Who sets the coding rules

Arjun Mehta has built conversational AI systems in Pune since 2018, including a banking chatbot that handled 300,000 sessions a month and a support copilot deployed to 200 agents. He has run the accuracy evals those launches required and knows how a 90% vendor claim dissolves into specific failure modes. He writes the evaluations context and vendor-claim labeling here. He does not endorse any vendor he has worked with.

Covers: why vendor benchmarks are shown as claims; failure modes behind accuracy complaints; agent assist vs autonomous agents.

Every quote on this site is verbatim from a review we opened, dated 2024-10-01 or later, and every code is published on the reviews page.

Claire Dubois has managed multilingual support teams in Lyon since 2015, including a 30-agent desk serving French, German and Spanish customers. She has read the non-English reviews most English-only sites skip and coded them the same way, which is why this site quotes reviewers from several languages. She writes the multilingual and translation-failure notes here. She does not cover translation vendor selection.

Covers: non-English reviews; multilingual agent failures; reviewer labeling conventions.

Every quote on this site is verbatim from a review we opened, dated 2024-10-01 or later, and every code is published on the reviews page.

Hana Kim has deployed and tuned AI support agents for consumer apps in Seoul since 2019, eleven deployments across three companies. She has read more than 20,000 escalated transcripts to find where the agent gave a confident wrong answer, and that reading shaped how this site codes accuracy complaints. She writes the accuracy and knowledge-quality categories here. She does not review products she has worked for.

Covers: accuracy coding; what "confidently wrong" looks like in a review; knowledge base limitations.

Every quote on this site is verbatim from a review we opened, dated 2024-10-01 or later, and every code is published on the reviews page.

Ingrid Bauer has led support automation for two Berlin software companies since 2017, including a Zendesk desk where she measured AI conversation pricing against actual resolution for 18 months. She has read the billing complaints on review platforms for every product this site covers and cross-checked them against published pricing. She writes the cost coding and per-conversation billing notes here. She does not give procurement advice.

Covers: cost coding; per-conversation and per-seat billing complaints; how shares are computed.

Every quote on this site is verbatim from a review we opened, dated 2024-10-01 or later, and every code is published on the reviews page.

Kwame Mensah has implemented and administered support platforms for companies in Accra since 2018, about 25 deployments. He has done the setup work himself on four of the products this site covers and knows which setup failures are vendor defaults rather than user error — and which reviewers say otherwise. He writes the setup coding here, from both his installs and reviewer accounts. He does not cover enterprise contract terms.

Covers: setup coding; default settings that surprise reviewers; admin and configuration complaints.

Every quote on this site is verbatim from a review we opened, dated 2024-10-01 or later, and every code is published on the reviews page.

Lucas Meyer has administered four different help desks for a Zurich software company since 2015, migrating between them with a combined 15,000 monthly tickets. He has kept configuration records for every AI feature he has switched on and off, and uses them to judge whether a reviewer's complaint reflects the product or the plan they bought. He writes the feature-availability and plan-tier notes here. He does not run evaluations or tests of these products.

Covers: which AI features sit behind which plan; availability gaps reviewers hit; admin-side complaints.

Every quote on this site is verbatim from a review we opened, dated 2024-10-01 or later, and every code is published on the reviews page.

Marco Ferretti has designed support workflows for banks and telcos in Milan since 2012, including one IVR-to-AI migration covering 4 million monthly contacts. He has kept the escalation matrices from those projects and uses them to judge whether a reviewer's "it never escalated me" complaint is a product flaw or a setup choice. He writes the handoff and escalation analysis here. He does not cover contact-center hardware.

Covers: handoff coding; escalation paths reviewers describe; when self-service walls people in.

Every quote on this site is verbatim from a review we opened, dated 2024-10-01 or later, and every code is published on the reviews page.

Mei-Ling Wong has designed and written dialogue for customer-service AI in Singapore since 2019, across telco, airline and government service desks with more than 500,000 monthly conversations. She has rewritten the exact kinds of replies reviewers complain about — over-apologetic loops, refused refunds, phantom escalations — and recognizes the patterns in their quotes. She writes the conversation-quality coding here. She does not design agents for the vendors covered.

Covers: conversation-quality complaints; tone and loop failures; refund and policy refusal patterns.

Every quote on this site is verbatim from a review we opened, dated 2024-10-01 or later, and every code is published on the reviews page.

Rachel Donovan has run support operations for SaaS companies in Toronto since 2014, most recently a 60-agent desk handling 40,000 tickets a month. She has piloted five AI agents on that desk and kept the before-and-after resolution and handoff numbers from each pilot. She writes the coding rules and job/accuracy judgments here based on what reviewers report, not on vendor demos. She does not score vendor marketing benchmarks.

Covers: how reviewer quotes are coded on job, accuracy, handoff, cost and setup; what counts as a handoff failure.

Every quote on this site is verbatim from a review we opened, dated 2024-10-01 or later, and every code is published on the reviews page.

Sofía Navarro has audited support desks for retail and travel companies from Mexico City since 2016, about 40 audits, most under 50 agents. She built the reviewer-interview habit this site borrows: asking what the AI was supposed to do before judging whether it did it. She writes the "job" coding — whether the AI does the work reviewers hired it for. She does not consult on staffing levels.

Covers: job coding; scope creep in AI agent promises; what reviewers expected vs received.

Every quote on this site is verbatim from a review we opened, dated 2024-10-01 or later, and every code is published on the reviews page.