How to Choose an AI Agent for WhatsApp: A 12-Question Checklist
A 12-question checklist and a criteria table to compare AI agent platforms for WhatsApp, from the official channel to the handoff to your team.

Marlos Carmo
Founder of Tolky
Updated on November 16, 2026

Share
Choosing an AI agent for WhatsApp means comparing vendors on 12 practical criteria: whether the channel is official, whether the AI answers from your content, whether it knows when to stop and call a person, whether it acts inside your systems, and whether you can test everything before going live. Any vendor can run a polished demo. The difference shows up in operation, in the second month.
This checklist is for people building a short list of platforms who need objective questions to bring to meetings. Use the table as a script and ask each vendor to show every answer on screen.
Updated in November 2026.
Why a checklist and not a feature list?
A checklist works better because a feature list rewards whoever has more items, while what matters is who delivers the right items for your operation. Two vendors can both say "we have AI, CRM and WhatsApp" and get opposite results once volume grows.
The underlying question is the same as in what an AI agent is: does the system only answer, or does it also look things up, record data and act within rules? The more the answer is "acts, with limits and with history", the closer you are to a conversational CRM.
What are the 12 criteria, in one table?
The 12 criteria cover channel, answer quality, control, integration and cost. The table summarizes what to ask and the warning sign for each.
| # | Criterion | What to ask | Warning sign |
|---|---|---|---|
| 1 | Official channel | Does it run on the official WhatsApp API? Do I have to change numbers? | A solution that relies on an unofficial number |
| 2 | Training on your content | How does the AI learn from my documents and how do I update it? | Training only with fixed Q&A pairs |
| 3 | Scope limits | What does the AI do when it does not know? | "It answers anything" |
| 4 | Handoff to your team | Does the person receive the history and a summary? | Customer repeats everything when transferred |
| 5 | Integrations | Which systems does the AI read and write? | Integration only through manual export |
| 6 | CRM and history | Is the customer history unified across channels? | Each channel with its own inbox |
| 7 | Test before publishing | Can I simulate conversations before going live? | You can only test with real customers |
| 8 | Metrics | Can I see resolution, response time and NPS? | Reports with volume only |
| 9 | Governance | Access profiles and an activity log? | Everyone with the same access |
| 10 | 24-hour window | How does it handle templates and campaigns? | Ignores WhatsApp rules |
| 11 | Other channels | Do web chat, Teams and voice join the same operation? | One product per channel |
| 12 | Pricing model | Do I understand what I pay and what happens if volume grows? | Price on request only, with no criteria |
The sections below give each criterion its right question and what a good answer contains.
Criteria 1 to 4: the core of the conversation
The first four criteria decide whether the conversation is safe, correct and human.
1. Is the channel official? A good platform runs on the official WhatsApp Business API and is clear about how the number is connected. Ask whether you keep the number you already use and whether the phone app keeps working. For context on the channel, see WhatsApp Business API: benefits for your company.
2. Does the AI learn from your content? The answer must come from your documents, policies and prices, not from generic knowledge. Ask how files come in, how a piece of content is updated and who approves it. The topic has its own post: how to train an AI agent on your knowledge base.
3. Is there a scope limit? A good answer describes what the AI does when the question is off topic: deflect, admit it does not know, or call a person. Be wary of any promise that the AI "answers everything". See also AI governance in customer service.
4. Does the handoff carry context? Ask to see the agent screen at the moment of transfer. The history, a summary and the reason for the handoff must be visible. The rules for when to hand off are in AI to human handoff.
Criteria 5 to 8: integration, testing and results
These criteria show whether the AI becomes an operation or stays an ornament on the channel.
5. Integrations. Ask which systems the AI reads and writes during the conversation. At Tolky, the native integration is with HubSpot, and there are connectors for about 20 tools, such as Pipedrive, Slack, Gmail, Google Sheets, Stripe and Notion. If your system is on no vendor's list, ask how it can be connected. The post AI without integration becomes an FAQ explains why.
6. CRM and history. A single history keeps the customer from being treated as a stranger on every channel. Check whether the platform has its own CRM or depends on another system to store the relationship.
7. Simulation before publishing. This is the cheapest criterion to verify and the one that best separates vendors. Ask to create a test conversation with a difficult customer and see the answer before any real customer does.
8. Metrics. Look for response time, AI resolution rate and NPS (Tolky measures NPS, not CSAT). To define what to measure, see AI resolution rate and AHT and first response SLA.
Criteria 9 to 12: control, channel rules and price
The last four criteria prevent surprises after you sign.
9. Governance. Ask about access profiles and an activity log with export. If your sector requires more, such as SSO, anonymization or configurable retention, confirm in writing whether it is available instead of assuming.
10. The 24-hour window. WhatsApp limits free-form messages to a window after the customer's last message. Outside it, approved templates apply. The vendor must handle this without you having to think about it. See 24-hour window and templates in campaigns.
