Sample output · publicimpact.ai · no mocks · real actions, real day

Today, the Swarm doubled the prime ad budget.

without anyone clicking a single button

Scroll
One half of the Swarm

This page shows the Performance Brain, the half that watches.

The cluster section on the home page shows what the Swarm produces: research, content, ads, funnels, conversion work. Sixteen specialised agents creating output every day.

This page is the other half, the Performance Brain. The layer that measures everything the producing clusters create, compares it against goals, surfaces decisions for the operator, and tunes the next day's run. Production happens above. Observation, learning, and constraint live here.

▸ See the production side on the home page →

SWARM · publicimpact.ai · MISSION CONTROL ▸ live data · sample day · 09 May 2026 · 14:32 UTC · day 38
SWARM ALIVE
CRON 04:00 UTC
LOOPS 6 / 6
NODES 16
What the swarm did today · last 12h● LIVE
14:32funnel-abvariant B wins on /blog · CTR +14%DONE
13:01ad-bidCPC reduced · prime ad 2.40€ → 0.95€DONE
08:00insight-synbriefing generated for the operatorDONE
07:42seo-scan2 keyword cannibalization conflicts foundFLAG
06:30watchdogbudget check across 3 clients · all greenOK
04:00ingestion5 sources pulled · 247 new metrics in DBOK
Voice of the swarm
"The prime ad has been running for 7 days. CPC dropped from 2.40 € to 0.95 €, that's 28% lower. Recommendation: double the budget. Expected: +14 leads per month."
▸ insight-synthesizer · 09 May 2026 · 04:30 UTC · confidence 0.84
Page views
974
▲ 12%
Visits
182
▲ 8%
Edge req.
1.5K
▲ 5%
Deltas · 7d
CPC−28% ▼
CTR+14% ▲
CPL−18% ▼
Followers+3 ▲
Bounce−6% ▼
Sessions+11% ▲

▸ snapshot of one real day · keep scrolling to see how the layers work

01
Ingestion

Every morning, the swarm pulls everything fresh.

04:00 UTC. The ingestion layer reaches into every channel a client has connected: Cloudflare for reach, Search Console for queries, social platforms for engagement, ad accounts for spend. Every metric lands in one central database.

Without good data, no agent can do its job. So this layer is paranoid: every source is logged, every failure recorded, every gap visible.

04:00 UTC · daily ingestion run● LIVE
04:00cloudflarepage views, visits, edge requestsOK
04:00searchclicks, impressions, positionOK
04:00socialfollowers, engagement, reach · 4 networksOK
04:00adsspend, conversions, CTR · 2 platformsOK
02
Performance Brain

One central memory for every channel.

The brain is the long-term memory. Every metric ever pulled lives here: indexed by client, by date, by source. Daily snapshots. Monthly aggregates. Year-over-year deltas, automatically rolled up.

Most marketing reporting is broken because tools live in silos: one dashboard for ads, another for SEO, a third for social. The brain ends that. Everything is one query away.

SQL flow · time horizons
yesterday
rolling 7d
rolling 30d
monthly aggregates
year-over-year
all available to every agent in one query
03
Optimization Loops

Six loops, continuously scanning.

Loops run independently, on different schedules, looking for different patterns. Some scan every two hours, some daily. Each one writes its findings into the daily briefing if it spots something worth your attention.

The point isn't more data. The point is fewer surprises.

Six loops in sync● LIVE
every 6had-bidflags rising cost without conversionsOK
every 6horganicdetects follower drift, engagement shiftsOK
05:00funnel-abpicks winning landing-page variantOK
06:00seo-scanspots keyword cannibalizationOK
06:30watchdogchecks ad spend vs. budget capOK
07:00boost-bridgefinds organic posts worth boostingOK
04
Insight Synthesizer

The voice of the swarm.

Numbers don't decide. People do. Every morning, Claude Haiku 4.5 reads the entire Brain: KPIs, deltas, what changed, what's blocked, and writes a fresh briefing. No templates, no formula. Real synthesis, every day.

Two streams: one for the operator (with action items, technical detail, decisions to approve), one for the client (status, context, no homework). Same data, different audience.

Sample briefing · operator stream · 10 May 2026
"Instagram is carrying 63% of reach, Facebook is watching. 428 impressions, 27 interactions over 30 days, system is collecting baseline. Google Ads is waiting on the final platform approval (should clear this week), then the prepared rent-law and employment-law campaigns go live in parallel."
▸ written by Claude Haiku 4.5 · 04:30 UTC · client stream below
"Instagram is delivering most of the reach right now while we build the baseline. Google Ads is in the approval phase, once that's through, we expand reach significantly with the prepared campaigns."
05
Customer Cockpit

Your client sees what's happening.

Every client gets a portal. The swarm's voice. The activity feed of the last hours. The live KPIs from every connected channel. Not a monthly PDF. Not a reactive email. A live window into what's being done for them.

Most agencies hide their work. We show ours. Trust is built in transparency.

What the client sees · today● LIVE
14:32A/B testvariant B wins on /blog · CTR +14%DONE
13:01adsCPC on the prime ad reduced significantlyDONE
11:00social2 reels published · 47 + 23 likes after 3hDONE
08:00briefingdaily analysis ready · top win + recommendationDONE
06:00SEO2 articles indexed by Google overnightDONE
06
Operator Constraints

The Swarm respects your brief.

Autonomous does not mean unsupervised. The operator layer is where we translate your monthly brief into hard limits: how much budget per platform per month, what needs explicit approval before going live, what triggers an automatic pause. The watchdog scans these limits every day and stops anything that drifts.

You define what the Swarm is allowed to do. We configure it. You see the results live in your portal. That's the difference between an AI that works for you and one that works on you.

Budget governance · sample client
monthly_budget_eur500
spent month-to-date47
soft cap (alert)90% → 450 €
hard cap (auto-pause)110% → 550 €
approval requiredYES · operator confirmed 09 May
07
Operational Transparency

When something blocks externally, you see it.

Most agencies hide their dependencies. When Google takes three days to approve an ads account, when a Search Console handover is pending, when a brand asset hasn't shipped yet, you usually hear nothing. The work just sits.

We made the opposite choice. The Brain monitors every ingestion source and detects friction automatically: failed runs, missing approvals, pending integrations. Then it translates the technical reason into customer-friendly language in your daily briefing. "Google Ads waiting on platform approval", never DEVELOPER_TOKEN_NOT_APPROVED.

Auto-detected · operational status● LIVE
detected from D1
source: google_ads · status: partial
err: DEVELOPER_TOKEN_NOT_APPROVED
→ translated by Claude Haiku 4.5
"Google Ads is waiting on the final platform approval, should clear this week, then we activate the prepared campaigns."
surfaced in today's briefing · operator + client

Ready to see it run?

This system isn't a slide. It's running right now, on this brand. We can build it for yours in a few weeks.