Implementation checklist
Agree on the call type, allowed actions, human fallback, and evidence you need before expanding.
Download checklist (PDF)Read the checklist on this pageBuyer resources
Useful tools for your first phone-agent pilot.
Define one workflow, test its handoffs, and estimate its full operating cost with your business owner and technical evaluator.
Open downloads. No signup required.
Cost worksheet
Include providers, unsuccessful attempts, implementation, and human follow-up. Replace the illustrative inputs with your own quotes and observations.
Read the worksheet instructionsUse the checklist to define a decision.
Start with one call type, a defined outcome, and a named human fallback. Record what the pilot must demonstrate and which requests it should leave with your team.
Read the full implementation checklist
A working evaluation sheet for business and technical owners. Complete one workflow before adding others. This is an evaluation aid, not a certification or a promise of QuickVoice functionality.
Project / workflow: __________ Business owner: __________ Technical owner: __________ Review date: __________ Intended callers: __________ Decision date: __________
1. Define the business job
- Choose one call type: appointment request, answering-service intake, support triage, or lead qualification.
- Write the exact successful outcome, including the system where it must be recorded.
- List questions the agent may answer and actions it may request. List prohibited topics/actions.
- Identify a human fallback, staffed hours, contact destination, and expected next step when unavailable.
- Record baseline call attempts, answered calls, relevant outcomes, staff time, and cost over a named period. Use aggregates, not caller transcripts, in the business case.
- Define decision thresholds before the pilot: acceptable completion rate, failed/escalated-call rate, duplicate-action count, and maximum cost. Enter actual thresholds: __________.
2. Confirm implementation dependencies
- Assign an engineer or implementation partner. Confirm the required console, API, worker, database and provider setup.
- Verify the exact deployment version and documented setup path. A local startup is distinct from a functioning carrier-connected call.
- Confirm LiveKit, telephony and model-provider accounts, quotas, permissions, credentials and expected costs.
- Check each calendar/CRM/helpdesk operation individually: available API, authentication, permissions, field mapping, errors and duplicate prevention. Do not assume a logo means a working integration.
- Prepare a small approved knowledge set with an owner and update date; remove contradictory or expired answers.
- Record who operates infrastructure, responds to incidents, reviews calls, changes prompts and pays provider bills.
3. Review data and caller handling
- Map audio, transcript, recording, contact data, tool requests, logs and backups to their actual storage/processors.
- Decide data access, retention and deletion requirements; verify them in the chosen deployment.
- Have the responsible business reviewer confirm caller notices, consent needs and any relevant sector obligations. Obtain required provider agreements before sensitive data is used.
- Use synthetic examples and consenting staff for initial tests. Keep credentials and personal information out of public screenshots.
- Document where the agent must stop, escalate or decline to act; test those cases.
4. Test a realistic call set
| Scenario | Expected result | Actual result / evidence | Owner |
|---|---|---|---|
| Normal request with all information | Correct next step; one recorded outcome | __________ | __________ |
| Missing or ambiguous information | Clarify without guessing | __________ | __________ |
| Caller interrupts or corrects details | Use corrected information | __________ | __________ |
| Unknown answer | Admit uncertainty; use documented fallback | __________ | __________ |
| Calendar/tool/provider unavailable | No false booking or false success | __________ | __________ |
| Duplicate request or retry | No duplicate side effect | __________ | __________ |
| Caller asks for a human | Follow the configured escalation path | __________ | __________ |
| Out-of-scope or sensitive request | Stop or escalate per agreed policy | __________ | __________ |
| No answer / voicemail / wrong contact | Record correct disposition; apply retry policy | __________ | __________ |
| Poor audio or unexpected language | Clarify or fall back; do not claim comprehension | __________ | __________ |
5. Decide using measured results
- Reconcile outcome counts against source systems, including failed attempts and human follow-up.
- Enter observed inputs and provider quotes in the cost worksheet; replace every illustrative input.