11. Other channels. Does your customer also use your website, your company Teams or the phone? Tolky runs WhatsApp, web chat, Microsoft Teams and voice on the same history. Instagram is coming soon; SMS and email are not channels of the platform.
12. Pricing model. Understand the billing unit and what changes with volume. Tolky publishes its plans at /precos, with a free plan to start. For an overview of costs, see how much it costs to implement an AI agent.
How do you score and compare vendors?
Give each criterion a score from 0 to 2 (0 does not meet it, 1 partly meets it, 2 meets it and was demonstrated) and add them up. The total runs from 0 to 24. More important than the sum is having no zero on criteria 1, 3, 4 and 7, because they protect your customer and your brand.
Two precautions avoid bad decisions. First, score 2 only for what was shown on screen, not what was promised in a presentation. Second, run the same script with every vendor, in the same order. To compare specific platforms, see best AI agent platforms for WhatsApp in 2026 and the alternatives page.
Which common mistakes derail the choice?
The most common mistakes are deciding on a polished demo, ignoring the cost of operating after launch, and leaving the support team out. The table shows how to avoid each one.
| Common mistake | Why it happens | How to avoid it |
|---|---|---|
| Deciding on the demo | The demo uses questions the vendor chose | Bring five real situations from your operation |
| Ignoring the handoff to the team | Focus stays on the AI's answer | Ask to see the agent screen at transfer |
| Training once and forgetting | The content changes and the AI keeps the old version | Define who updates the content and how often |
| Forgetting the 24-hour window | The test only uses conversations with an active customer | Also simulate contact outside the window |
| Choosing only on entry price | The quota does not cover real volume | Estimate conversation volume before comparing plans |
Another frequent mistake is treating the pilot as final. Start with a narrow topic, such as a duplicate invoice or order status, measure for a few weeks and only then widen the scope. That reduces risk and gives the support team time to trust the AI.
How does Tolky answer this checklist?
Tolky is a conversational AI CRM platform: AI agents that serve customers on WhatsApp, know the customer history and hand off to the team with context. Against the criteria above:
- Channel: official WhatsApp API, without changing numbers.
- Training: the AI uses the company's knowledge base and stays within the defined scope.
- Testing: simulations let you rehearse conversations before publishing.
- Team: the operator pauses the AI with one click and takes over with history and summary.
- Channels: WhatsApp, web chat, Microsoft Teams and voice, with a unified history.
- Management: NPS, access profiles and an activity log.
Platform-wide figures are: more than 5 million AI messages per month, more than 67% AI resolution, 60% lower response time and 99.9% availability. These are platform data, not a guaranteed result for any one customer. See the WhatsApp support solution.
Frequently asked questions
How many platforms should I compare?
Between three and five is usually enough. Beyond that, the script becomes expensive to run and the comparison loses depth. Use the checklist to cut the list before asking for demos.
Do I need to change my number to use an AI agent on WhatsApp?
Usually not. The point is to confirm with the vendor how the number is connected through the official API and whether the phone app stays available. At Tolky, coexistence keeps the number on the WhatsApp Business app and also makes it available on the platform.
Which criterion matters most?
The handoff to the team with context and the simulation before publishing. An agent that errs and does not pass the conversation to a person harms the customer; one that cannot be tested makes you discover the error in production.
Does the AI agent replace the support team?
No. The AI resolves the repetitive work and the team handles what needs judgment. The goal is AI with humans, with a transition that preserves context.
How do I test before signing?
Bring five real situations from your operation, including a complaint, an off-topic question and a request that needs one of your systems. Ask the vendor to run them on the platform and watch the answers and the handoff to a person.
Is it worth starting on a free plan?
Yes, to validate the channel and the AI's tone at low risk. Afterward, assess the quota you need for your real volume.
Next step
If you want to run this checklist on your own operation, bring your real cases and watch Tolky answer on screen. Shall we talk? Talk to the Tolky team or see the plans.
Share
Tags
- how to choose an AI agent for WhatsApp
- AI agent checklist
- AI platform for WhatsApp
- #conversational AI CRM
- WhatsApp AI agent
- compare customer service platforms
- #WhatsApp Business API
- criteria for choosing an AI chatbot
About the author

Marlos Carmo
Founder of Tolky
Marlos Carmo is an AI entrepreneur and founder of Tolky, the conversational-era infrastructure and AI CRM that unifies intelligent service, multi-channel support (such as WhatsApp and voice), live CRM, and operational intelligence in a single ecosystem. He is a finalist for the SXSW Innovation Awards and a member of Francesco's Economy, a global network of young entrepreneurs focused on innovation and social impact. He works connecting Artificial Intelligence and digital transformation in projects for large organizations.
Next step
Want to bring this to your operation?
Tati, Tolky's AI assistant, understands your scenario in a few questions, suggests the best path and connects you with someone from the team, with full context.


