Scope and inclusion
Included products materially overlap website chat and digital customer-service workflows, have a current official product source, disclose a verifiable price or contact-sales state, and provide enough evidence for at least five material fields. Products outside the channel, without a comparable buyer job, or without minimum current evidence are excluded or labeled insufficient evidence.
Included product types
Included products materially serve website chat and digital customer-service workflows, match a defined buyer job, and have enough current official evidence to evaluate the relevant gates and criteria.
excluded product types
A product is excluded when its primary channel or buyer job is not comparable, or when the requested scenario falls outside its stated public scope. Exclusion is not a negative score.
minimum public evidence
A candidate needs a current official product source, a verifiable price or contact-sales state, and evidence for material gate and criterion fields. Missing material evidence produces insufficient evidence, not an assumption.
Comparison accountability
- Accountable comparison reviewer
- Roscoe Morgan, factual comparison-claim review and sign-off
- Corrections and source updates
- support@firstaiemployee.com
Evidence collection
- Official pages establish current product, help, pricing, limitation, and legal facts.
- Public demos support only the behavior actually visible in the dated demo.
- Controlled tests record tier, configuration, inputs, expected and observed results.
- Support confirmations are dated, attributed, and narrower than published terms.
Official sources
Official product, help, pricing, limitation, privacy, and legal pages establish vendor-controlled facts. Each material source keeps its URL, checked date, evidence state, and review deadline.
public demos
A dated public demo supports only the actions that can actually be observed. It does not establish production reliability, unshown integrations, plan eligibility, or contractual commitments.
controlled tests
A controlled test records tier, configuration, inputs, expected behavior, observed result, captured evidence, and limitations. Findings remain bounded to that reproducible setup.
support confirmations
A support confirmation is dated and attributed, but remains narrower than current published terms. It cannot silently create a contractual promise or erase a contrary official limitation.
Must-pass gates
| Gate | Published rule |
|---|---|
| Channel eligibility | Evidence must show this gate passes for the stated scenario. An unknown does not pass, and a failed must-have eliminates the candidate regardless of score. |
| Geography | Evidence must show this gate passes for the stated scenario. An unknown does not pass, and a failed must-have eliminates the candidate regardless of score. |
| Regulated-data fit | Evidence must show this gate passes for the stated scenario. An unknown does not pass, and a failed must-have eliminates the candidate regardless of score. |
| Required system/integration | Evidence must show this gate passes for the stated scenario. An unknown does not pass, and a failed must-have eliminates the candidate regardless of score. |
| Action requirement | Evidence must show this gate passes for the stated scenario. An unknown does not pass, and a failed must-have eliminates the candidate regardless of score. |
| Human-workspace requirement | Evidence must show this gate passes for the stated scenario. An unknown does not pass, and a failed must-have eliminates the candidate regardless of score. |
| Budget/contract requirement | Evidence must show this gate passes for the stated scenario. An unknown does not pass, and a failed must-have eliminates the candidate regardless of score. |
| Implementation model | Evidence must show this gate passes for the stated scenario. An unknown does not pass, and a failed must-have eliminates the candidate regardless of score. |
| Source freshness | Evidence must show this gate passes for the stated scenario. An unknown does not pass, and a failed must-have eliminates the candidate regardless of score. |
Safety
Safety is evaluated through the published regulated-data-fit, implementation-model, source-freshness, and other applicable universal gates. A material safety unknown does not pass the scenario.
required action
The action-requirement gate asks whether the product can complete and confirm the buyer’s required action, not merely answer or route a request. Unknown action support does not pass.
geography
The geography gate requires current official evidence that the service and required channel are offered where the buyer operates. A broad availability claim is not enough.
commercial ceiling
The budget-and-contract gate checks required plans, usage, seats, add-ons, commitments, and unknown charges against the scenario. A material cost unknown prevents a calculated winner.
data handling
The regulated-data-fit and required-system-integration gates use published boundaries and tested controls. The method never infers a privacy, security, or compliance capability from silence.
- Define the scenario’s must-have requirement.
- Pass with current evidenceContinue to the published weighted criteria.
- Unknown or materially staleReturn insufficient evidence and publish no winner.
- FailEliminate the candidate for that scenario regardless of score.
