- 3 AugAI agent answers the refund questionTicket closed, no handoverRESOLVED
- 11 AugSame customer, same questionRepeat contact on the same issueREPEAT
- 14 Aug212 more customers ask itLoop opens. Outcome set: contacts on this issue down 40% for 30 daysLOOP OPEN
- 4 SepCause fixed at the sourceArticle rewritten; an expert approved it, on recordFIX
- 4 OctOutcome test passesContacts on this issue down 81%CLOSED
OnClarity Insights
Resolved isn’t fixed.
Research, standards and tools on what customers keep contacting you about, and whether anything actually changed. Public data and stated method, free to read.
RESOLVEDREPEATLOOP OPENFIXCLOSED
Standard
The Proof Standard
How to tell whether AI in customer service fixed anything: what counts as a repeat, when a loop opens, who grades.
Open standard, v1
Read
Research
OnClarity Labs
One finding a week from public complaint data. One number, one chart, one rule.
6 findings, method published
Read the findings
Report
How accurate does AI need to be before it talks to customers unsupervised?
There is no single accuracy number for AI in customer service. Set the bar per type of contact by what a wrong answer costs, and decide who signs each one off.
6 min read
Read
Report
Will our customers accept talking to AI?
Customers accept AI that fixes the problem, keeps a person one step away and gives them no reason to come back. What the data says, and how to measure it.
6 min read
Read
Report
AI models are becoming commodities. The harness is what you’re buying.
Leading AI models are converging and getting cheaper. What makes a customer service AI agent work is the harness around the model. What it is and what to ask.
6 min read
Read
Report
On-premise AI: everyone asks for it. When is it worth it?
Is on-premise AI worth it for customer service? Four deployment steps, what each protects and costs, who needs it, and the questions to answer before you choose.
7 min read
Read
Report
Build vs buy for AI customer service: the full inventory
Build vs buy AI for customer service: the 15 components you would build and run, a three-year cost formula, what buying includes, and when building is right.
22 min read
Read
Report
How do we grow without cost to serve growing with us?
Cheaper contacts cut cost to serve once, then growth brings volume back. Track contacts per customer, rank savings by how long they last, report it monthly.
6 min read
Read
Report
How do you stop an AI agent from making things up?
You can’t make an AI agent incapable of inventing, but you can stop invented answers reaching customers. Five controls, a test for each, and a Monday checklist.
6 min read
Read
Report
The AI bill is due. Customer service is where it gets paid
Service teams put more of their budget into AI than any other function, and few can show a return. Where the AI bill gets paid, and the number finance will accept.
6 min read
Read
Report
If AI costs nearly as much per contact as a person, where is the saving?
When AI costs nearly as much per contact as a person, the saving is in contacts that stop. Why costs converge, and one calculation to redo yourself.
6 min read
Read
Report
Surveys vs conversations vs public feedback: which tells you first?
Three sources of customer insight, scored on nine criteria with a reason in every cell, and the order that shortens time to know.
9 min read
Read
Report
AI in customer service: faster answers, the same problems
Independent 2025 and 2026 research: spend is up, satisfaction is flat, returns are rare. Repeat cost is the missing number.
Evidence review, 18 min read
Read
Framework
Know, Fix, Prove: a framework for customer problems that stay fixed
A customer service framework in three clauses, five stages and three numbers, each with an owner and an exit test.
10 min read
Read
