Overview

Last 30 days · ending Jun 12, 2026
☆ Save view ✦ Ask 🧪 1 experiment 🔔3
⌘K
Yesterdayin expected range
$6,646
1,186 conversions · within the holdout-era norm
June pacing→ 7% under plan
93%
pacing alarm below has the fix
Health
checks passed · 0 disapprovals · guardrails green
Decisions needed
start review mode →

⚡ Needs a decision today

📉 Pacing alarm — 7% under plan

June is tracking ~$21k under the $300k plan. Best homes for the extra budget, pre-checked against guardrails:

Teacher AI · +$280/day headroom
$2.97/signup · best efficiency in the account
Conquest · losing 33% IS to budget
1,573 cross-funnel signups hiding here
Hold the rest until Jun 16
the holdout verdict may free brand budget anyway
WAIT

🌙 While you were out

System · /check · 9:02
Daily health check passed — pacing normal, 0 disapprovals.
🧪
Experiment
Holdout day 8 of 12 — organic recapture steady at ~250 signups/day.

Shortcuts:
Monthly spend plan$278.6k / $300k · on pace
Cost / signup vs ceiling$3.19 / $3.50 ✓
Q2 pipeline target$2.4M / $2.0M · exceeded

Spend & conversions — last 60 days

🪜 The funnel — this window

🧪 Live experiment

Brand-search incrementality holdout — brand campaign paused since Jun 4; protected by a guardrail until the verdict.
📉
Early read: brand clicks are only ~35–55% incremental.
💰
~$4,000/mo of related fixes deliberately held so they don't contaminate the test.

⚡ Needs your attention

Channels

Click a channel to focus the whole dashboard on it.

Top campaigns

Every action in Compound is attributed — an agent, the orchestrator, or you. Nothing happens anonymously.
Click a campaign name for the full drill-down. CPA shows “—” where it isn’t the right scoreboard.

What-if: reallocate budget estimate — directional only

🔑 This is where the Keyword Researcher hunts. Converting terms get promoted into managed keywords (bids and ads under explicit control); waste gets blocked with surgical match types; brand-adjacent terms stay held while the holdout runs.
Auto-written by the reporter agent at the end of every cycle.

Weekly brief — Acme Learning

Week of Jun 6–12, 2026 · written by Compound · reviewed by your strategist

The week in one paragraph

A quiet, deliberate week: spend held at the reduced level you set in late May (~$48k across Google + LinkedIn), signups stayed efficient at $3.19 — comfortably under the $3.50 ceiling — and the brand-search experiment entered its final stretch with a verdict due Jun 16. The June 11 audit queued 12 recommendations worth ≈$38,000/mo; none have been applied yet, pending your approval.

What we found

  • $21,000/mo flagged: Demand Gen bought 177,000 clicks for 60 signups (~$350 each). We queued an analytics trace before judging — junk clicks or broken tracking, the data will say.
  • $12,000/mo of proven volume: ~7,600 signups/60d are flowing through unmanaged search terms. Fourteen promotions are queued.
  • A creative formula: compliance-proof ads (FERPA, 20k districts) earn 16.8% CTR vs 8.4% generic — now feeding every new B2B brief.

What we deliberately did NOT do

~$4,000/mo of brand-negative fixes are held — applying them mid-experiment would contaminate the holdout. The fix is queued for the moment the verdict lands.

What happens next

Jun 16: holdout verdict → brand budget decision + held fixes unlock. Pending your approval: the 12 audit recommendations (the five negatives are one click — they're surgical and risk-free). The retro grades this cycle's changes ~Jun 18.

Every number above links to its evidence in the dashboard. Nothing in this brief was written by a human — and nothing in your account was changed without one.

Experiments are first-class objects: designed up front, protected by guardrails, judged on evidence.

🧪 Brand-search incrementality holdout RUNNINGverdict in 4 days

Jun 4started
Jun 16verdict due
35–55%early incrementality read
$4,000/mofixes held to protect the test
Statistical confidencedirectional now → verdict-grade by Jun 16 (~4 more days of data)

Question: how many brand-search signups are truly incremental vs recaptured free? Design: pause Brand_Search_Demo, watch total signups vs brand volume. The campaign is guardrail-protected — even an approved enable request is blocked until the verdict. Early read: other campaigns recapture ~250 signups/day organically.

📉 Budget-cut staircase COMPLETE · May 20–26

Question: does cutting ~33% of spend hurt efficiency? Verdict: CPA held flat while volume tracked spend — the account had healthy headroom. Learning recorded so no agent ever misreads the dip as decay.