Comparison criteria
| Criterion | Measurement | Evidence |
|---|---|---|
| Task coverage | Exact eligible jobs, not generic feature count | official docs/test |
| Action completion | Read/write/confirm/recover behavior | official docs/test |
| Human escalation | Trigger, availability truth, context transfer | official docs/test |
| Setup effort | Buyer work, vendor work, configuration control | official source/user test |
| Ongoing operation | Content updates, QA, reporting, staffing | official source/test |
| Price predictability | Unit, allowance, overage, dependencies | official pricing |
| Total cost | Same scenario, required plans/seats/usage | formula + official pricing |
| Limitations | Geography, data, channels, use cases | official source/terms |
| Evidence confidence | Quality and freshness of proof | source ledger |
| Accessibility/privacy | Public commitments and tested behavior | official policy/test |
| Buyer scenario | Task/action | Human workspace | Setup/operation | Channels | Price predictability | Limitations fit | Evidence |
|---|---|---|---|---|---|---|---|
| Managed small-business website assistant | 25% | 5% | 25% | 10% | 15% | 10% | 10% |
| Live-chat team with AI | 20% | 25% | 15% | 15% | 10% | 5% | 10% |
| Help-desk-native AI | 25% | 20% | 10% | 15% | 10% | 10% | 10% |
| Visual bot builder | 20% | 10% | 25% | 10% | 15% | 10% | 10% |
| Complex action-taking agent | 30% | 15% | 10% | 15% | 10% | 10% | 10% |
Capability
Capability combines exact task coverage and action completion: eligible jobs, read/write/confirm behavior, failure recovery, and evidence for the scenario, not a generic feature count.
reliability and fallback
Reliability and fallback examine escalation triggers, actual human availability, transferred context, recovery, and operating evidence. A marketing promise is not treated as a measured service level.
setup and control
Setup and control cover buyer work, vendor work, configuration authority, testing, content maintenance, quality assurance, reporting, permissions, and ongoing operational burden.
pricing
Pricing compares the same scenario using units, allowances, dependencies, overages, required plans or seats, usage, commitments, and formula-backed total cost. Contact-sales remains unknown.
privacy
Privacy assessment is limited to published commitments and tested behavior, including material channel and data limitations. The method does not infer certifications or legal compliance.
support
Support assessment uses documented routes, escalation conditions, human-workspace requirements, and operating evidence. It does not invent availability or response-time promises.
Pricing normalization
Reference profiles
Reference profiles hold the material website inquiries, answers, handoffs, and follow-up actions volumes, complexity, handoffs, and maintenance assumptions constant so products are compared against the same buyer job.
assumptions
Every model exposes the base plan, units, allowances, required seats or add-ons, setup allocation, usage assumptions, and formula. A changed assumption produces a new scenario, not a hidden adjustment.
unknowns
Contact-sales pricing, unstated overages, and configuration-dependent charges remain null. A material unknown is neither free nor expensive and can yield an insufficient-evidence verdict with no winner.
| Billing model | Comparable monthly formula | Required disclosed inputs |
|---|---|---|
| Per seat | required platform plan + required seats × per-seat rate + AI or usage add-ons + allocated setup | required base plan, seat count, seat rate, AI add-ons, setup allocation period |
| Per message | required base plan + max(0, measured billable messages - included messages) × message rate + required seats and add-ons + allocated setup | vendor message definition, included messages, rate, required seats and add-ons, setup allocation period |
| Per conversation | required base plan + max(0, measured billable conversations - included conversations) × conversation rate + required seats and add-ons + allocated setup | vendor conversation and reset definition, allowance, rate, required seats and add-ons, setup allocation period |
| Per automated outcome | required base plan + measured vendor-defined automated outcomes × outcome rate + required seats and add-ons + allocated setup | vendor outcome definition, excluded and disputed outcomes, allowance, rate, required seats and add-ons, setup allocation period |
| Platform plus usage | required platform subscription + required seats + each disclosed usage unit × its rate + required add-ons + allocated setup | platform tier, seats, every usage unit and allowance, add-ons, setup allocation period |
Contact-sales and unknown charges remain unknown. Currency, cadence, annual commitment, allowances, overages, taxes, trials, and promotional pricing stay separately labeled.
How recommendations are written
A verdict is conditional: “best for” names the scenario and passed gates; “avoid when” names a disqualifying requirement or tradeoff; “neither” explains when a human service, existing workflow, or no purchase is the better decision. The publication does not claim a universal winner or use an ordinal list to hide different product classes.
| Score | Meaning |
|---|---|
| 0 | explicitly does not meet |
| 1 | severe limitation |
| 2 | partial |
| 3 | meets |
| 4 | strong |
| 5 | exceptional, supported by evidence |
Unknown is stored as null and receives neither a positive nor a negative score. A material unknown yields “insufficient-evidence” and no winner. Scores are decision aids; every short answer must explain the actual requirement and tradeoff.
