Comparison
Evaluating AI Receptionist Vendors
Every provider in this category demonstrates well. The questions below are the ones that separate a polished demonstration from a deployment that still works in a year, and you should ask us all of them.
The five that matter most
- What happens when it does not understand
- Who owns your data and can you take it
- Can it write to your systems, not just read
- Who maintains the content after launch
- What the exit looks like
Test the Failure, Not the Success
Any vendor can show you a call that goes well. What distinguishes deployments is behaviour when things go wrong: an unclear question, a caller who interrupts, a request outside scope, a noisy line.
Insist on testing that yourself during the demonstration. Ask something the system was not set up for and watch whether it escalates cleanly, admits it cannot help, or invents an answer.
A system that invents answers is disqualifying, regardless of how impressive the rest of the demonstration was.
Questions Worth Asking Every Vendor
| Area | Ask this | What a weak answer looks like |
|---|---|---|
| Failure handling | What happens when it does not understand? | Vague reassurance rather than a described behaviour |
| Integration | Can it write to our diary, or only read? | “Integrates with everything” without naming your system |
| Data ownership | Can we export transcripts and knowledge base? | Hesitation, or terms that are not in writing |
| Language | Have you tested this with our actual callers? | A long language list with no testing offered |
| Maintenance | Who updates content when our prices change? | Unclear ownership, or an unexpected charge |
| Claims | Can you evidence that compliance statement? | Certification implied but not held |
| Exit | What happens to our data if we leave? | Not addressed in the contract |
| Results | What should we measure to know it worked? | Guaranteed percentages with no basis |
Signals Worth Noticing
Encouraging
- They describe what the system will not do
- They ask about your call mix before quoting
- They let you test failure cases yourself
- They put integration limits in writing
- They tell you when you should not proceed
Worth probing further
- Guaranteed percentage improvements
- Compliance claims without evidence
- Scripted demonstrations you cannot deviate from
- Pricing quoted before scoping
- No description of escalation behaviour
Applying This to Us
These questions apply to our proposals as much as anyone’s. We do not publish guaranteed improvement figures, we do not assert compliance certifications, and we scope before pricing, precisely because the alternative would fail the tests above.
Where a capability depends on your systems, we say so. Where language coverage is uncertain, we test before committing. Where the numbers do not support a deployment, we would rather say that than sell one.
Our approach to what the system may and may not say is set out under knowledge base management, and escalation design under emergency escalation.
Questions People Ask
How many vendors should we evaluate?
Two or three is usually enough. Beyond that the differences become marginal relative to the time spent.
Should we run a paid pilot?
Often worthwhile for larger deployments. It reveals things no demonstration will.
What contract length is reasonable?
Long enough to justify setup effort, short enough that a poor fit is escapable. Terms should be explicit before you sign.
How do we compare pricing?
Insist on total cost including setup and ongoing support, not just the headline usage rate.
What if we are already committed elsewhere?
These questions are still worth asking of your current provider, particularly the ones about data export.
Ask Us the Difficult Ones
Bring this list to the demonstration. If we cannot answer something well, that is worth knowing before you commit.