← All screens
What the agent can put in a conversation
Every reply is built from these blocks. The agent picks them; the operator never learns a syntax.
Answer
Prose with sources. Never a bare claim.
Spend is pacing 8% above plan. At this rate you finish the month at ¥109k against a ¥101k budget.[1]
1 source · campaign daily report
Metric tiles
Two to four numbers, each against its target.
Spend (MTD)
¥32.8k
▲ 8% vs plan
ACoS
34.2%
target 30%
ROAS
3.42
target 3.30
Data preview
The first rows, then a way into the real table.
Zero-conversion search terms 3 of 12
wireless earbuds¥3,200
bluetooth earphone¥2,100
earphone repair¥1,800
Total ¥9,340 Open full table →
Chart preview
Same chart library as the dashboards, so they match.
50% 40% 30%
ACoS vs target, daily Open dashboard →
Navigate
Buttons that move the main pane. No copied IDs, no hunting.
Take me to
SP-Auto-Discovery
Search terms
Budget pacing
System action
Anything that writes stops here for a human.
NEEDS APPROVAL Writes to Amazon
Lower bid on 2 keywords
wireless earbuds ¥120 ¥102
Approve and run
Cancel
Action result
Partial success is a normal outcome, not an error.
2 applied, 1 skipped 09:14
wireless earbuds¥120 → ¥102
bluetooth earphone¥95 → ¥81
cheap earphonedata changed
Undo
Previous values saved
Question back
When the ask is ambiguous, the agent asks instead of guessing.
"Cut the wasted spend" could mean two things. Which do you want?
Lower bids on the 12 zero-conversion terms
Add negative keywords so they stop serving
Working
Which tools ran, not a spinner. Collapses when done.
Analysing search terms… 6.4s
Read 24 campaigns
Compared to previous 30 days
Grouping by search term
Follow-ups
Two or three, never a wall of chips.
What changed since yesterday?
Draft this month's report
Run this weekly