Best for
“Best for” names the buyer scenario, confirms that every must-have gate passed, and explains the weighted strengths and tradeoffs that support the conditional recommendation.
avoid when
“Avoid when” identifies a failed gate, missing requirement, material unknown, operating burden, or tradeoff that should disqualify the product for the stated scenario.
neither
“Neither” explains when a human service, existing workflow, different product class, or no purchase better fits the requirement. The method does not force a vendor winner.
| Candidate | Task/action | Human workspace | Setup/operation | Channels | Price predictability | Limitations fit | Evidence | Weighted result |
|---|---|---|---|---|---|---|---|---|
| Synthetic Provider Alpha | 4 × 25% | 2 × 5% | 3 × 25% | 4 × 10% | 3 × 15% | 4 × 10% | 4 × 10% | 3.5 |
| Synthetic Provider Beta | 3 × 25% | 5 × 5% | 4 × 25% | 4 × 10% | 2 × 15% | 3 × 10% | 3 × 10% | 3.3 |
Published sample verdict: For the fictional managed website-assistant scenario, Synthetic Provider Alpha is the sample best fit because task coverage, price predictability, limitations fit, and evidence outweigh Synthetic Provider Beta's stronger human-workspace and setup scores. This is an arithmetic demonstration, not a real recommendation.
Owner-product safeguard
- The ownership disclosure appears before comparison conclusions.
- At least one realistic competitor-win or neither scenario is stated.
- First AI Employee uses the same evidence states and unknown handling.
- Accountable comparison review: Roscoe Morgan, factual comparison-claim review and sign-off.
Disclosure placement
The ownership relationship appears before comparison conclusions and commercial actions so a reader can evaluate First AI Employee’s material interest before using the verdict.
competitor-win requirement
Every direct comparison must publish at least one realistic scenario where a competitor or neither option is the better fit. The owned product never receives automatic placement advantage.
review sign-off
Accountable comparison review is assigned to Roscoe Morgan, factual comparison-claim review and sign-off. Sign-off checks evidence, gates, weights, null handling, disclosure, and the conditional verdict.
Freshness and revisions
Material commercial sources are checked at least every 30 days, and the methodology itself at least every 6 months. Price, plan, ownership, limitation, legal, or product changes trigger an earlier review. Each material change keeps its check date, evidence state, methodology version, and a dated internal editorial record.
30-day commercial sources
Material commercial sources are rechecked within 30 days. A source that exceeds that limit cannot support a current price, plan, or commercial verdict.
event-driven updates
A price, plan, ownership, limitation, legal, integration, or product change triggers review before the ordinary cadence. A materially affected verdict is recalculated from current evidence.
versioning
Methodology version 1.0.0 changes immediately when a criterion, weight, source hierarchy, or disclosure changes. Source refreshes retain their own dated internal editorial record.
Methodology version
Current methodology version 1.0.0, effective . A version changes immediately when criteria, weights, the source hierarchy, or the disclosure changes. Each comparison exposes dated evidence; substantive changes are recorded internally. A new check date alone is not a substantive revision.
The machine-readable JSON download appears once in the version artifact below.
Download methodology version 1.0.0 as JSON Includes the gates, criteria, scenario weights, scoring rules, and safeguards.
How much official evidence exists
How much of an AI answering-service purchase can a buyer settle from official pages? We counted every normalized procurement fact this publication tracks and read the evidence state we recorded for it. A fact counts as settled only when a current official vendor page states it plainly.
Across 9 vendors, 32 of 44 facts are settled. 12 leave the buyer with an open question. The least documented fact in the category is live-human-backup: of the 8 vendors that carry a record, 0 state it plainly on an official page.
Evidence as of . The counts move when the evidence moves, and the same figures travel in the methodology download.
What was measured
Every normalized procurement fact tracked for every vendor, read against the evidence state recorded for it. A fact counts as settled only when a current official vendor page states it plainly. Configuration-dependent, not-stated and not-offered-explicit each leave a buyer with work to do, so none of them counts as settled. A vendor with no record for a field is absent from that field rather than counted either way.
what the counts show
live-human-backup: 0 of 8 stated plainly. setup-model: 7 of 9 stated plainly. billing-unit: 8 of 9 stated plainly. starting-price: 8 of 9 stated plainly. product-class: 9 of 9 stated plainly. Product class is the only tracked fact with full official coverage. Every vendor says what it is; the category is far less clear about what happens when the software cannot finish the job.
limitations
The measurement covers what official pages state, not what a product does. A vendor may support a capability it does not document. Five normalized facts do not describe a whole purchase, and sales conversations, trials and private documentation are out of scope. First AI Employee is one of the measured vendors and this publication’s owner sells it; it carries no live-human-backup record either, so it is absent from that row rather than scoring well on it.