Intercom, Zendesk, Ada, and Sierra count a chat as resolved when it ends without a human handoff. That is containment, not resolution. We score your bot-handled conversations on one test: did the customer's problem actually get solved? €2,000. Fixed fee. 7 days.
EXHIBIT A -- THE FINDING
Illustrative sample -- not real customer data
Dashboard containment
71%
True resolution rate
38%
The gap
33 pts
33-point gap -- tickets the dashboard claims but the bot did not resolve.
This is the real report format with illustrative numbers. The company is fictional; the conversations are rewritten composites.
To commission this audit for your own bot, complete the request form. The fit check is free.
Free fit check. You pay nothing until we confirm your export is usable.
01 · Background
Three facts about the number on your dashboard.
Most bot dashboards call it “resolved” when the chat ends without a human. A customer who gave up counts the same as one who got help.
You review a few dozen conversations a month by hand. The failure modes hiding in the rest stay hidden.
A bot that invents refund policy or sends customers in circles doesn’t just look bad -- it creates tickets, refunds, and churn.
02 · Methodology
No customer logos to show you yet -- so we show the rubric instead. Every conversation in your sample 1 is scored on four questions.
TABLE 1 -- SCORING RUBRIC
Did the bot do the work?
Was the ticket handled entirely by the AI, handed off to a human, or abandoned? A handoff counts as a handoff, even if the bot had a good conversation first. If the customer gave up and left, that is a failure, not containment.
Was the problem actually solved?
A resolved ticket is one where the customer’s request got what it needed: an answer that matched their question, a fix, a refund issued, an order changed -- within the thread. Not “the bot replied.” Not “the customer didn’t write back.”
Who deserves the credit?
When a ticket involves both the bot and a human, we split it. If the bot answered and the human fixed the actual problem, the bot did not resolve it. Bot-resolved, human-resolved, and unresolved are tracked separately.
What is the true resolution rate?
True resolution rate = tickets the bot fully handled AND fully resolved, divided by all tickets the bot touched. Shown next to the containment number your dashboard reports, with exactly which tickets make up the gap.
Clear rules, same answer every time. Bot never replied = unresolved. Handoff before the issue was addressed = not bot-resolved. Customer wrote “that worked” = resolved. Ticket reopened within 7 days on the same issue = original verdict re-scored as unresolved.
The customer stopped replying after the bot's answer -- silent is not solved 2, counted as “unclear” and shown separately. Partially correct answer the customer accepted -- flagged, with the excerpt. Two questions asked, one answered -- unresolved, flagged. You see every flag. Nothing goes into the score quietly.
EXHIBIT B -- ANNOTATED CONVERSATION
Illustrative composite -- rewritten and redacted
CUSTOMER (a)
My order still hasn't arrived and it's been two weeks.
BOT (b)
I've checked on this for you -- your package is on its way and should arrive soon.
[customer never replies] (c)
Vendor verdict: Resolved.
Our verdict: Unclear -- the customer went quiet for 48h and the bot logged “resolved.” We marked it unclear. Silence is not solved.
NOTE A
The request is specific: a late order. The only test that matters is whether the thread resolves it.
NOTE B
No tracking number, no date, no action taken. A status-shaped sentence, not an answer.
NOTE C
Forty-eight hours of silence, logged as “resolved” by the vendor. We mark it unclear -- silence is not solved.
03 · Procedure
No API keys. No code changes. No new tools to learn.
Export the conversations your bot handled from Intercom or Zendesk as a CSV. We send you a column-mapping template so it lines up with our checks. The fit check is free -- you pay nothing until we’ve confirmed your export is usable.
Once the audit fee is paid, deterministic checks score every conversation in your sample on handling quality, outcome, attribution, and verdict. Anything ambiguous -- sarcasm, multi-issue chats, edge cases -- gets flagged for a human instead of guessed.
Within 7 days: your true resolution rate with the math shown, the ranked causes of failure, your ten worst conversations with excerpts, and a walkthrough of what to fix first.
04 · Report contents
A written audit of what your bot actually does with real tickets.
Resolved vs. contained, with the calculation shown for every number. If we can’t defend a figure, it isn’t in the report.
Handling quality, outcome, attribution, and an overall verdict -- pass, needs review, or fail.
The reasons your bot fails, ordered by how often they happen, so you know what to fix first.
Full excerpts of the chats where your bot did the most damage, with notes on what went wrong in each one. This is usually the part teams act on first.
Every report documents what was checked, how many conversations were flagged for human review, and where the method is weak.
Your data is processed on EU-hosted infrastructure, encrypted in transit and at rest, and never used to train models.
Every audit is run personally by PrimoQA's founder, a former Meta support-ops architect.
