- 1
Write your rules
Each line of your scorecard becomes one rule: something that must happen, or something to watch for. Mark how serious a failure is, and limit it to the channels, teams or situations it’s for. Assay shows what each answer leads to before you save, and suggests clearer wording when a rule is vague.
“Before cancelling, the agent tells the customer about any cancellation fee.” Critical, phone calls only. - 2
Bring in conversations
Connect Zendesk, Freshdesk, Intercom, Front or Gorgias and finished tickets arrive every 15 minutes, with the agent who handled them. Connect tl;dv and recorded customer calls arrive too. Calls, chats and emails can also be dropped in an Amazon S3 folder, pasted in, uploaded as a file or sent through the API.
Helpdesk: 214 tickets imported this week. - 3
Every conversation is checked
Each rule is answered yes, no or not sure, with the line that decided it highlighted and one plain sentence on why. Rules that don’t apply are skipped. Clear results are recorded; unclear ones go to a person.
Failed. The agent cancelled the plan without mentioning the £68 fee. - 4
Work the queue
Failed and unclear conversations wait in one queue, with a small random share of passed ones as spot checks. Turn on blind review to decide before you see Assay’s answer, and add a reason when you disagree. Every decision stays in the result’s history.
7 conversations need you today, 1 of them a spot check. - 5
Send results where you work
Each result, with the lines behind it, goes to your CRM, helpdesk or data warehouse as soon as it’s decided or changed. Change a rule and check the affected conversations again when you’re ready.
conversation.checked: CRM-4821 failed “Cancellation fee explained”, critical. - 6
Coach the pattern
See each agent’s pass rate, how it moved since last month and the rule they fail most. One click writes a short coaching note from their results, ready for the next one-to-one.
Priya: refunds over £50 need a supervisor’s approval before anything is promised.
What happens to each conversation
Calls, chats and emails arrive
Imported from your helpdesk or bucket, uploaded, or sent through the API.
Checked against your scorecard
Each check is answered yes, no or not sure.
Clear results are recorded
With the line in the transcript that decided them.
Uncertain ones go to a person
They land in your review queue.
What your reviewers see
Refund query, J. Patel
Call, 9 min
Late payment, Ashworth Ltd
Call, 14 min
Cancellation request
Chat, 22 messages
Delivery complaint
Email thread, 5 messages
Account update, M. Owen
Call, 4 min
Refund query, J. Patel
CustomerThe charge is still on my statement and nobody has called me back.
AgentI’ll refund it today, you have my word.
CustomerRight. I’ll hold you to that.
Refunds approved by a supervisor failed on line 16
DisagreeAgreeAccuracy by check
Agreement with your reviewers, last 30 days
- Identity confirmed98%
- Complaint recognised93%
- Refund within policy96%
- Call summarised at end88%
Each check is one sentence. You can see exactly what it does.
Say whether something must happen or is something to watch for, how serious a failure is, and which conversations it applies to. Before you save, Assay shows what each answer leads to, so nobody has to guess what a check will do to the numbers.
Written the way you’d explain it to someone new. Assay suggests clearer wording if it’s vague.
Checked by Assay
yes, no or not sure
Coach the pattern, not the one bad call.
Every agent’s pass rate, how it moved since last month and the check they fail most. Pick someone and Assay writes a short note for their next one-to-one, from the results it already has.
Choose an agent to see their coaching note.
And after every check
- Every result says why
- Each failed or unclear result comes with one plain sentence explaining it, next to the line that decided it. The reason is written from the result; it never changes it.
- Passes get double-checked too
- A small random share of passed conversations also goes to a person, who can decide before seeing Assay’s answer. That keeps the agreement figure honest.
- Tickets arrive on their own
- Connect your helpdesk and finished tickets are imported every 15 minutes, with the agent who handled them. Or drop transcripts in an Amazon S3 folder, paste them, or send them through the API.
- Critical failures stand out
- Mark each check as info, warning or critical. A missed greeting never looks the same as an address changed without checking who was calling.
- Nothing is rewritten
- Editing a check starts a new version. Every result keeps the version it was judged against, and every change a person makes is kept with who, when and why.
- Results go back to your systems
- Each result, with the lines behind it, is sent to your CRM or helpdesk as soon as it’s decided or changed by a person.
Try it on your own calls
Tell us about your team. We’ll reply within two working days to set up a trial with 50 of your transcripts and your current scorecard.
- No card or contract for the trial
- Your conversations are never used to train models
- We go through the results with you
Thanks, we’ve got it.
We’ll be in touch within two working days.