WhatsApp is the most widely used messaging channel in Spain: according to Statista data from 2025, 87% of adults use it daily. For an SME with a small customer service team, that means the bulk of customer enquiries arrives through a channel that, without automation, requires continuous manual attention. A WhatsApp Business customer service chatbot solves that bottleneck: it handles queries outside business hours, classifies and routes requests, and — when integrated with generative AI — answers complex questions about products, orders or services without human intervention.
The question almost every SME asks before getting started is the same: how much does it cost to implement? The honest answer is that the range is wide, because the price depends on very specific technical and business decisions. In this article we break down the cost components, provide real indicative market ranges (not any provider's tariffs) and explain which factors make one project cost twice as much as an apparently similar one.
How a WhatsApp Business API chatbot works
WhatsApp, for commercial entities, is divided into two products: WhatsApp Business App (free, for sole traders or very low-volume businesses) and WhatsApp Business API (also called Cloud API since 2022, managed directly by Meta). Automated chatbots are only possible through the API, which requires registering a dedicated phone number and connecting via a Business Solution Provider (BSP) or directly with Meta.
The basic architecture of a functional chatbot includes four layers:
- Messaging layer: the connection to Meta's WhatsApp Cloud API, where messages are sent and received.
- Conversation engine: the logic that decides what to reply. This can be a simple decision tree, a natural language understanding (NLU) engine, or a large language model (LLM) with RAG over the company's catalogue.
- Backend integrations: connection to the CRM, ERP, e-commerce platform or ticketing system to query real-time data (order status, stock availability, customer history).
- Supervision and human escalation panel: an inbox where human agents take over conversations when the bot cannot resolve them.
The greater the complexity in each layer, the higher the implementation cost and the ongoing monthly cost.
Indicative market price ranges in 2026
The ranges shown below are market estimates compiled from prices published by BSP platforms, European integrators and comparative proposals gathered from specialist forums such as Chatbot Summit and GSMA events. They do not represent the tariffs of any specific provider.
| Project profile | Implementation cost (one-off) | Monthly recurring cost | What it includes |
|---|---|---|---|
| Basic bot with decision tree | €800 – €2,500 | €50 – €150/month | Predefined flows, static FAQs, human handover, no backend integrations |
| Bot with NLU and knowledge base | €3,000 – €8,000 | €150 – €400/month | Natural language understanding, intent classification, dynamic responses, 1-2 simple integrations (CRM or Sheets) |
| Bot with generative AI (LLM + RAG) | €8,000 – €20,000 | €400 – €1,200/month | Contextual responses over own catalogue, ERP/e-commerce integration, customer history, omnichannel escalation |
| Enterprise solution with advanced analytics | €20,000 – €50,000+ | €1,200 – €3,000+/month | Multilingual, multiple numbers and departments, guaranteed SLA, BI dashboards, Contact Centre integration |
Reference sources for the ranges: Sinch (European BSP, published prices 2025), Vonage/Ericsson Business Messaging, Gartner Market Guide for Conversational AI Platforms 2025, and comparative analysis of proposals from the Spanish market gathered at industry events.
Factors that push the price up or down
1. The BSP platform chosen
To connect to the WhatsApp Business API, it is necessary to do so through a Meta-authorised BSP (Business Solution Provider), or directly through Meta's Cloud API. The BSP platforms most commonly used in the Spanish market include Vonage, Infobip, Sinch, Twilio, MessageBird (now Bird) and Gupshup. Their prices vary considerably in terms of licensing model and cost per conversation.
Between 2024 and 2025, Meta made significant changes to the WhatsApp Business API pricing model. From November 2024, user-initiated service conversations became free with no monthly cap. From July 2025, template messages (marketing and utility) are billed per delivered message rather than per 24-hour conversation. Exact per-message costs depend on destination country and category (marketing, utility, authentication); for Spain, indicative references published by leading BSPs in 2025 place a marketing message at around €0.05–€0.07 and a utility message at around €0.01–€0.02. On top of this comes the BSP's margin, which typically increases the base price. Always check the latest rates in the Meta Business Manager panel, as tariffs are reviewed periodically.
