# AI receptionist cost workbook: method and input dictionary

Version 1.0. Official-source review: **17 November 2026**. Produced by Valory AI, which sells managed AI reception and has a commercial interest in this comparison. This is a budget model using published information and explicit synthetic workloads. It is not an observed customer-spending study, product performance test, provider ranking or complete quote.

## Scope and sources

The roster is the seven providers already present in the cost guide: Valory AI, Johnni AI, BotBloke, Nexwin, Smith.ai, RingCentral and TransferToAI. Twelve rows retain selected published plans, including Johnni's monthly/annual alternatives, Smith.ai's two entry tiers and unverified-provider quote placeholders. Plan selection is not an assessment that the scopes are equivalent. Enterprise subscription and Valory setup are published **from** amounts, not fixed quotes.

`pricing-source-register.csv` records the official URL, page section, fact, status, qualification, review date and country/currency context. `retrieved_at_utc` has date precision only. Exact request seconds were not retained. The access date is not a claim that every page was freshly crawled that day: the public extraction tool can return cached text. No enquiry, account, sales form, call, paid API, private customer record or test booking was used. No inaccessible source was bypassed.

A verified price field does not verify a provider's functional capability. Grouped plan rows cite the named card/table; unresolved policies remain separate source rows. `UNKNOWN` means not established in the bounded public inspection. `UNAVAILABLE` records a failed read, not proof that an offer/business no longer exists. `QUOTE_REQUIRED` means a required component must be priced separately. No historic inaccessible price was carried forward as current.

Valory prices/rates remain 149/299/499/from 1,299 AUD per month, with applicable GST included and setup from 990. Johnni's Pay As You Go is 49 monthly or a 44 monthly equivalent paid annually, plus GST; SMS is the separately advertised 50/month unmetered add-on. Nexwin's Business offer remains 249 plus GST, with 600 minutes and 0.49/minute extra; no zero-setup assertion was re-established from pricing/terms. Smith.ai's Free/Pro call rates are retained in **unlabelled source dollars**; currency and Australian availability remain unresolved. RingCentral's 62 add-on and 0.80/minute extra are retained separately from the required, unpriced RingEX base; its AU add-on page did not explicitly settle currency/tax. BotBloke and TransferToAI have no current numeric inputs after failed public reads. These are source snapshots, not guaranteed future offers.

## Files and reproducibility

- `ai-receptionist-cost-comparison.xlsx`: editable buyer workbook. Blue values are editable; amber cells mark quoted/model assumptions. Results are formulas, not pasted CSV totals.
- `pricing-inputs.csv`: published facts, explicit model assumptions and initially blank quote components. Every plan cites source IDs. Empty numeric cells are missing, not zero.
- `scenario-inputs.csv`: three synthetic equal-duration workloads and reproducible duration vectors.
- `calculated-scenarios.csv`, `first-year.csv`, `sensitivity.csv`: deterministic baseline calculations. A blank numeric result remains unavailable; read `result_status`, basis and limitations.
- `calculate.py`: Python standard-library, offline calculation and validation. No network calls. No CSV is modified in check mode.
- `test_calculate.py`: independently stated arithmetic and boundary fixtures.
- `build-workbook.mjs`: Artifact Tool workbook authoring/recalculation/verification using the existing bundled runtime, with no repository dependency change. Its PNG checks go to the operating-system temporary directory, not the public package.
- `manifest.json`: SHA-256 hashes of this reviewed package, excluding itself. A hash establishes file integrity, not factual correctness or freshness.

From this directory:

```bash
python3 -B test_calculate.py
python3 -B calculate.py --check
```

To intentionally regenerate baseline CSV outputs after reviewed input changes:

```bash
python3 -B calculate.py
```

Workbook generation requires the bundled Artifact Tool and Node runtime. Set `ARTIFACT_TOOL_MODULE` to the file URL of its documented `dist/artifact_tool.mjs` export, then run `node build-workbook.mjs`. The task's bundled runtime loader supplies that location. There is no dependency download or automatic vendor refresh. Regeneration performs 88 baseline comparisons, input-mutation checks, recalculation, formula-error scan, XLSX export/re-import and sheet rendering. Temporary verification images are not part of the download. Run `python3 -B calculate.py --manifest` only after final source, formula and workbook review; subsequent `--check` verifies those file hashes as well as the calculated CSVs.

