Skip to content
Buying kit · RFP

RFP template for anAI agent on WhatsApp

Questions to ask any conversational AI CRM vendor, with what a good answer includes. Download it, adjust the weights and compare everyone with the same yardstick.

11

sections

40

questions

CSV, Markdown and PDF

For a PDF, choose Save as PDF in the print dialog.

RFP (request for proposal) is the document in which a company describes what it needs and asks each vendor to answer the same questions, so the proposals can be compared side by side.

Reviewed on 11/16/2026

How to score

The same yardstick for everyone

  1. 01

    Weight from 1 to 3

    Weight 3 for what is decisive in your operation, 1 for what is nice to have.

  2. 02

    Score from 0 to 3

    0 does not meet it, 1 meets it with a workaround, 2 meets it, 3 meets it and showed it working.

  3. 03

    Always evidence

    Ask for a demo, document or link. An answer without evidence is worth at most 1.

  4. 04

    Weighted total

    Add up weight × score for each question and compare the totals, looking first at the weight 3 items.

01

Channels

The agent needs to be where your customers already talk, with an official connection.

  1. 1.1

    Which channels are available today and which are still in development?

    What a good answer includes

    A list that separates what is in production from what is on the roadmap, with a date only when there is a commitment.

  2. 1.2

    Does the WhatsApp connection use the official API? Directly with Meta or through a partner?

    What a good answer includes

    Official WhatsApp Business API, with the provider's name and who pays Meta's fees.

  3. 1.3

    Does the same agent serve several channels with the same contact history?

    What a good answer includes

    A demo of one contact with conversations on different channels brought together in the same record.

  4. 1.4

    How do outbound campaigns and message templates work?

    What a good answer includes

    Sending of Meta-approved templates, contact segmentation, sending limits and a delivery and reply report.

02

AI and answer quality

This is where most AI projects in customer service succeed or fail.

  1. 2.1

    How does the agent avoid making up answers?

    What a good answer includes

    Answers grounded in the knowledge base, an explicit rule for when it does not know and a real example of a refusal.

  2. 2.2

    Can I test the agent before it goes live?

    What a good answer includes

    Simulations with scenarios and success criteria, with a report, in an environment separate from production.

  3. 2.3

    Can I see why the AI gave a specific answer?

    What a good answer includes

    A trail for each answer: instructions used, knowledge base excerpts and calls to external systems.

  4. 2.4

    Which AI models are used and can I choose?

    What a good answer includes

    The providers' names, model choice per agent and what happens when a model does not reach the confidence level.

  5. 2.5

    How is quality measured after the agent goes live?

    What a good answer includes

    AI resolution rate, satisfaction surveys such as NPS and conversation review, with the definition of each metric.

03

Knowledge base

The agent is only as good as what it knows about your company.

  1. 3.1

    What content formats does the knowledge base accept?

    What a good answer includes

    Files, links and text, with the size and quantity limits of each plan.

  2. 3.2

    How is the knowledge base updated when a policy changes?

    What a good answer includes

    Updates made by your team, without relying on the vendor, taking effect immediately or within a stated time.

  3. 3.3

    Does the agent query systems in real time, such as orders or invoices?

    What a good answer includes

    Integration through an API or tools, with an example of a query to a system similar to yours.

04

Handoff to people

Good AI customer service knows when to call someone in.

  1. 4.1

    When and how does the conversation pass to a person on the team?

    What a good answer includes

    Rules configurable by topic, customer request or AI uncertainty, with the full history for whoever takes over.

  2. 4.2

    Can the operator pause the AI and take over at any time?

    What a good answer includes

    Pause and resume per conversation, without the customer having to switch channels.

  3. 4.3

    How are conversations distributed among the people on the team?

    What a good answer includes

    Queues by department, automatic distribution and the reason for each assignment visible.

05

Integrations

The agent needs to read and write in the systems you already use.

  1. 5.1

    Which integrations are native and which depend on a connector or development work?

    What a good answer includes

    A list that separates native, connector and API, without calling native what requires technical setup.

  2. 5.2

    Is there a documented API and webhooks?

