Choosing a WhatsApp AI sales assistant: 8 criteria

Short answer

inTusell is an AI sales assistant that answers customer messages on WhatsApp, Instagram and website chat, and completes appointments and orders inside the conversation. It is software (SaaS), not a medicine, spray or health product. This article explains which questions to ask when choosing a tool of this kind. The criteria apply to any tool; you can evaluate inTusell with the same questions. If you want a short read on what inTusell is, see the What is inTusell? page.

A distinction up front: three different jobs are sold under the name "AI assistant". One is automation tools that build rules and flows (e.g. ManyChat, Botpress). Another is shared-inbox and WhatsApp API platforms that a team works in together (e.g. WATI, respond.io, Supsis). The remaining one is sales assistants that write to the customer and move the sale or the appointment forward themselves. The criteria below are written mainly for the last group, but they help when comparing the others too.

1. Which channels does it work on?

It has to be where your customer writes. A tool that covers only one channel leaves messages on the other channel unanswered. Ask: are WhatsApp, Instagram and website chat supported together, and are conversations collected in one place?

Example: the number of channels you can connect on ManyChat depends on the plan: up to 2 channels on the free and Essential plans, 3 on Pro, and all eligible channels on Business and Advanced (ManyChat pricing page, accessed 29.10.2026). inTusell covers WhatsApp, Instagram and a web chat widget added to your site with one line of code, and collects conversations in one inbox.

2. Where does it get its information?

Everything the assistant says is said on your behalf. So you must be able to define the knowledge source. Ask: can I upload PDFs, spreadsheets and web addresses? What happens if sources contradict each other? What stops it from making up a price?

In inTusell the knowledge base is built from PDF, Word, Excel, images and web addresses; amounts in replies come only from a defined source; if there is a contradiction the panel warns you and the assistant gives no figure on that topic and hands it to your team.

3. Does it complete the appointment and the order inside the conversation?

Many tools answer the question but send the job to a form link. The customer may give up at that point. Ask: does it offer an open slot and book the appointment? Does it send reminders and confirmations? Does it open the order and take the customer to the payment step?

inTusell offers an available slot and sends appointment reminders and confirmation messages; it opens the order and sends the customer an order portal link, and your own payment link can be shared.

4. When and how does it hand over to a person?

Not every conversation should be left to AI. Ask: what does the assistant do when it is unsure? When your team writes, does the assistant go quiet, or does it keep sending replies on top? After handover, can the team see the conversation history?

In inTusell, if your team writes, the assistant waits in that conversation; on handover a notice goes to the customer and an internal note is left for the team.

5. How are personal data and regulation handled?

In Turkey a customer message is personal data. Ask: is a KVKK (Turkish data protection law) privacy notice given when a new customer makes contact? In sensitive fields (health, law, education) is the assistant's boundary defined?

When inTusell makes contact with a new customer it sends a separate KVKK privacy notice; in clinics it is set up not to diagnose and not to give treatment guidance, and it does not give legal opinions in a law firm. For schools, the advertising rules for private educational institutions of the Turkish Ministry of National Education (MEB) also apply; every piece of information the assistant gives should stay limited to content the school management has approved. Compliance with this regulation is the institution's responsibility, not the tool's.

6. Can I observe before I release?

Giving a new assistant full authority from day one is risky. Ask: can I approve replies before they are sent? Is there a trial stage such as a shadow mode?

In inTusell, behaviour modes are split into Live, Approval, Shadow and Pause; in a new setup the health field starts in full approval, and the education field starts in the mode the owner chooses.

7. What is the pricing model based on?

The same "monthly price" can depend on different units. Ask: is the fee based on the number of contacts, agents, sessions or messages? Are WhatsApp's own message fees passed on separately?

Examples from the market. Price and plan information was taken on 29.10.2026 from the official pricing pages of the companies concerned; see their pages for the current list:

• ManyChat: sets its fee by the number of "Active Contacts"; every plan includes a monthly Active Contact limit and a charge per additional contact applies beyond it; there is a free plan and the paid plans offer a 14-day trial (pricing page).
• WATI: shows its Growth plan at 59 dollars a month billed annually, with 3 users included; it states that per-message charges apply separately and offers a 7-day free trial (pricing page).
• respond.io: works on a "pay only for the contacts you talk to" model; its Starter plan is 79 dollars a month billed monthly, with a 7-day free trial (pricing page).
• Supsis: scales its packages by the number of agents and chatbot sessions; it shows prices in Turkish lira (pricing page).
• inTusell: current plans and limits are published on the pricing page; no amounts are given in this article.

When comparing, equalise the unit: work out, with your own volume, what each tool costs in total for 100 customer conversations.

8. Does it comply with WhatsApp's rules?

In October 2025 Meta added a section to the WhatsApp Business API terms that restricts AI providers and the distribution of general-purpose AI assistants on the platform; the change took effect on 15 January 2026. As its reason it said the API is designed for businesses to serve their customers (TechCrunch, 18.10.2025; accessed 29.10.2026). An assistant that serves a business's own customers is the use this framework targets; still, before buying a tool, ask whether it works through the official WhatsApp Business API and through which provider it is connected. For current terms, see the text Meta publishes.

Extra questions for sensitive fields such as schools and clinics

In addition to the general criteria, ask these:

• School: On topics such as enrollment fees, quotas or success, does the assistant rely only on text the school has approved? When a figure in the knowledge base becomes outdated, who notices? Where is the personal data boundary drawn when sharing information about a parent, student or child? Because MEB private education regulation limits advertising and promotion language, there should be a rule that stops the AI from making promises such as "guaranteed" or "certain success". This is why approval mode helps in schools, at least at the start: the school manager sees the reply before it goes out.
• Clinic: What happens if the assistant starts to diagnose or recommend treatment? In inTusell this boundary is a system rule, but it is not an absolute guarantee; it is right to verify the behaviour with your own test conversations. In our own records, this capability is also marked as "partial".
• Both fields: The assistant should not offer a service your own business does not provide (e.g. an on-site visit, a callback by phone). inTusell offers these options only if the business provides them.

Three common mistakes when choosing

1. Watching the demo with one easy question. Every tool looks good on an easy question. The hard ones are a product whose price is not in the table, an angry customer, a half-finished appointment and an off-topic message. Try those.
2. Comparing price by the monthly amount alone. As the examples above show, the unit can be a contact, an agent, a session or a message. The same amount gives different results at different volumes.
3. Not counting the setup effort. Preparing the knowledge base, collecting frequently asked questions and reviewing replies in the opening weeks of use takes time. Ask how much support each tool gives you in this work and whether ready-made sector templates exist.

Decision table

How do you try it?

Write the same five questions to each candidate tool in real customer language: a price question, an appointment request, a complaint, a question that is not in the knowledge base, and an off-topic message. Put the accuracy, tone and handover decision of the answers side by side. Whichever tool needs fewer corrections for the work your business does is the one that suits you. Do not leave the trial to one person: have three people who represent sales, customer service and management ask the same questions and read the results together. Watching a real message flow in shadow mode for a week teaches far more than a single demo call. When deciding, look not at the number of features but at how the tool answers the ten questions your own customers ask most often.

To try inTusell with these questions, see the pricing page and the What is inTusell? page.

Price and plan information was taken on 29.10.2026 from the official pricing pages of the companies concerned. Sources: ManyChat, WATI, respond.io, Supsis.