 2026 · Replio

“AI customer service” is quoted three completely different ways — per conversation, per seat, or one flat monthly fee — and the same support volume can produce wildly different bills depending on which model you signed. This guide breaks down what AI-powered customer service actually is, what makes up the bill, and how to work out your real number before you commit.

What “AI-powered customer service” actually means

Strip away the marketing and it comes down to one thing: an AI agent that reads a customer's message, understands what they're asking, and answers it from your business information — your prices, your hours, your policies, your stock — rather than from a script somebody hand-wrote for twenty predicted questions.

A genuine AI support layer does four jobs. It answers instantly, 24/7, in whatever language the customer wrote in. It acts — captures a lead, logs a complaint, takes a booking, remembers the customer for next time. It works everywhere your customers already are, so WhatsApp, Instagram DMs, Messenger, Telegram and your website chat widget all draw on the same knowledge. And it hands off to a human the moment the conversation needs one.

That's the product you're pricing. Anything that only does the first job — a decision-tree bot with a chat bubble — is a much cheaper thing, and should be priced like one.

The three pricing models, and who each one favours

Almost every vendor uses one of three models. The difference matters far more than the headline number, because each one moves the risk somewhere different.

Per-conversation pricing is fine at low volume and painful at scale. Per-seat pricing suits an operation that is mostly humans with a bit of AI. A flat plan suits the opposite — mostly AI, with humans stepping in on the hard ones — and it's the only model where a viral week doesn't arrive as an invoice.

The line items that actually make up the bill

Whatever the model, a real AI customer service budget has up to five components. Price them separately or you'll compare two quotes that aren't measuring the same thing:

• The platform subscription — the AI, the inbox, the knowledge base, the integrations. The number on the pricing page.
• Channel fees paid to the channel, not the vendor — the big one is WhatsApp, where Meta charges its own per-conversation rate. Our WhatsApp Business API pricing explainer walks through how those rates work. Telegram, Instagram, Messenger and a website widget have no equivalent fee.
• Seats — how many humans need inbox access. Some plans include a handful; extra seats are usually per-user per-month.
• Volume — whether messages are metered at all, and if they are, what happens when you exceed the allowance: soft cap, top-up pack, or automatic upgrade. A vendor that meters volume is charging you more in the months your business does well.
• Add-ons — extra languages, analytics, API access, voice. On some platforms these are the majority of the bill.

What it costs with Replio

Replio is built for businesses of every size — a solo shop through to a large multi-brand team — and prices flat, so the bill doesn't move because you had a good month:

Three things worth knowing, because they're where budgets usually go wrong elsewhere:

• Only incoming customer messages count. Replies from the AI or your team are free and unlimited, so a chatty resolution doesn't cost more than a curt one.
• There is no message cap to hit. Messages are unlimited on every plan, including the free one, so a month that goes well costs the same as a quiet one and there is no allowance to watch, top up or get upgraded out of. Plans differ by channels and seats instead.
• No per-message markup. On WhatsApp you connect your own number and pay Meta directly at Meta's rates. On every other channel there is no per-message fee at all.

Extra seats are $20/seat/month, annual billing is roughly two months free, and there's a free 14-day trial of everything with no card required. If you genuinely only need one thing, the à la carte products sit below the plans: Telegram Agent at $49/month, QR Chat at $29/month, or Links & QR at $9/month.

The costs that aren't on the pricing page

Two quotes with the same headline can differ by a wide margin once you add these up. Before signing, ask specifically:

• Is there a markup on channel fees? Some platforms resell WhatsApp conversations at a margin. Ask for the per-conversation number they charge versus Meta's published rate.
• Is AI included, or an add-on? On seat-priced helpdesks, the AI is frequently a separate line — sometimes per resolution.
• Is every channel included, or per channel? Paying separately for WhatsApp, Instagram and a web widget is how a modest plan triples.
• What does onboarding cost? Setup and implementation fees are common in enterprise support tooling and rarely appear on the public page.
• What does a second language cost? With Replio, 50+ languages are automatic — see multilingual AI customer support — but on many platforms each additional language is a project, not a setting.

How to work out your own number

You don't need a spreadsheet model. Four questions get you within range:

1. How many customer messages arrive in a normal month? Count incoming only, across every channel — most businesses underestimate the DMs.
2. Which channels genuinely matter? If it's WhatsApp alone, a WhatsApp-only plan is the cheapest honest answer. If customers reach you three or four ways, an all-channel plan almost always beats stitching separate tools together.
3. How many humans need the inbox? Not the whole company — the people who actually reply.
4. What's your busiest month, not your average? Price the peak. That's the question that decides whether a per-conversation model is safe for you.

Then compare that against the cost of covering the same hours with people. We ran that comparison in detail in why AI support beats hiring more agents — the short version is that the gap isn't mainly about salary, it's that one hire can't cover nights, weekends and five languages at once, and an AI layer does that by default.

Frequently asked questions

Related