🎨 Compliance-creative split COMPLETE · Apr–May

Question: what message wins district queries? Verdict: compliance-proof creative at 16.8% CTR vs 8.4% generic — 2×. The formula was adopted account-wide and now writes the Creative Studio briefs.

June 2026

plan $300k · tracking 93%

🌡 Seasonality

Teacher searches dip hard from late June (summer break) and ramp from late July (back-to-school). The trend researcher pre-stages the B2S budget plan in July — and won't let anyone misread the June dip as decay.

📌 Coming up

🧪
Jun 16 — holdout verdict; unlocks ~$4k/mo of held fixes
⚖️
Jun 18 — retro grades this cycle's changes
🎪
Jun 22 — ISTE conference; district traffic spike expected

Auction landscape — impression-share overlap

Who you're meeting in the auction and which way they're trending. Watched continuously by the competitor analyst.
📣 New this week: a competitor shipped a comparison page targeting “vs” queries. The landing-page auditor was notified, and a counter-brief is drafted and waiting in Creative Studio.
Keyword cluster: “exit ticket generator”
est. 14k searches/mofits the $3.50 ceilingKeyword Planner
Adjacent to your converting quiz/rubric cluster — similar intent converts at ~$3.10. Drafted as its own ad group with three RSAs.
Keyword cluster: “ai grading rubric”
est. 9k searches/moKeyword Planner
Search demand grew 3× this semester; your rubric-maker term already converts at $3.36 — this is the head term above it.
Audience: lookalike of demo-booking admins
reach ~84kLinkedIn
Built from the 24 lead-form admins + CRM demo bookers. Tighter than the current role targeting that produces engagement clicks.
Channel gap: TikTok teacher-creators
competitor presence detectedtrend researcher
Two competitors run teacher-creator Spark Ads; you have zero presence. Paired with Proclaim's UGC pipeline, this is a fast test.
Budget headroom: Conquest losing 33% IS to budget
+$300/day capacityauction insights
Strong cross-funnel economics (1,573 signups) and it's leaving a third of its auctions on the table.
Health score
86/100
2 issues open · checked daily by the account auditor
Conversion scoping
verified Jun 11 — campaigns count only their own goals
CRM join
94%
of deals matched to ad UTMs

Open issues

⚠️

Demo form double-tracked (GA4 + HubSpot)

Demo counts read ~2× high until one source of truth is chosen.

fix queued · rec-09
⚠️

/ai-tool/presentation-maker — 1,900 clicks, ~0 conversions

Siblings convert 16–23% of clicks; this smells like a broken tag, not a bad page.

trace queued
⚠️

3 LinkedIn ads missing UTM parameters

Their landing clicks are invisible to the CRM join until tagged.

Google conversion goals correctly scoped

Verified during the Jun 11 audit.

No disapprovals, no policy flags

All ads serving normally.

Managed spend · 30d
$530k
across 3 clients
Clients
3
+ 1 onboarding
Pending approvals
23
all human-gated
Learnings promoted
6
client → global library

Cross-client learnings — one brain, many accounts

When a learning generalizes, it's promoted to the global library and every client benefits on the next cycle.
Promoted · found at Acme
CRM conversions must be PRIMARY for smart bidding to use them
AppliedSaved Lexbridge the same blind-bidding mistake during onboarding.
Promoted · found at Acme
Conquest terms masquerade as waste — check strategy before judging
AppliedStopped a "cleanup" that would have paused Northwind's best prospecting campaign.
Promoted · found at Northwind
Budget-capped winners are the cheapest growth available
AppliedSurfaced $13k/mo of capped winners across the portfolio, Acme included.
Jun 11 audit · 5 agents in parallel nothing executes without your approval
CONCEPT PREVIEW — Proclaim Studio integration

The loop the partnership closes: Compound's learning loop discovers which creative wins and why → the insight becomes a structured brief → Proclaim renders finished variants at scale → they land in the ad account paused, awaiting one-click approval → performance feeds the next brief.

1 Insight — from the learning loop

Verified · Jun 11 audit

District queries reward compliance-proof creative

Same campaign, same audience: the ad leading with FERPA compliance + “trusted by 20,000 districts” + admin dashboards earns 16.8% CTR vs 8.4% for generic teacher-freebie creative.

Formula: lead with compliance & scale proof, not features.

📈 Trending now · scraped nightly via Apify
🪝 “POV: my admin asked…” hook format +340% in teacher TikTok
🎵 Trending sound flagged licensed-safe for ads use
🎬 Fast carousel cuts beating talking-head 2:1 this week

All three folded into the brief → so Proclaim renders into the current moment, not last quarter's.

