Your company wishes to take the turn of generative AI to gain efficiency. The management has decided to launch a call for proposals for an AI chatbot dedicated to customer support.

You are in charge of drafting the call for proposals and you may risk spending weeks deciphering online offers, organizing RFIs, and multiplying exchanges with providers.

This article offers you a structured roadmap to prepare your RFP with more clarity. To go further, download our complete 10-page guide, which contains a step-by-step guide for your RFP.

Objectives and ROI of a Chatbot Call for Proposals: What KPIs to Include?

Implementing an AI chatbot is not just about following a technological trend: it is a strategic decision with measurable impacts.

• Reduction of operational costs
  A bot can handle up to 40% of basic requests.
  
  

• Smoothing of peak loads
  You no longer need to hire temporary agents for peak seasons: the chatbot handles low-value requests 24/7 and smooths out demand.
  
  

• Improvement of CSAT (Customer Satisfaction Score)
  Less waiting means more satisfaction. By automating factual requests (order status, schedules, procedures), you reduce the average wait time and increase your CSAT score.
  
  

• Optimization of the customer journey
  Beyond support, the bot can qualify a lead, offer an interactive FAQ, or guide to scheduling an appointment, ensuring a seamless end-to-end experience.

What is a chatbot with generative AI?

An AI chatbot is an intelligent conversational agent based on a LLM (Large Language Model) like ChatGPT (🇺🇸) or Mistral AI (🇫🇷).

These agents can be configured to follow predefined customer journeys, connect to your systems to perform actions, and answer questions from a knowledge base (online FAQs, PDFs, CSVs…).

The different types of flows (FAQ, Transactional, Scripted)

1. Answers to various questions from a knowledge base: the agent pulls information from your documents to provide personalized and contextualized responses.
   
   

2. Transactional/API: the agent performs actions via API calls to your CRM, ERP, or internal tools (order status, billing, authentication).
   
   

3. Scripted flows: here, the bot follows branches with predefined responses for total control. This is the technology of first-generation bots, which ensures that each message is controlled. However, you can integrate AI to make these journeys more flexible and efficient.
   
   

4. Human escalation: switch to a human agent via LiveChat, ticketing, or web callback.
   
   

5. Rich media: image carousels, PDFs, videos to enrich the experience.

Deployment Channels

Web widget, mobile application (SDK/webview), WhatsApp Business, Facebook Messenger, SMS…

Each channel has its technical constraints; clearly mention them in your RFP (section 2.3 of the guide).

For a complete overview of the types and channels, refer to section 2 of the guide, which details each use case with examples and diagrams.

Typical Structure of an AI Chatbot Specification Document (CCTP): 6 Blocks to Copy

To ensure a consultation that is both clear and comparable, structure your RFP around these six essential blocks. Each plays a key role: aligning your teams, guiding providers, and securing your decision.

1/ Project Presentation & Objectives

Why? Providing context avoids generic offers. Moreover, by communicating your support volumes, you allow suppliers to adjust their commercial offers and establish a reliable ROI calculation.

Content: your support organization, current volumes (web traffic, emails, calls, existing chatbot) and clear objectives (cost reduction, CSAT, scalability).

Tip: a quantified objective better guides responses ("reduce TTR by 20%").

2/ Expression of Functional Needs

Why? Defining precisely what your bot should do prevents scope creep.
Content:

• Use Cases: detailed scenarios (FAQ, authentication, invoices, ticket searches, escalation), key steps and systems involved.

• Types of Flows: FAQ, transactional/API, scripted flows, rich media, escalation.

• Tip: accompany each use case with a concrete example to illustrate the "journey".

3/ Technical Requirements (Hosting, GDPR, API)

Why? Ensure feasibility and compliance from the RFP phase.

Content:

• Hosting: GDPR SaaS vs on-premise, server location.

• Security & GDPR: encryption (TLS/AES), consent, retention period, DPA.

• System Integration: APIs (REST, GraphQL, SOAP), authentication (OAuth2, JWT), access to specifications (Swagger/Postman).

• AI-agnostic: architecture capable of switching between multiple models or providers without major overhaul.

• Tip: involve the IT department as early as the drafting of the RFP to validate the connections to internal systems and avoid any unpleasant surprises.

4/ Selection Criteria and Scoring Grid

Why? Objectify comparison and limit arbitrariness.

Content:

• Functional: coverage of use cases, quality of responses, management of multi-turn context.

• Technical & UX: native connectors, no-code editor, dashboards.

• Support & Guidance: SLA (response time), training, Customer Success.

• Costs: build (set-up, tests), license, TCO over 3 years.

• Tip: assign a weight to each criterion in an Excel grid for transparent scoring.

5/ Methodology and Planning

Why? Harmonize the format of responses to facilitate evaluation.

Content:

• Response plan (summary, intro, functional, technical, planning, budget).

• Submission format, key dates (RFI, demos, closure), selection committee.

Tip: request a demonstration schedule.

6/ Annexes & Templates

Why? Provide ready-to-use resources and standardize documentation.
Content: glossary, conversation diagrams (Visio/Draw.io), Excel scoring grid, DPA model, technical checklist
Tip: attach an example of a diagram to clarify the expected format.

Pitfalls to Avoid in Your Consultation: GDPR, API and Use Cases

• Not defining use cases precisely

Without detailed scenarios, the scope remains vague and open to interpretation. Detail each journey to be able to clearly communicate your expectations in demos.
Consequences: Too generic demos, not aligned with your priority needs. Risk of suboptimal choice and surprises regarding implementation cost.
Solution: map your dialogue journeys with tools like Draw.io/Whimsical or use the simple and universal template provided in the appendix of our guide.

• Not communicating your KPIs from the start

Ticket volumes, web traffic, average resolution time (TTR), escalation rate, CSAT, automation rate… Without concrete numbers, providers cannot offer a realistic pricing nor provide you with a reliable ROI calculation.

• Ignoring sovereignty and GDPR compliance

• Localization & portability of models

Demand hosting on servers in France or the EU and an architecture designed to support multiple AI model providers (for example, switching from a cloud LLM API to a self-hosted model) without having to redo everything. This avoids vendor lock-in, controls your costs, and ensures your compliance with regulatory changes. The other points are detailed in section 3.2 of our guide.

• Omitting to involve the IT department from the beginning

Not validating API integrations leads to unpleasant technical surprises. Involving the IT department before sending the RFP ensures the availability of connectors and network compliance.