Spend check · usage units (assists, credits or actions)
Simulated service desk
2026-05-01 to 2026-05-31 · Billed in assists from a 45,000-assist pool at $0.50 each (illustrative) · $3,262 · 6,523 assists × $0.50
$2.57Per held resolution
Spend in the period
$3,262
6,523 assists × $0.50
Assists used
6,523
14.5% outside production
Resolutions that held
559
of 672 claimed · 17% did not hold
Cost per resolution that held
$2.57
range $2.45 to $2.70
Pool used
69%
runs out 2026-08-05
From 2026-05-01 to 2026-05-31, the AI used 6,523 assists ($3,262), 14% of them outside production. Of the assists its conversations used, 49% went to conversations that were handed to a person, abandoned or came back as a ticket. Each resolution that held cost $2.57 (range $2.45 to $2.70). At this pace the pool runs out on 2026-08-05, before the term ends.
1What to raise
Finding
Assists
Amount
Detail
Assists used outside production
944
$472
14.5% of the period's assists ran in acme-dev, acme-test. Ask whether testing should draw from the production pool, and test with smaller samples.
Assists spent on conversations that did not resolve
1,403
$702
49% of the assists linked to conversations went to ones that were handed to a person, abandoned, or came back as a ticket.
The pool runs out before the term ends
5,465
$2,733
At the last 30 days' pace (212 a day), the 45,000-assists pool runs out on 2026-08-05, before the term ends on 2026-08-31.
2Where the assists went
Feature
Assists
Share
Amount
Outside production
Now Assist in Virtual Agent
2,871
44.0%
$1,436
0
AI agent: password reset
1,140
17.5%
$570
315
Incident summarization
964
14.8%
$482
237
Chat reply recommendation
868
13.3%
$434
223
Resolution notes generation
680
10.4%
$340
169
Instances: acme (5,579) · acme-dev (635, treated as non-production) · acme-test (309, treated as non-production). Names containing dev, test, uat, qa, sandbox, staging or demo are treated as non-production; pass a file with --nonprod to mark one yourself.
3What the AI's conversations delivered
The AI's conversations in the period, by what happened next. "Came back" means the same person opened a ticket about the same type of issue within 7 days.
Outcome
Conversations
Share
Assists
Amount
Resolved, and the person did not come back
483
56.3%
1,468
$734
Resolved, but the person came back
189
22.0%
555
$278
Handed to a person
129
15.0%
762
$381
Abandoned or unanswered
57
6.6%
86
$43
4Each resolution that held
Of the period's 672 claimed resolutions, 28.1% came back within 7 days, against 13.7% after your agents' resolutions. Beyond that normal rate, 16.7% did not hold (range 12.9% to 20.8%; audit confidence medium), so about 559 held.
$1,436 (conversation units) ÷ 559 resolutions that held = $2.57 per resolution that held (range $2.45 to $2.70). It is the number to compare across vendors and billing models.
How the platform closed the conversations it counted as resolved
Platform's reason
Claimed
Share
Came back
Did not hold
Topic Complete
247
36.8%
19.0%
6%
Auto Closed
174
25.9%
18.4%
5%
No User Response
146
21.7%
54.1%
47%
User Ended
83
12.4%
30.1%
19%
5The assists pool
Pool for the term 2025-09-01 to 2026-08-31
45,000
Used from 2025-11-10 to 2026-05-31 (all instances)
30,958 (69%)
Pace over the last 30 days
212 a day
Runs out on 2026-08-05, then over the pool by the term's end
5,465 · $2,733
Usage data starts on 2025-11-10, after the term began; use before that is not counted, so the pool may be further along than shown.
6Notes
237 interactions without the Virtual agent flag were left out.
The conversations carry no category, so a return was any ticket from the same person.
7Data lineage
File
Rows
SHA-256
incidents.csv
74,477
a5dd9e825b4798c2…
interactions.csv
4,978
6ece6678ea983285…
now_assist_usage.csv
6,183
4ee980773418388b…
Probative Labs 0.4.0 · settings 6df558ff604d · run 2026-10-17T05:00:49+00:00. Anyone with the same files and settings can rerun this check and get the same result.