Stop-loss on time
A creative that died on Tuesday should not spend until Thursday. The agent checks every batch each morning and drafts the stop the day it qualifies.
Picaso SV reads your ad data daily, tracks every creative batch through its test lifecycle, applies your team's rules and drafts the actions your ops team sends. Media buyers approve instead of digging through dashboards.
| Batch | Type | Day | Spend | Purchases | CPA | Status | Origin |
|---|---|---|---|---|---|---|---|
| v12.2 | Video | D3 | $412 | 19 | $21.7 | Scale | Request #214, Sep 29 |
| v12.1 | Video | D4 | $286 | 3 | $95.3 | Stop-loss | Request #214, Sep 29 |
| v8.0 | Text | D1 | $58 | 1 | $58.0 | Testing | Request #221, Oct 6 |
| v5.3 | Image | D7 | $730 | 31 | $23.5 | New strategy | Request #209, Sep 22 |
Sample data. Layout matches the production dashboard.
One pipeline, run on schedule. The agent drafts, a person approves, nothing spends on its own.
A creative that died on Tuesday should not spend until Thursday. The agent checks every batch each morning and drafts the stop the day it qualifies.
Ads are grouped into creative batches and followed from launch through D0 to D3. Older versions roll off when a newer one arrives.
Rules live in a file the team owns, not in someone's head. Thresholds, exceptions and review days are explicit and versioned.
D3 spend ≥ 2 × target CPA and purchases = 0CPA under target for 3 consecutive daysD0 to D2: no action unless spend > 3 × targetscaled ad with 7+ days data earns a new strategyThe draft uses the exact format your ops channel already reads, so it is pasted, not rewritten.
Product A, Oct 8
STOP NAV v12.1 ad 4, ad 6
reason: D3 $94 spend, 0 purchases
SCALE NAV v12.2 ad 2, budget +30%
reason: CPA $18 vs target $25, 3 days
The agent scans release notes in Slack and links each batch back to the request that created it. No more "which brief was this?"
Nothing goes to the ops channel without a click. Every decision is logged with the rule that triggered it.
The job is not to find one winner. It is to run more batches, kill losers faster and keep the few that work. The chart is what the agent reports every Monday.
Sample month for one product. Hit rate 20 to 25% is typical for Meta creative testing.
"I ran paid social by hand for years. The rules were always the same, only the time to apply them was missing. So I wrote them down and built the agent that applies them every morning."
Meta Ads today, since that is where our own spend is. The pipeline reads a daily export from your reporting tool, so other platforms with a daily export can follow.
No. It drafts the action and the reason. A media buyer approves it, and your ops team executes it the way they do today. Nothing spends without a human click.
The product was built with Claude Code and runs on the Claude Agent SDK. Claude classifies new batches, matches them to Slack requests, applies the playbook with its exceptions and writes the action in your team's wording.
One working day for a team that already has a daily export and a Slack channel for creative requests. Most of it is writing down the rules you already use.
We are taking on a few e-commerce advertisers in Vietnam and Southeast Asia for pilots. Write to us with your monthly spend and the number of creatives you test per week.
admin@picasosv.io.vn