Changes a reader makes in Excel do **not** update the baseline CSVs or the original source review. Record the new quote source/date before substituting values, retain the dated original, and recheck formulas and terms. The source register remains the original evidence snapshot. No macros, external workbook/data connections, hidden private sheets or automatic prices exist.

## Scenario dictionary

`calls` is the number of answered, billable calls, not all calls made to a business. `seconds_per_call` is the duration of **every** synthetic call in that row, not an estimated average. `sms_units` is assumed billable SMS units, not long-message segments or a promise about a provider's metering. Vectors use `seconds x count`, with semicolon-separated groups for independent rounding fixtures.

| Scenario | Calls | Seconds per call | Raw voice minutes | SMS units |
|---|---:|---:|---:|---:|
| Ordinary overflow | 100 | 120 | 200 | 100 |
| Busy overflow | 150 | 180 | 450 | 200 |
| Sustained load | 300 | 180 | 900 | 450 |

The call-length comparison holds Valory Core, 100 calls and 100 SMS units fixed initially, with 60/120/180/300 seconds per call. This is a specified scenario comparison, not a forecast, native Excel What-If Data Table or typical client distribution. Its formulas refer to the same plan/quote inputs as the monthly models.

## Plan, assumption and quote fields

| Fields | Meaning and missing-value treatment |
|---|---|
| `currency`, `tax_basis` | AUD, unlabelled source dollars or unknown; tax is inclusive, exclusive or unknown. Only explicit AUD plus a known basis permits an AUD including-GST value. No FX conversion or tax-credit claim. |
| `billing_frequency`, `commitment_months`, `upfront_subscription_months` | Monthly or annual payment basis, published/quoted minimum term and the number of base subscription months paid initially. The monthly amount on an annual row is an equivalent, not a month-to-month offer. Unknown upfront months produce Unknown upfront cash. |
| `subscription`, `voice_unit`, `included_voice`, `voice_overage` | Source/quoted monthly amount, MINUTE or CALL, included units and extra-unit price. All are required for a voice subtotal. CALL plans never convert minutes into calls. Unknown/malformed/negative required inputs do not yield a subtotal. |
| `published_rounding_seconds`, `published_minimum_seconds` | Only explicit public policies populate these facts. Blank remains unresolved. RingCentral establishes a 30-second increment; no minimum is inferred. |
| `model_rounding_seconds`, `model_minimum_seconds` | Editable modelling assumption or verified quote rule. The baseline assumes a one-second increment and zero minimum where unknown. These are not verified tariffs. Clearing either required minute-plan value invalidates its subtotal. A supplied published increment/minimum must agree with the model. |
| `sms_mode`, `included_sms`, `sms_overage`, `sms_addon` | METERED charges units above allowance. FLAT_ADDON adds the published charge only when SMS is positive. UNKNOWN leaves SMS unpriced. Explicit zero SMS workload has zero modelled SMS cost; this does not establish a free SMS service. |
| `transfer_wait_mode` | UNKNOWN excludes unpriced additional time. NO_EXTRA is an explicit reader/quote assumption of no extra charge. IN_VOICE adds transfer plus waiting seconds to each minute-plan call. SEPARATE_PER_MINUTE prices that time at a separately entered rate. The model does not implement a separate carrier's different tariff, so include those in a quote instead. |
| `transfer_seconds_per_call`, `wait_seconds_per_call`, `transfer_wait_rate` | Explicit quote/model values. IN_VOICE needs both durations and a minute plan; clearing a required duration blocks the model. Separate-rate mode needs both durations/rate/increment. The same increment rounds each separate time block; confirm this convention before using that mode. No additional time is double-counted in the main voice amount for separate-rate mode. |
| `required_base_monthly`, `custom_monthly`, `custom_onetime` | Editable quoted base-phone, custom recurring and one-off charges. Blanks stay unpriced. Only entered non-negative amounts join the partial subtotal. A blank never proves no base/integration charge applies. All components on a plan row must share its currency and tax basis. |
| `setup`, `setup_basis` | Fixed quoted amount, published minimum or unknown. A known minimum is added as a floor, not a complete scope quote. An unknown setup blocks the first-year subtotal. |
| `quote_reference`, `source_ids`, `limitations` | Reader quote source/date, dated original-source lineage and material limitations. Do not place personal/customer data here. A quote does not silently supersede the original reviewed record. |

## Formula contract

For each minute-plan call, add quoted transfer/wait duration only in IN_VOICE mode, apply the selected minimum, then round **that call** up to the selected increment. Sum the rounded seconds and divide by 60. Never round only aggregate minutes when per-call increments apply. Two 30-second calls and calls of 1 and 59 seconds have equal raw time but different billed time at a 30-second increment.

