A conversational assistant can serve many different purposes: handling incoming calls, qualifying requests, automating part of your support, improving service availability, or relieving teams of repetitive tasks.
But when it comes to consulting the market, many companies fall into the same trap: they evaluate solutions based on highly marketing-oriented demonstrations, without a precise framework for comparing offers.
The result: poorly defined projects, gaps between the promise and the reality, or solutions that are technically attractive but difficult to deploy at scale.
To avoid this, an effective set of specifications must cover three dimensions: business use cases, technical requirements, and operational performance criteria. Here are the 15 essential criteria to include.
1. The use cases covered
The first point to clarify: what should the conversational assistant actually do?
The specifications must specify the priority use cases : handling incoming calls, lead qualification, appointment scheduling, routing, answering frequently asked questions, customer support, follow-ups, or others.
A good provider must be able to demonstrate how their solution addresses your actual use cases, and not just demonstration scenarios.
To be specified in the requirements:
- priority use cases
- estimated volume per use case
- criticality level
- language(s) involved
2. The channel involved
Not all conversational assistants cover the same channels with the same level of maturity.
You need to specify whether your requirements concern:
- phone
- website
- messaging apps
- omnichannel
An voice assistant and a chatbot are not designed in the same way. The channel impacts the user experience, technical integration, expected quality, and performance indicators.
3. Natural language understanding
This is one of the most visible criteria, but also one of the most poorly evaluated.
Beyond the "wow" factor of a demo, you must verify the solution's actual ability to understand varied phrasing, hesitations, accents, incomplete or ambiguous requests, and to handle deviations from the ideal scenario.
Questions to ask:
- examples of comprehension based on your real-world cases
- handling reformulations
- robustness in noisy voice environments
- handling out-of-scope intents
4. The quality of the user experience
A high-performing conversational assistant must be more than just technically capable. It must also provide a fluid, clear, and reassuring experience.
The specifications must include criteria for:
- naturalness of responses
- clarity of instructions
- ability to guide the user
- handling misunderstandings
- ability to hand off seamlessly
An effective solution is one that users actually want to use.
5. The level of customization
Not every company needs a fully bespoke assistant, but few projects succeed with a completely standardized tool.
You must specify the level of customization expected:
- tone of voice
- business scenarios
- specific vocabulary
- routing rules
- qualification criteria
- brand identity
This helps distinguish rigid solutions from truly adaptable platforms.

6. Ease of configuration and scalability
A conversational assistant is never set in stone. It must evolve with your offerings, your processes, your peak activity periods, and your business priorities.
The specifications should therefore evaluate:
- team autonomy in modifying a scenario
- simplicity of updates
- administration interface
- time required to deploy a change
- whether or not technical intervention is needed
This is a key criterion for avoiding constant reliance on the service provider.
7. Integrations with your existing environment
An AI voice agent isolated loses much of its value.
The specifications must list the systems to be connected:
- CRM
- ticketing tool
- calendar
- telephony
- document database
- ERP
- internal tools
You must also specify the expected level of integration: read-only, data writing, real-time synchronization, automatic action creation, or routing to the right person.
8. Security and hosting
As soon as a conversational assistant processes customer data, internal data, or sensitive information, security is becoming a non-negotiable requirement.
To be defined in the request for proposal:
- hosting model
- data residency
- encryption
- access management
- logging
- retention policy
- GDPR compliance
This is often a deciding factor for IT, security, and legal departments.
9. Data management and compliance
A conversational project quickly raises very practical questions: where does the data go? who has access to it? how long is it kept? can it be reused?
The specifications must ask the provider:
- what data is stored
- how it is used
- who owns it
- what guarantees for deletion or export exist
- what compliance measures are in place
10. The ability to handle handoff to a human agent
A conversational assistant should not claim to handle everything.
A key criterion is to evaluate its ability to detect the limits of automated processing and to efficiently transfer to an agent or an appropriate department.
Be sure to ask about:
- escalation rules
- contextualized transfer
- history retention
- conditions for human takeover
- handling of sensitive or urgent cases
11. Available performance indicators
If you cannot monitor the assistant, you cannot improve it.
The specifications must define which indicators must be accessible :
- resolution rate
- automation rate
- conversion rate
- reasons for contact
- comprehension quality
- average handling time
- satisfaction
- missed calls avoided
- qualified leads or appointments booked
A good service provider must provide rigorous KPI tracking as well as a clear view of performance, not just a decorative dashboard.
12. Service level commitments
The project doesn't end with deployment.
You must include criteria in your request for proposals regarding:
- platform availability
- response time
- technical support
- maintenance
- incident management
- onboarding support
This allows you to compare offers that might otherwise look similar on paper but differ significantly in execution.

13. The business model
Price alone is not enough. You need to understand how the solution is billed and what the offer actually covers.
To be clarified:
- implementation cost
- subscription
- usage-based cost
- cost per call or interaction
- cost of updates
- support costs
- potential hidden costs
A well-written specification document should allow you to compare the total cost of ownership, not just the entry price.
14. Provider references and maturity
A conversational assistant might look convincing in a demo, but the real question remains: has the provider already delivered this type of project under conditions similar to yours?
To ask for:
- client references
- comparable use cases
- feedback and experience
- level of support
- ability to deploy at scale
This criterion is particularly reassuring for strategic projects or those involving high volumes.
15. Deployment methodology
Finally, the specification document must evaluate the provider's ability to transform a solution into fully operational project.
To be specified:
- project stages
- timeline
- resources required from the client
- expected deliverables
- pilot phase
- success criteria
- continuous improvement plan
Good technology without a clear deployment method often leads to frustration and little value.
Conclusion
A well-managed tender process should do more than just compare suppliers. It should secure a solution capable of meeting your business needs, integrating into your environment, and evolving over time.
By structuring your specifications around these 15 criteria, you give yourself the means to compare providers in a more objective, operational, and sustainable way.
The goal is not to choose the most impressive solution during a demo. It is to choose the one that will truly perform once deployed within your organization.
How many calls can your voice agent absorb?
Estimate in one minute the number of automated calls and the savings on your own volumes, with our ROI calculator.