2 Brief — auto-generated

AudienceDistrict admins & curriculum directors (B2B)
MessageCompliance-first: FERPA / SOC 2, 20k districts, admin visibility
Proof16.8% CTR formula from the learnings log · district logos · privacy score
Formats15s video · 30s UGC-style · 6s bumper · RSA copy set
Trends“POV: my admin asked…” hook · licensed-safe trending sound · fast carousel cuts (Apify, last night)
Landing/curriculum-intelligence
GuardrailBrand kit + claims allow-list enforced by Proclaim governance

3 Variants — rendered by Proclaim

🏛️0:15
District Trust READY
“20,000 districts already said yes. Here's the 15-second version of why.”

🎤0:30
Teacher Voices · UGC RENDERING
“My admin asked if it was FERPA-compliant. It already was.”
0:06
Compliance Counter READY
FERPA ✓ SOC 2 ✓ 20,000 districts ✓ — six seconds, three proofs.

The loop closes: once live, each variant's CTR, cost-per-signup, and CRM pipeline flow back into the learning loop. Winners get scaled, losers get explained, and the next brief starts from evidence — not taste.

Copy Studio — everyday text ads claims allow-list enforced

Headline variants in the brand voice, char-validated, every claim checked against what the landing pages can prove. Built on the 16.8%-CTR compliance formula.

Variant performance — live creative, watched for fatigue

The loop doesn't stop at launch: every variant's CTR is tracked, and decay automatically queues the next Proclaim brief — fresh creative before performance falls off a cliff.
Fatigue detected: “Compliance Counter” CTR fell 12.4% → 8.1% over six weeks. A refresh brief was queued to Proclaim automatically — same formula, new execution.
The same operation, drawn as a flow — click any node to inspect & configure it, drag to rearrange, or run a full weekly cycle. Open full screen ↗
The account's institutional memory. Every agent reads this before touching the account — which is why a chart dip doesn't trigger a panic, and an “obviously wasteful” conquest campaign doesn't get paused. Entries are never deleted; reversed ones are marked superseded.
Trust isn't a promise here — it's enforced in code. The same gates apply to the AI agents, the CLI, and this dashboard. There is no bypass path.

The hard rules

No write without approval

Every change is a recommendation until a human clicks approve. Approving one change is never approval for the next.

🔒

Guardrails are law

Daily budget cap ($10,000), max budget move per change (±30%), protected campaigns. Violations return GUARDRAIL BLOCKED — to agents and humans alike.

Everything new lands paused

New campaigns, ad groups, and ads are created in a paused state for final review. Always.

🧾

Everything is logged

Every applied change records when, what, why (the hypothesis), and from where — so the learning loop can grade it later.

📡

Real data or loud failure

If a source breaks, the dashboard says so. No stubs, no estimates standing in for facts.

What a write actually looks like

$ ./ads set-budget --client acme --campaign "Teacher AI_Search_Signup" --amount 1400
DRY RUN — no changes applied.
  Campaign: Teacher AI_Search_Signup
  Budget: $1,050/day → $1,400/day (+33%)
──────────────────────────────────────────
✗ GUARDRAIL BLOCKED — change exceeds max_budget_change_pct (30%).
Nothing was modified. Adjust the request or update guardrails with the account owner.

$ ./ads set-budget --client acme --campaign "Teacher AI_Search_Signup" --amount 1330 --apply
✓ APPLIED — budget $1,050 → $1,330/day (+27%, within guardrails)
logged → clients/acme/changes.log (hypothesis: capture 5% lost-IS-to-budget; grade in 7d)

The change log feeds the learning loop

Illustrative entries — every line is later graded against real before/after data by the retro cycle.

WhenActionHypothesisVerdict (after 7d)
Jun 18*add-negative “edu-app” (EXACT)~$630/mo waste removed, CTR upgraded by /retro
Jun 18*promote 14 search terms → keywords~$12,000/mo under explicit controlgraded by /retro
Jun 16*brand holdout decisionkeep / restart brand spend on evidenceexperiment

*Future-dated: these are the queued recommendations as they'll appear once approved.

Add a client

From OAuth to first audit in about 20 minutes.
Client name
Connect ad accounts (OAuth — read access first, write unlocks after guardrails)

Slide preview

Add a memory annotation

Pin context the APIs can't see — agents read it before judging this window.

Day
What happened

🚀 Plan a launch

Compound stages the budget lift inside guardrails, queues creative briefs to Proclaim, arms watch alerts, and reports incremental impact afterward.

Name
Window
Budget lift
↑↓ navigate↵ runesc close