2. Complexity of the conversation engine
A decision-tree bot (buttons, lists, predefined replies) can be configured in days with low-code tools such as Tidio, Kommo or even Meta Business Suite itself. The implementation cost is low, but the user experience is rigid: if the customer does not select exactly one of the available options, the bot gets stuck.
A generative AI bot based on an LLM and RAG over the company's documents (catalogue, FAQs, returns policy, pricing) can answer open questions in natural language. The implementation cost is 4 to 10 times higher, but the resolution rate without human intervention rises significantly. For an SME with a complex catalogue or a high volume of varied questions, the difference in return justifies the outlay.
If you want to understand how the AI layer behind these bots works, Summum IA's AI customer service chatbot service explains the architecture and selection criteria according to the use case.
3. Number and intensity of integrations
The difference between a bot that only answers generic questions and one that queries in real time the status of an order, checks whether a product is in stock or retrieves a customer's incident history is, essentially, a difference in integration. Each connection to an external system (ERP, CRM, e-commerce platform, ticketing database) adds between €800 and €3,000 to the implementation cost, depending on the quality of the target system's API and whether standard connectors exist or custom development is required.
4. Monthly conversation volume
The recurring cost scales with volume. An SME with 500 conversations per month has a very different BSP platform spend from a company with 10,000 monthly conversations. When sizing the project, it is important to project the expected volume with a realistic growth margin so as not to have to change providers six months in because the contracted plan is too small.
5. Maintenance, training and continuous improvement
An AI chatbot is not a finished product on the day of delivery. It requires periodic review of failed conversations, updating the knowledge base when the catalogue or commercial policy changes, and retraining or adjusting prompts when new questions arise that the bot does not handle well. This maintenance work typically represents between 15% and 25% of the implementation cost per year, and it is a budget line that many SMEs underestimate at the evaluation stage.
WhatsApp Business App vs. WhatsApp Business API: which one applies to you
Before starting a chatbot project, one point that frequently causes confusion needs to be clarified:
- WhatsApp Business App (free): for businesses with 1-4 people handling queries manually. It allows simple automatic replies (welcome and away messages), but does not support chatbots or advanced automated flows.
- WhatsApp Business API / Cloud API: for real automation. It requires a dedicated phone number (which cannot be the same one used with the App), business verification and connection via a BSP or directly with Meta. It is the only route to a genuine chatbot.
SMEs already using the App that want to scale to the API must manage the number transition carefully: Meta allows migration, but it involves temporarily losing access to the App during the process and following the Business Manager steps to verify the account.
Expected ROI and metrics for evaluating it
The return on a WhatsApp chatbot is not measured solely in cost saved per conversation. The metrics that matter most to operations directors at an SME are:
- Autonomous resolution rate (containment rate): the percentage of conversations the bot resolves without routing to a human agent. A well-configured bot with NLU or generative AI can reach 60%–80% for frequently asked question categories.
- Average first response time: with a bot, the first response time drops from hours (or days, outside business hours) to seconds, regardless of the time of day.
- Cost per handled interaction: divide the total monthly cost of the chatbot (platform + maintenance + Meta cost) by the number of conversations handled. Compare it with the cost of handling those same conversations manually.
- Qualified lead conversion rate: if the bot also acts as the initial commercial filter, measure how many leads captured via WhatsApp end up as opportunities in the CRM.
In projects with a volume exceeding 1,000 conversations per month, the chatbot payback period typically falls between 6 and 18 months, according to implementation data published by Sinch and Infobip in their 2025 reports. Below that volume, the argument is more about service quality (24/7 support) than cost savings.
Platforms and tools commonly used in the Spanish market
Without recommending any specific platform — the choice depends on the use case, volume and technology stack of each company — these are the tool categories most commonly used:
| Category | Representative examples | Usage profile |
|---|---|---|
| BSP with visual builder | Kommo, Tidio, Intercom | SMEs with low volume and decision-tree bots |
| Enterprise BSP | Infobip, Sinch, Vonage, Twilio | Mid-sized companies with high volume and SLA requirements |
| Conversational AI platforms | Botpress, WATI, Respond.io, Landbot | SMEs that want NLU without custom development |
| Custom development with LLM | Direct integration with model APIs (OpenAI, Anthropic, Mistral) + own RAG | Companies with complex catalogues or data privacy requirements |
Legal aspects you cannot ignore: GDPR and Meta policies
A WhatsApp chatbot that handles customer data (name, email, order history, queries) processes personal data subject to the General Data Protection Regulation (GDPR). This entails several specific obligations:
- Users must give explicit and informed consent before starting an automated conversation for purposes other than pure service (for example, if the bot is going to send marketing messages or newsletters).