`Extra voice units = max(0, model voice units − included units)`.

`Voice overage = extra voice units × extra-unit rate`.

`Metered SMS = max(0, billable SMS units − included SMS units) × extra-SMS rate`.

`Known monthly subtotal = subscription + known voice overage + known SMS + known separate transfer/wait charge + entered required-base and custom recurring amounts`.

An explicitly selected METERED, FLAT_ADDON or SEPARATE_PER_MINUTE mode with incomplete required inputs blocks the subtotal in both calculation engines. A published minimum, like a published rounding increment, must agree with the model. UNKNOWN modes intentionally retain a partial known-component subtotal. Unknown additional components remain text/blank in their own fields and are excluded from this **partial** sum. All outputs retain a partial status. No output changes to “complete quote” automatically, even when a reader fills fields. There is no cheapest-provider formula. Negative values and fractional call/SMS counts are rejected by the script; workbook formulas and validation also prevent those from appearing as plausible costs.

For explicit AUD plus-GST amounts, multiply the final subtotal by 1.1. For GST-inclusive amounts, retain it. Unknown tax/currency yields no comparable AUD value. Full precision is kept through each monthly build. Source-basis and AUD including-GST monthly subtotals are each rounded independently to two decimals from that build, so tax is not calculated on an already rounded subtotal. For example Johnni's 100 calls of 121 seconds plus 100 SMS units gives a 169.83 source-basis subtotal and 186.82 including GST. Annual source-basis totals sum the displayed two-decimal monthly amounts, then add setup/one-off components; no tax-inclusive annual amount is inferred.

First year covers **12 paid months**, initially ten ordinary and two busy. `Recurring subtotal = ordinary subtotal × ordinary months + busy subtotal × busy months`. Counts must be whole, non-negative and sum to 12. `First-year known subtotal = recurring subtotal + known setup + entered custom one-off`. Unknown setup prevents this output; unpriced custom work remains excluded and the total remains partial. The average divides by 12. The base-subscription-upfront column is `subscription × upfront months`; it deliberately excludes setup, SMS and other add-ons, so it is not a complete initial invoice. The minimum contractual obligation can extend beyond this horizon and must be considered separately.

Trials, discounts not frozen in the input, outbound calls, usage rollovers, proration, custom integration scope, carrier forwarding, tax eligibility, downstream staff time and commercial outcomes are outside these calculations. No bookings, revenue, savings or ROI is inferred.

## Checks and limits

Regression fixtures retain Valory Core/Premium values of 299/444/784/604 for the four existing examples, and Core's **base-only** first year of 4,578 and 381.50/month equivalent. The workbook's default 10 ordinary plus 2 busy months intentionally gives Core 4,868 instead. These are different workloads, not conflicting totals. The published rates were checked on the review date; exact public rounding and transfer policy remain unresolved.

Boundary tests cover allowance below/at/above, zero use, per-call mixed-duration rounding and minimums, call-versus-minute billing, unknown setup/SMS/currency, annual upfront commitment, quoted transfer/base/custom components and cleared/invalid inputs. Workbook verification compares 88 baseline results to independent Python outputs and tests live input changes before restoring defaults. Eight sheets are exported, recalculated and re-imported with Artifact Tool; rendered views are inspected. Desktop Excel and Google Sheets behaviour has not been independently tested. Custom print areas are not configured because the bundled authoring API did not expose suitable print settings; review page layout in the target application before printing. Formatted sheet ranges are bounded to populated content. CSV input validation does not establish supplier invoice accuracy.

To propose a correction, contact Valory AI with the source ID, plan and dated official evidence. Retain prior source snapshots and record changed facts separately. Material price/policy changes require both source and formula review before republishing.