05 · Fee
No subscription. No per-seat math.
Resolution audit -- bespoke sample, sized to your volume
€2,000.00
True resolution rate, with the math shown
incl.
Verdicts across handling, outcome, and attribution
incl.
Ranked list of failure causes
incl.
Ten worst conversations with excerpts
incl.
Walkthrough call on what to fix first
incl.
Total due
€2,000.00
Terms -- invoiced after the free fit check confirms your export is usable. The 7-day audit starts on receipt of payment. Late delivery: €200 back per day we're late.
Most teams run this 60-90 days before renewal. Bring this to your renewal negotiation.
€2,000 invoiced upfront, only after the fit check passes.
Your real decision is between the dashboard you already have and an independent check.
Who defines “resolved”
VENDOR DASHBOARDThe vendor
INDEPENDENT AUDITWritten rubric, published above
Silent customers
VENDOR DASHBOARDCounted as resolved after 24h
INDEPENDENT AUDITCounted as “unclear,” shown separately
Reopened tickets
VENDOR DASHBOARDUsually still counted as resolved
INDEPENDENT AUDITRe-scored as unresolved
Per-conversation verdicts
VENDOR DASHBOARDNo
INDEPENDENT AUDITYes, every ticket
Who you can argue with
VENDOR DASHBOARDA help article
INDEPENDENT AUDITThe person who scored it, on a call
06 · Terms
Before we invoice anything, we run a free fit check on your CSV. If the data is missing, truncated, or corrupted, we tell you exactly what’s wrong and you pay nothing. No invoice is sent.
The audit is delivered within 7 days of payment. For every day we’re late, we rebate €200.
Within 14 days of delivery, send any ticket ID and your reasoning. We re-score it in writing. If re-scoring moves your true resolution rate by more than 3 percentage points, the full re-run is free.
If it contains no gap breakdown, no flagged tickets, and no fix you didn’t already know, tell us within 14 days and we refund €1,000. We keep half because the scoring work was done.
07 · Cover letter
PrimoQA
SUPPORT BOT RESOLUTION AUDIT
September 2026
Dear reader,
I started PrimoQA after watching the same scene repeat: a support leader presenting a resolution rate they privately didn't trust.
Not because anyone was dishonest -- because the dashboard gave them one number and no way to check it.
Here's the problem in one sentence: vendors define “resolved” as “the chat ended without a handoff.” Your customers define it as “my problem got solved.” Those are different numbers, and renewals get signed on the first one.
So this audit does the unglamorous work: read your conversations, score each one against written rules, and show the math. Every verdict traces back to the customer's own words. I run every audit personally -- the scoring, the flagged-case review, the fixes list. If your bot is genuinely doing well, the report will say that too. I'd rather hand you a good number you can trust than a bad number you already suspected.
PrimoQA is new. I don't have a client list to show you, and I'd rather say that plainly than dress it up. Judge the method, not the logo wall.
Iulian Stanescu
Founder, PrimoQA -- hello@primoqa.com
08 · Questions & answers
A1 -- A CSV export of conversations your bot handled, from Intercom or Zendesk. We send you a column-mapping template so your columns line up with our checks in a few minutes. No API keys and no code changes.
A2 -- We set the sample after the free fit check, based on your volumes: a random, stratified pull across the audit period, large enough that the headline rate holds to ± 5 points at 95% confidence, and small enough that every sampled conversation gets human review against the rubric. For most teams that is 400 to 800 conversations. The sampling plan -- period, strata, exclusions -- is printed in the report, so you can see exactly what the numbers stand on.
A3 -- Dashboards usually report containment: the chat ended without a human stepping in. We check whether the customer’s problem was actually fixed, using the conversation text and your ticket fields. Those two numbers are rarely the same -- that’s the point of the audit.
A4 -- After the free fit check. Once we’ve confirmed your export is usable, we invoice the €2,000 upfront, and your 7-day audit starts as soon as payment is received.
A5 -- Straightforward conversations are scored by deterministic checks. Anything ambiguous is flagged for human review rather than guessed, and every report shows the split between decided and flagged conversations. The method and its limits are documented in the report itself.
A6 -- Yes. Your export is processed on EU-hosted infrastructure, encrypted in transit and at rest, and never used to train models. We’re working toward SOC 2 certification.
A7 -- You get the report plus a walkthrough call on what to fix first. If you want the audit run on a regular schedule, mention it in your request.
COMMISSION
Know your true resolution rate in 7 days.
Fixed €2,000 -- results in 7 days. No sales call required.
One form. One export. One walkthrough call. That's the whole ask on your side.
09 · Requisition
One form. We reply within one business day, starting with the free fit check -- you pay nothing until your export is confirmed usable.