- You must have a signed Data Processing Agreement (DPA) with the BSP you use and, where applicable, with Meta (which acts as data processor for the Cloud API).
- Conversation data stored on the BSP platform must have a defined retention policy. Meta stores messages for 30 days on its servers; BSP platforms may store them indefinitely if not configured otherwise.
- If the chatbot collects data from minors or health data, the obligations intensify.
Additionally, WhatsApp's commercial policies prohibit certain sectors and types of messages (weapons, illegal products, fraudulent investment schemes, gambling in non-permitted jurisdictions). Non-compliance can result in the suspension of the business number without prior notice.
To correctly size the AI layer of your chatbot and ensure that data does not leave the corporate perimeter, the Summum IA AI customer service chatbot team can help you choose the right architecture and GDPR-by-design compliant providers.
What you should demand from your provider before signing
Regardless of the provider chosen, there are ten questions that every operations director or technology manager at an SME should ask before approving a WhatsApp chatbot project:
- Does the project include a training and adjustment period after launch, or does the provider deliver the bot «as is» on go-live day?
- Who is the holder of the WhatsApp Business API number: my company or the provider? (It must always be the client company.)
- What happens if I want to change providers? Can I migrate the number and the flows?
- Where are conversations stored? On servers within the EU?
- Does the provider have a DPA available and has it signed Meta's?
- What autonomous resolution rate do they estimate for my use case, and on what basis?
- How is escalation to a human agent handled? Is an inbox included or does another tool need to be contracted?
- What is the per-conversation cost that Meta will pass on to me, separately from the provider's fee?
- Can the bot personalise responses by querying data from my ERP or CRM in real time?
- What performance reports does it deliver and how often?
Frequently asked questions
Can a small SME (5-10 employees) get a return from a WhatsApp chatbot?
Yes, as long as the volume of repetitive queries is high. A dental clinic, an estate agency, an optician or an e-commerce store with a catalogue of hundreds of references can recover a basic bot (€800–€2,500 implementation) in less than six months if it handles more than 300–400 conversations per month. The key is to identify the five most frequently asked questions and make sure the bot handles them well before thinking about advanced features.
Do I need a new phone number for WhatsApp Business API?
Yes, in practice. Although it is technically possible to migrate an existing WhatsApp number to the API, during the process you lose access to the App and all previous chats (which cannot be exported to the API). The most common and recommended approach is to get a new number (a data SIM or a VoIP line verifiable by SMS) dedicated exclusively to the automated customer service channel. The cost is minimal (€5–€15/month for a virtual line) compared to the rest of the project.
Can a WhatsApp chatbot send proactive messages (notifications, reminders)?
Yes, but with conditions. WhatsApp calls these messages «templates» and they must be approved in advance by Meta. They are permitted for transactional notifications (order confirmation, appointment reminder, shipping notice) but require the user to have given explicit opt-in. Marketing templates are also permitted but have a higher cost per conversation and a frequency limit to prevent spam. Using WhatsApp to send bulk messages without prior opt-in is a violation of Meta's policies that can result in the permanent suspension of the number.
Is it better to develop the chatbot from scratch or use a low-code platform?
It depends on the use case. For simple flows and a limited vocabulary (bookings, schedule enquiries, quote requests with fixed fields), a low-code platform such as WATI, Landbot or Respond.io delivers quick and cost-effective results. For chatbots that need to understand open questions, query internal systems or handle complex exceptions, custom development with an LLM and RAG over the company's own data is more efficient in the long run. The common pitfall is starting with low-code for a simple feature and, as requirements grow, trying to push the platform to its limits — that path usually ends in a costly migration.