How much does it cost to implement AI in customer service
heysteff is an AI platform for customer support and sales across WhatsApp, Instagram, Messenger, Gmail and Shopify; Steff is the AI agent that runs it. Before signing with any vendor, it pays to break the real cost into its parts: the monthly license, the setup fee, what Meta charges for each WhatsApp conversation, and the internal time your team spends training the agent. Comparing only the first number against another vendor's first number is the most common way to compare badly.
The four components of the real cost
Almost no conversational AI quote is a single number. There are at least four separate line items, and every vendor groups or hides them differently:
- Platform license. What you pay the software vendor each month, under some pricing model (per seat, per resolution or per conversation).
- Setup fee. A one-time implementation charge, if the vendor charges one.
- WhatsApp conversation cost. A charge billed directly by Meta, independent of whichever software you use.
- Internal cost. Your team's time curating the knowledge base, defining escalation rules and reviewing conversations.
Ignoring any of the four makes two quotes that look different end up costing the same in practice — or the other way around.
Pricing models: per seat, per resolution or per conversation
Per seat. You pay based on how many people on your team have access to the platform. It's easy to budget, but it has no relationship to how much the AI actually handles: a three-person team with a huge message volume pays the same as one with little traffic, and adding a new human agent raises the cost even if the bot resolves most cases.
Per resolution. You pay for each case the AI closes without human intervention. In theory it aligns cost with value delivered, but it opens an uncomfortable question: who defines what counts as "resolved"? If the criterion isn't transparent and auditable by you, it's a model that's easy to inflate in the vendor's favor.
Per conversation. You pay based on how many distinct customers you served in a period, regardless of how many messages went back and forth or how many people on your team use the platform. That's heysteff's model: a "conversation" is one customer served within 24 hours, on any channel. Cost scales with your real support volume, not with your team size or an ambiguous definition of "success".
Setup fee: the market range and why heysteff charges $0
A setup fee usually covers the initial channel configuration, connecting integrations (store, calendar, CRM) and, in some cases, a first round of training the bot with the catalog and the brand's tone. In the conversational AI category this charge commonly exists and varies quite a bit depending on how much manual implementation work each vendor requires: it ranges from zero to several hundred dollars.
heysteff charges no setup fee: it's $0 on every plan. The Starter plan is up and running the same day; Pro takes 3 to 7 days depending on how many integrations you add; Max, built for multi-brand or multi-country operations, is implemented in 2-3 weeks with dedicated onboarding. None of that carries a separate charge.
The cost almost nobody factors in: Meta's per-conversation WhatsApp fees
If you use the WhatsApp Business Cloud API (the official verified number), Meta charges directly for certain conversations, based on the message category (marketing, utility, service or authentication) and the recipient's country. This fee is set and billed by Meta, not by the AI software vendor — it exists whether you use heysteff or any other platform on the Cloud API. It varies by country and category, so there's no single number valid for every business; the right move is to check Meta's current rate for your country before projecting total cost.
An alternative within heysteff is an unofficial WhatsApp number (via Baileys) for businesses that already have a number in use and want to avoid Meta's per-conversation charge, in exchange for operating outside the official API.
The internal cost: who trains and curates the knowledge base
No AI agent starts out knowing your catalog, your policies or your tone by heart. Someone on your team has to load and maintain the knowledge base, define which keywords trigger an escalation to a human, and review real conversations to fix what isn't working. That time is a real cost, even though it never shows up on an invoice: underestimating it is one of the most common reasons an implementation ends up feeling more expensive than quoted.
◆ checklist for comparing quotes
Before deciding, ask every vendor for a clear answer to these five questions.
- Is the license charged per seat, per resolution or per conversation — and how exactly is that unit defined?
- Is there a setup fee? What does it cover and how long until we're operational?
- Is Meta's WhatsApp conversation cost included in the license, or is it billed separately to me?
- What happens if I exceed my contracted volume — is there overage, and how much does each extra unit cost?
- How much internal curation and training work will my team need to put in during the first month?
heysteff's prices, as a transparent example
heysteff charges per resolved conversation, not per seat or per message. Free is $0/month (50 conversations, 1 channel, no card required). Starter is $149/month (500 conversations, 2 channels, 3 users, scheduling with 1 calendar, 3 custom tools, unlimited flows). Pro is $299/month, the most popular (3,000 conversations, 4 channels, 8 users, 1 Shopify store with cart recovery, unlimited campaigns). Max is $549/month (5,000 conversations, unlimited channels, 20 users, 3 Shopify stores). Overage is 0.5 / 0.4 / 0.3 USD per additional conversation depending on the plan, there's a 20% discount for annual payment, and the setup fee is $0 on all of them.
Related
◆ next step
Get a demo and we'll show you the real cost of implementing Steff in your operation, with no fine print.