Skip to content
Tool · 5 minutes

Voice AI Readiness Diagnostic — score your enterprise programme

Twenty questions across seven dimensions. The four survival dimensions — evaluation/QA, security, integration depth, operating model — are weighted ×4 because they reliably make or break enterprise voice AI at production scale. You get a defensible /100 score, a unit-economics sanity check, and a board-ready summary page you can paste into a steering deck.

1. Score yourself, honestly

0 = not started · 1 = partial or informal · 2 = in place, documented, named owner.

Business case

Whether the spend is anchored to intent-level economics rather than vendor cost-per-minute headlines.

  1. Q01
    Top three intents identified with volume, AHT, and containment baselines.
    You can name them and quote the numbers from memory.
  2. Q02
    Cost-per-resolved-call target agreed for in-scope intents (not cost-per-minute).
  3. Q03
    Recontact and escalation cost of a partially-contained call modelled, not just the deflection saving.

Data & telephony

Whether your call data, carrier path, and fallback patterns can actually carry a production deployment.

  1. Q04
    Last 90 days of recordings and transcripts exportable in under two weeks, with consent flags.
  2. Q05
    SIP / carrier / CCaaS supports low-latency media streaming without breaking recording or QA.
  3. Q06
    Clear DTMF fallback and warm-transfer-with-context path documented.

Evaluation & QA

×4 survival

Whether you can tell, with evidence, that the AI is improving or regressing call-over-call.

  1. Q07
    Turn-taking latency, barge-in behaviour, and intent-accuracy targets defined per intent.
  2. Q08
    Labelled evaluation set of 200+ real conversations per priority intent, refreshed quarterly.
  3. Q09
    Human-in-the-loop QA scores AI calls on the same rubric as human-agent calls.

Security, privacy, compliance

×4 survival

Whether the legal and data-protection groundwork is in place before a synthetic voice talks to a customer.

  1. Q10
    DPIA / privacy assessment complete for synthetic voice, biometrics, and cross-border flows.
  2. Q11
    Documented position on vendor training-data use, with contractual carve-outs where needed.
  3. Q12
    Approved disclosure script and human opt-out path for regulated intents.

Integration depth

×4 survival

Whether the agent can actually transact against your systems of record, not just answer questions.

  1. Q13
    Voice agent performs the top three transactions end-to-end (authenticate, read, write) — not just FAQ.
  2. Q14
    Identity / auth pattern approved by fraud and security for voice (KBA, OTP, or biometrics).
  3. Q15
    Single source of truth for customer state, read by AI and human agents in real time.

Operating model

×4 survival

Whether there is a named owner, a runbook, and a forecast that treats AI as a permanent channel.

  1. Q16
    Named product owner with weekly time to review failed calls and ship prompt / flow changes.
  2. Q17
    Ops runbook for AI-agent incidents (degradation, hallucination, mis-routing) with a kill switch.
  3. Q18
    AI-handled calls in the workforce-management forecast, not bolted on as a separate channel.

Measurement

Whether the weekly numbers are honest enough to defend in front of finance.

  1. Q19
    Containment, resolution, CSAT, and cost-per-resolved-call reported weekly, AI vs human side-by-side.
  2. Q20
    Closed-loop process feeds failed calls into prompt / flow / training-data changes within a sprint.

2. Unit-economics sanity check

Optional. Stress-tests cost-per-resolved-call against vendor cost-per-minute headlines.

True contained rate
45%
Cost per resolved call
£1.52
Monthly net saving
£96,816

3. Unlock the personalised PDF

Email optional — your score and results page work either way. Add an email to receive the personalised board-ready PDF and subscribe to The Voice AI Brief.

We store your answers, score, and (if provided) email to generate your personalised PDF and aggregate anonymised insights. Nothing is shared with vendors.