Decision sample / Customer support

Evidence before a shortlist.

9 October 20263 providers8 original sources

A worked example of how to compare public claims, price components and unanswered questions before selecting an AI customer-support provider.

By SG Evidence Intelligence / Superintelligence SG Lab · Published 9 October 2026 · Research version 0.2
An uncommissioned research sample. Supplier statements are distinguished from interpretation; no product trial or customer engagement is represented.

Download brief (PDF · 3 pages)Download evaluation workbook (XLSX)View 8 sources

SG Evidence Intelligence / Superintelligence SG Lab. (2026, October 9). AI customer-support procurement sample, version 0.2. https://superintelligencesg.ai/field-notes/customer-support-procurement.html

The decision

Which AI customer-support option deserves a scoped evaluation for a particular workflow, and what evidence must the buyer still obtain?

A useful comparison exposes the gaps.

Public pages can establish what a supplier says. They cannot establish performance on your support cases, a complete deployment price or the contractual handling of your customer data. This sample makes those boundaries visible before a purchasing decision.

Three options, with sources.

A deliberately small comparison. These providers are examples, not an exhaustive list, ranking or endorsement.

01 / AiChat

Messaging and voice require different scope.

AiChat markets messaging and Voice AI, with voice listed under Enterprise. Its public pages describe live-agent intervention. The Team monthly plan lists US$149 with 800 credits and US$5 per additional 100-credit block, using the stated GPT-4o mini rate. S1 · S2

Buyer questionWhich channels, credit rules, integrations and escalation steps apply to the proposed workflow? Obtain its service terms and processing locations; the policy does not establish Singapore-only hosting. S3 / Policy ↗

02 / WIZ.AI

Separate product terms from website privacy.

WIZ.AI describes inbound Wizlynn, outbound Talkbot and human transfer for complex cases. Its trust centre presents supplier security claims. These have not been independently audited for this sample. S4 · S5

Buyer questionWhat are the product-specific data region, retention, subprocessor and handoff terms? The website-visitor privacy policy explicitly excludes talkbot customer service terms; it cannot answer those questions. No full numeric price was established on the reviewed pages. S6 / Policy ↗

03 / AI Rudder

Evaluate the transition to a person.

AI Rudder describes voice, chat, contact-centre assistance and bot-to-human transitions. Its policy describes contractual customer-data processing and retention while needed, with possible additional protections in customer agreements. S7 · S8

Buyer questionWhat context reaches the human agent, how are unresolved cases routed and what service agreement applies? A specific hosting region, fixed retention period and complete scoped price remain unverified here.

Price a component, then ask what is missing.

Illustration only: 1,800 assumed chargeable credits on AiChat Team monthly gives US$149 + 10 × US$5 = US$199 for the stated platform component. This is our arithmetic based on S2, not a vendor quote. Credits are not unique customers. Annual-plan prices and other model rates should not be mixed into this example.

A full deployment can also involve messaging, telephony, setup, integrations, support and tax. Confirm which items apply and request a workflow-specific quote. Public prices and policies can change after the reading date.

Turn research into an evaluation.

The workbook contains two sheets: Evaluation and Sources. Enter the decision owner, workflow, channels, case volumes, constraints and approval route. For each provider, record three proposed checks: a routine answer, uncertainty or a missing answer, and transfer to a human.

Save the actual input, observed output, escalation result and evidence before choosing Pass, Fail or Needs review. Trial rows start as Not tested; acceptance rows start as Not reviewed. No customer results are prefilled. Its price section is a component illustration, not a total-cost quote; the file has not been recalculated in native Excel during preparation.

For a tailored brief, email the workflow, decision question, existing shortlist and desired decision date. We will clarify the scope, evidence and acceptance criteria in writing before any paid engagement. Do not include personal customer records or credentials in an initial enquiry.

Describe your decision by email ↗

What this sample can prove.

It demonstrates a dated research method, explicit source links, transparent unknowns and a reusable evaluation structure. It does not prove supplier performance, customer demand or a completed deployment. SG Evidence Intelligence is a project in formation run by the same operator as Superintelligence SG; no incorporation, client, partnership or revenue is claimed.

Use the three sites together.

.COM / NAVIGATORFind source-linked organisations and programmes; compare their public records.Find candidates ↗

.ORG / COMMONSRead organisation-authored profiles and their visible source status. A profile is not independent verification.Read Commons profiles ↗

.AI / LABExamine the evidence, limits and questions behind a decision.Explore the research archive ↗

Eight original sources.

All read on 9 October 2026. Reading dates are not publication dates. Claims remain supplier statements unless explicitly described as our interpretation.

Version and corrections

0.2 · 9 October 2026 — First public release of the revised sample. Corrects the earlier draft's channel framing, adds price-component evidence and separates website privacy from product terms. HTML and downloadable brief use the same evidence version; the evaluation workbook is version 0.1.

Found a missing or changed source? Send a correction. Material revisions will be recorded here.