- ai
WhatsApp AI Agent for Business: Cost and Timeline (2026)
What a custom WhatsApp AI agent costs: €6,000 to €12,000 and 2 to 4 months to launch. The factors that drive price, the phases and running costs.
Short answer: a custom WhatsApp AI agent for a business typically costs between €6,000 and €12,000 to build and takes 2 to 4 months to go live. Where your project lands in that range depends mostly on how many of your systems it has to connect to (CRM, ERP, booking tool, online store), whether it only answers questions or also takes actions, and how much tuning the pilot needs. On top of the build you pay running costs every month: Meta’s per-message charges on the WhatsApp Business Platform, AI model usage, and maintenance.
This guide breaks down where those numbers come from, so you can compare quotes on the same terms.
What a WhatsApp AI agent is, and how it differs from a chatbot
A classic WhatsApp chatbot runs on menus and rules: “reply 1 for opening hours, 2 for orders”. When a customer types something the script didn’t expect, it gets stuck.
An AI agent understands free-text messages, looks up your business information and can act inside your systems. It can check an order status, book an appointment, change a delivery address, or hand the chat to a person with the full context. We cover the difference in more depth in chatbot vs AI agent.
Being able to act is what makes an agent useful. It is also why building one costs more than switching on a chatbot platform.
How much does a WhatsApp AI agent cost?
For a custom build, the range we see for business projects is €6,000 to €12,000 of implementation. When you ask for quotes, split the cost into three parts:
- Build and implementation (one-off): conversation design, integrations, WhatsApp Business Platform setup, testing and pilot.
- Usage costs (variable, monthly): what Meta charges per message and what the AI model costs per conversation.
- Maintenance (monthly fee): tuning, keeping the knowledge base current, monitoring and support.
If a vendor gives you a single figure without separating these, ask what it covers.
Table: what drives the price
| Factor | Closer to €6,000 | Closer to €12,000 |
|---|---|---|
| Integrations | One or none, with a well-documented API | Several (CRM, ERP, bookings, ecommerce), some with a weak API or none |
| Conversation complexity | Answers common questions | Takes actions: books, charges, changes orders |
| Knowledge sources | A website and an FAQ document | Catalogs, policies, manuals and data that changes daily |
| Channels and languages | WhatsApp only, one language | WhatsApp plus web chat or other channels, several languages |
| Human handoff | Email alert | Live handover to a support team, with history and business hours |
| Security and GDPR | No sensitive data | Personal, payment or health data; audit logs |
| Pilot and tuning | Short pilot, few scenarios | Long pilot at real volume, several tuning rounds |
The factors behind price and timeline
Integrations
This weighs the most. An agent that only answers questions doesn’t need access to your systems. One that checks orders or creates bookings has to connect to your CRM, ERP, booking engine or store (Shopify, WooCommerce or similar).
The quality of each integration matters more than the count. A modern, documented API is quick to connect. Legacy software with no API, or an API nobody on your team really knows, means investigation, sometimes a middle layer, and a lot more testing.
Conversation complexity
“What time do you close on Saturday?” and “I want to swap the size on yesterday’s order and ship it to a different address” are very different requests. The second means identifying the customer, finding the order, checking whether it can still be changed, applying the change and confirming it.
Every action the agent can take needs clear rules: what it can do alone, what needs confirmation, and what it must never do. Designing and testing those limits takes time.
Knowledge sources
The agent answers from what you give it: your website, return policy, catalog, prices. If that information is organised and current, there is less work. If it is spread across old PDFs, spreadsheets and one colleague’s memory, part of the project is putting it in order. That work pays off outside the agent too.
Channels and languages
Many businesses want the same agent on WhatsApp and on their website chat or Instagram. Reusing the logic on a second channel costs less than building it twice, but each channel has its own formats and limits. Languages work the same way: current models handle English, Spanish or French well, but tone still needs checking in each language with real cases.
Human handoff
A good agent knows when to stop. You need to decide when a chat goes to a person (an angry customer, a complaint, a large sale, a question it can’t resolve), who receives it, what happens outside business hours, and how the history reaches your team so nobody asks the same thing twice. Connecting this to the helpdesk your team already uses adds work, and it is what keeps the switch invisible to the customer.
Volume and Meta, API and model costs
To run an AI agent on WhatsApp you need the WhatsApp Business Platform (the API), not the regular WhatsApp Business app. Meta charges per delivered message, and the rate depends on the message category (marketing, utility, authentication, service) and the recipient’s country.
One recent change matters for AI agents in particular. According to Meta’s own WhatsApp Business FAQ, from 1 October 2026 service messages (your replies inside the 24-hour customer service window) are also charged per message: each business phone number gets the first 1,000 free every month, and Meta charges from the 1,001st at its published rates. Since most of what an agent sends are service replies, check your expected volume against the current rate card before you budget.
