Human-AI Interaction Assessment Suite · AXA
AI Conversation Quality Review
See the conversation from the customer’s side—not only whether the bot completed a task, but whether it understood what the person was trying to accomplish, kept the exchange moving, and responded well when the conversation became complicated.
Choose a Review Ask Susie a QuestionWhat the review adds
Customer meaning
What the customer said directly, what their next replies suggest they were trying to reach, and where the assistant’s answer did—or did not—meet that need.
Conversation evidence
Findings are tied to the messages that support them, so your team can see why each strength, concern, or recommendation appears in the report.
Practical next steps
Response changes your team can test against the product’s intended behavior, without treating every unhappy customer or policy-based refusal as a bot failure.
Two ways to begin
Founding Client Rate · First 10 companies
Quick Conversation Checkup
$99
One authorized, de-identified customer-and-bot conversation, up to 20 total messages and 1,500 words. One founding-rate checkup per company.
- Limited to the first 10 companies
- Concise, human-reviewed, evidence-linked report
- What appeared to work well
- Where understanding, clarity, momentum, or customer experience changed
- Two to three practical improvements to test
- One clarification by email
For patterns across several conversations
Five-Conversation Review
$1,000
Up to five conversations from one chatbot or product, each up to 40 messages and 1,500 words.
- One consolidated, evidence-linked report
- Cross-conversation patterns and priorities
- Prioritized recommendations
- One review meeting
- One clarification round
Additional conversations, longer records, specialist evidence checks, and other assessment-suite services are quoted separately.
What you provide
An authorized, de-identified conversation transcript; a short explanation of what the customer was trying to accomplish; and any relevant bot instructions, company policies, or product constraints you want considered. The review evaluates the visible exchange and supplied context. It does not claim access to the model’s private reasoning.
What you receive
A clear account of how the exchange unfolded, supported by specific conversation evidence. The report separates supported strengths from genuine friction, considers the customer’s changing language and apparent experience, and recommends changes only where the record supports them.
Why Susie
Susie Sarkozy is a Human-AI Interaction Research Analyst with more than 30 years of customer-service experience. She brings the buyer’s perspective to the technical conversation: how people ask indirectly, repeat themselves, correct a misunderstanding, become more or less engaged, and decide whether the assistant is helping them get somewhere.
Before work begins
Scope, authorization to review, human and AI-assisted processing consent, submission method, delivery date, and any sensitive-data requirements are confirmed before the transcript is processed. Do not submit personal identifiers, payment data, health information, student records, or other regulated production data through ordinary email.
Research and review standards
These are developmental, evidence-limited reviews. They are not legal advice, accredited certification, clinical assessment, security testing, or product-wide validation. Findings apply to the reviewed records and supplied configuration. Client materials are not used in research, publication, or promotional case studies without separate permission.
Start with One Conversation Request a Larger ScopeSusie@SusieSarkozy.com · About Susie