    What a good answer includes

    Public documentation, usage examples and the list of events available through webhooks.

  3. 5.3

    How does the integration with our current CRM work?

    What a good answer includes

    Which data goes back and forth (contacts, deals, tickets), when, and who maintains the integration.

06

Security and LGPD

Customer conversations carry personal data and call for clear control.

  1. 6.1

    Where are the application and the data processed and stored?

    What a good answer includes

    Cloud providers, database and subprocessors, including the AI providers.

  2. 6.2

    How is data protected in transit and at rest?

    What a good answer includes

    Encryption in transit (TLS) and what is encrypted at rest, with the algorithm used.

  3. 6.3

    What security certifications does the vendor have?

    What a good answer includes

    Current certifications with scope and date, or a clear answer that it has none.

  4. 6.4

    How are LGPD data subject requests handled?

    What a good answer includes

    A channel for requests and tools to find, correct and delete a contact's data.

  5. 6.5

    Is conversation data used to train AI models?

    What a good answer includes

    An answer per AI provider, with the contractual terms for data use and retention.

07

Governance and audit

Who changed what, and how to undo it.

  1. 7.1

    How are access profiles defined?

    What a good answer includes

    Permissions by module and by action, with the option to limit a person to their own conversations.

  2. 7.2

    Is there an audit log? Can it be exported?

    What a good answer includes

    Actions recorded by person, module and date, with export in an open format.

  3. 7.3

    How does a change to the agent go to production and how is it undone?

    What a good answer includes

    A draft separate from production, comparison, backup on publish and restore of the previous version.

  4. 7.4

    What login methods exist (SSO, 2FA, social login)?

    What a good answer includes

    A straightforward list of what exists today and what does not.

08

Implementation

Timeline, effort from your team and what happens in the first week.

  1. 8.1

    What are the implementation steps and who does each one?

    What a good answer includes

    A plan with steps, owners on both sides and your team's estimated hours.

  2. 8.2

    What is needed to start a pilot?

    What a good answer includes

    Minimum scope, channel, initial knowledge base, success target and review date.

  3. 8.3

    How is the team trained to operate the platform?

    What a good answer includes

    Training, documentation and support material, including for people who join later.

09

Pricing and billing model

The billing model defines the cost as volume grows.

  1. 9.1

    Is billing per user, per contact, per conversation or per message?

    What a good answer includes

    A well-defined billing unit, public pricing or a written proposal and an example with your volume.

  2. 9.2

    What is included and what is charged separately?

    What a good answer includes

    A list of extra items, such as Meta fees, voice, implementation and integrations.

  3. 9.3

    What happens if volume exceeds the contracted amount?

    What a good answer includes

    The overage price, a notice before charging and the option to change plans.

  4. 9.4

    Is there a free plan or trial period?

    What a good answer includes

    Trial terms and limits and what happens to the data at the end.

10

Support

When something goes wrong, who responds and how fast.

  1. 10.1

    Which support channels and hours are included in the plan?

    What a good answer includes

    Channels, hours and response time for each plan, in writing.

  2. 10.2

    Is there an account manager or regular check-ins?

    What a good answer includes

    Who follows the account, how often and what is reviewed at each meeting.

  3. 10.3

    How are product changes and incidents communicated?

    What a good answer includes

    A public changelog or direct notice, with recent examples.

11

References

What other customers have experienced is worth more than the sales pitch.

  1. 11.1

    Which customers similar to our company use the solution?

    What a good answer includes

    Names that can be cited, industry and how long they have used it.

  2. 11.2

    Can we talk to a reference customer?

    What a good answer includes

    Contact with a customer in the same industry or with a similar volume.

  3. 11.3

    Are there published case studies?

    What a good answer includes

    Cases with the context, what was done and how the result was measured.

What about Tolky?

Tolky answers this RFP like any other vendor. To get ahead, the official answers are in the security questionnaire and on the facts page, and pricing is public.

Reviewed on 11/16/2026
Frequently asked questions

Questions about the RFP

Send us your RFPand we will answer it

In the demo, we answer question by question and show each answer working in the dashboard.