Then there is the AI model, which scales with the number and length of conversations. At low and medium volume it is usually a small line item. At thousands of conversations a day, it is worth designing the agent to use fewer tokens: shorter replies, leaner lookups.
Security and GDPR
If the agent handles personal data, and it almost always does, GDPR applies to customers in the EU: tell users they are talking to an AI, limit what you store and for how long, sign data processing agreements with your vendors and know where data is processed. Payments, health or financial data raise the bar, and the budget with it. Outside the EU, check your local equivalents too.
Pilot and tuning
No agent is perfect on day one. The pilot, with real customers at limited volume, is where you find out how people actually ask things. Answers get adjusted, knowledge gaps get filled, actions get fixed. Cutting the pilot short to save money usually costs more later.
Ongoing maintenance
Your prices change, you launch products, you update your shipping policy, and Meta and model providers update their services. So after launch there is a monthly maintenance fee covering conversation monitoring, knowledge updates, behaviour tuning and incident response. Ask what it includes and what is billed separately.
How long does it take? Timeline by phase
Total time runs from 2 to 4 months. A simple project with few integrations and good APIs sits near 2 months. One with several systems, sensitive actions and a long pilot gets closer to 4.
| Phase | What happens | Typical duration |
|---|---|---|
| 1. Discovery | Use cases, real conversations, systems to connect, agent limits | 1 to 2 weeks |
| 2. Design | Flows, tone, handoff rules, GDPR decisions | 1 to 2 weeks |
| 3. Build and integrations | WhatsApp Business Platform, CRM, ERP and others; knowledge base | 3 to 7 weeks |
| 4. Internal testing | Your team tests normal and edge cases | 1 to 2 weeks |
| 5. Pilot with real customers | Limited volume, daily review, tuning | 2 to 4 weeks |
| 6. Launch and maintenance | Full volume, monitoring, ongoing improvements | Ongoing |
Start one task early: registering your number on the WhatsApp Business Platform and verifying your business with Meta. It is rarely hard, but it depends on Meta and can delay the pilot if left to the end.
A worked example
This example is illustrative and does not describe a specific client.
Picture an online sports equipment store receiving around 150 WhatsApp messages a day. Most ask where an order is, how to return something, or whether a size is in stock.
The agent connects to the store to look up orders and stock, and to the shipping provider for tracking. It can start returns following the store’s policy. Complaints and orders above a set value go to a team member with the full history. It works in English and Spanish.
With two integrations on modern APIs, limited actions and a one-month pilot, a project like this would sit in the middle of the range for both cost and time. Add a legacy ERP with no API and both go up.
Custom agent or off-the-shelf chatbot platform?
If you only need to answer frequent questions at low volume, a chatbot platform with AI, or Meta’s own built-in agent, may be enough and cheaper to start. A custom build pays off when:
- the agent has to take actions in your systems, not just reply;
- your systems don’t fit the platforms’ standard integrations;
- you want control over data, behaviour and cost per conversation;
- you expect volume to grow and per-seat or per-conversation fees to get expensive.
For a broader view of AI beyond WhatsApp, see our guides on what an AI agent costs and AI agents for business (both currently in Spanish), and on MCP servers for connecting AI to your internal tools.
FAQ
How much does a WhatsApp AI agent cost?
A custom WhatsApp AI agent for a business costs between €6,000 and €12,000 to build. Monthly costs come on top: Meta’s message charges, AI model usage and a maintenance fee.
How long does it take to launch a WhatsApp AI agent?
Between 2 and 4 months from discovery to full volume. Integrations with systems that lack a good API and the length of the pilot are what stretch the timeline most.
Can I use the regular WhatsApp Business app?
Not for a custom AI agent. You need the WhatsApp Business Platform (API), accessed directly through Meta or through an approved provider, plus business verification.
Does Meta charge for the agent’s replies?
Yes. From 1 October 2026 Meta charges service messages per message, with the first 1,000 free each month per business phone number. Rates vary by country.
Can a WhatsApp AI agent take bookings or change orders?
Yes, if it is integrated with your booking system, online store or ERP. You set rules for what it can do alone and what needs a person to confirm.
Is a WhatsApp AI agent GDPR compliant?
It can be, if it is designed for it: disclose that users are talking to an AI, minimise and limit data retention, sign data processing agreements and control where data is processed.
Want a price for your agent?
At We Are Nubi we design and build custom AI agents that plug into the systems you already use. Tell us what your customers ask on WhatsApp and which tools you work with, and we will give you a cost and timeline range for your case. See our AI agent services or get in touch.