- Compare the pilot with a comparable baseline period and disclose staffing, demand or process changes.
- Review failures with the business owner and engineer; fix and retest material issues.
- Record a clear decision: expand, revise and retest, or stop. Keep a rollback plan and a reachable owner.
Decision: __________ Reason/evidence: __________ Next review: __________
Make the worksheet your own.
The CSV contains spreadsheet formulas. Its sample values are illustrative, and zero supplier rates mean unknown, not free. Read the instructions, replace the inputs, and keep the validation flags unconfirmed until you have supporting evidence.
Read the full cost worksheet instructions
Open the CSV in Excel, LibreOffice Calc or Google Sheets using comma-separated import and formula interpretation. Keep the header on row 1 and the original row order: formulas reference column B by row. This file intentionally contains spreadsheet formulas. Do not import it as a plain-text-only table if you want calculations.
What to edit
Edit column B in rows 2–30 and baseline row 51. The nonzero values are illustrative arithmetic inputs, not typical customer usage, supplier prices or deployment estimates. Zero supplier rates mean unknown, not free. Replace every applicable input with measured values or dated quotes, then set B29 and B30 to 1 only when the underlying checks are actually complete. Record quote URLs, quote dates, provider/model/region, billing increment and exclusions in column D.
All costs use USD. Convert other currencies at your approved dated exchange rate before entry. Fractions are decimals: enter 0.6 for 60%, not 60. Duration is in minutes; labor is USD per hour; LLM price is USD per one million tokens; TTS price is USD per one million characters. If a provider uses another unit, convert it before entering a rate. A blended LLM rate must account for your input/output/cached-token mix. Do not add a separate component charge if it is already included in a bundled runtime quote.
The model includes unanswered attempts, connected runtime, speech/model costs, phone numbers, infrastructure, storage, human follow-up, quality review, amortized implementation and contingency. Human follow-up uses answered calls as its denominator; add follow-up on unanswered attempts in other monthly costs if applicable. Carrier legs, transfer legs, rounding, minimum commitments, retry rules, taxes, peak-capacity charges, training, supervision and support may require additional amounts in row 28. Avoid counting a bill twice.
How to interpret results
- Row 46: modeled monthly cost, including amortized implementation and contingency.
- Rows 47–48: modeled cost per attempt and per answered call. These are different denominators.
- Row 49: initial engineering cash cost. It is also spread over row 25's amortization period in the monthly model; do not add it a second time when making an amortized comparison.
- Row 50: input check. The default is explicitly ILLUSTRATIVE OR INVALID — NOT A QUOTE. “Confirmed inputs” means your flags and numeric ranges pass; it is not independent validation of source evidence.
- Row 51: comparable measured baseline monthly cost. Leave zero if unknown. Use the same call types, period, service level and cost scope.
- Rows 52–53: modeled difference from baseline, shown only after inputs are confirmed and baseline is positive. These are not measured savings, ROI, or a revenue forecast. A negative value means the modeled workflow costs more.
Arithmetic check with the untouched sample
The illustrative inputs yield 600 answered calls, 400 unanswered attempts and 1,800 connected minutes. Because supplier rates remain unknown/zero, the visible subtotal is incomplete: human follow-up is $360, quality review $120, amortized implementation $125, subtotal $605 and contingency $90.75, for $695.75. This is a formula sanity check, not a usable price estimate. The one-time illustrative implementation cost is $1,500. The validation flag must stay unconfirmed and baseline differences blank.
Before using a decision estimate, reconcile one actual provider invoice or approved quote against the model, compare an independent hand calculation, test a zero-call and invalid-fraction input, and verify the validation flag changes appropriately. Retain the assumption sheet with the date of the business decision; do not reuse an old quote silently.
Keep exploring
Match the tools to your workflow.
Work through one workflow with us.
Bring your call type, approximate volume, destination systems, and the decision you need to make. Keep caller records and sensitive information out of the initial enquiry.