The agent dashboard: 24/7 operation, transparent and with human approval

Live metrics (checked hourly)

• ROAS per channel
• CPA / CAC trend
• Budget utilisation (pacing)
• Pipeline value from paid
• Delivery status per campaign
• Open A/B tests

Automated A/B testing

A/B tests run automatically: the test agent launches variants (headline, CTA, audience, format), evaluates them for statistical significance, stops losers and promotes winners into the live campaign – without anyone having to set up a test.

Self-learning status

Every completed A/B test, every pacing hour and every daily ROAS trend feeds the learning loop: the orchestrator recalibrates its channel weights and target-CPA margins daily and writes derived rules into the standard policies of all campaign and bidding agents – week by week, its recommendations get more accurate without anyone fine-tuning by hand.

Actions that need your approval

• ! Budget increase > 20% per channel
• ! Pausing a live campaign > 24 h
• ! New targeting outside approved audiences
• ! Launching new landing pages

Live metrics (checked hourly)

• Impression share (Google + Bing)
• Search-volume drift per ad group
• Quality score trend
• Conversion rate per match type
• Feed error rate
• CPA trend 7/28 days

Automated A/B testing

A/B tests run automatically: the test agent launches variants of headlines, bidding strategies, landing pages and PMax assets, evaluates them for significance and scales only the winners. Every search campaign therefore carries fresh tests permanently – without manual testing.

Self-learning status

The learning cycle turns daily around search queries: from the performance of every keyword, match-type class and ad variant, the orchestrator learns which search intents actually convert – and automatically updates bid modifiers, negative keywords and quality-score forecasts in the agent rules for Google and Bing.

Actions that need your approval

• ! Bidding strategy switch (e.g. to tROAS)
• ! Budget reallocation > 25%
• ! Disabling whole ad groups
• ! New conversion events as primary goal

Live metrics (checked hourly)

• Rankings top-10/top-3 per cluster
• Organic traffic (trend)
• Core Web vitals
• Indexation rate
• AI visibility (GEO citations)
• Open technical fixes

Automated A/B testing

A/B tests run automatically: titles, meta descriptions and content structures are tested in variants (CRO methodology); click and ranking data decide which version stays.

Self-learning status

The orchestrator analyses ranking and AI-visibility trends per keyword cluster: which topics gain traffic, which content AI search cites – from this it updates content priorities, spots keyword-cannibalisation pairs early and sharpens cluster boundaries before they cost rankings.

Actions that need your approval

• ! High-volume content gets changed
• ! URL cluster restructuring (301s)
• ! Schema.org markup changes
• ! Removing existing content

Live metrics (checked hourly)

• CAC per audience
• CPM & frequency trend
• Creative fatigue score
• Learning status per ad set
• Policy warnings
• Scale factor (scale ad sets)

Automated A/B testing

A/B tests run automatically: formats, hooks, CTAs and audience variants are tested continuously. The test agent stops losers once significant, scales winners into dedicated scale ad sets and starts the next test round – social budget works on new variants every single day.

Self-learning status

From every creative-fatigue pattern, every hook A/B test round and every CAC trend per audience, the orchestrator learns which formats, hooks and messages work in which market – and sharpens the creative and audience agent briefings from that: today’s creatives are tomorrow’s learning patterns.

Actions that need your approval

• ! New creatives with brand claims
• ! Budget > approved daily frame
• ! Audience change > 30%
• ! Scaling into a new country/market

Live metrics (checked hourly)

• Conversion rate per landing page
• Running tests (significance)
• Drop-off per form step
• Scroll depth / click hits
• Load time (CLS, LCP)
• Won tests (Δ-CR)

Automated A/B testing

A/B tests run automatically: the test agent keeps 2–4 tests running in parallel, computes significance itself and promotes winners to the live variant. New test ideas are fed in by the agent team from the UX data.

Self-learning status

Every completed test, every heatmap anomaly and every form-exit session flows into the learning base: the orchestrator recognises recurring friction points across pages and re-prioritises the test backlog – the next hypothesis is born from the analysis of the last one, not from gut feel.

Actions that need your approval

• ! Changes to checkout/request forms
• ! Tests on pages with > 50% traffic
• ! New pricing or offer variants

Live metrics (checked hourly)

• Partner ROI per partner
• New partners / onboarding status
• Fraud flags (open)
• Commissions (planned vs. paid)
• LTV per partner channel
• Click-to-conversion rate

Automated A/B testing

A/B tests run automatically: landing pages, offers and bonus structures are variant-tested per partner channel – conversion and LTV data decide on scaling.

Self-learning status

The orchestrator learns from each partner’s ROI trajectory: which partner types deliver sustainable conversions, which show fraud patterns – from that it automatically recalibrates bonus logic and risk thresholds and spots new partner opportunities from market trends before they show up in the numbers.

Actions that need your approval

• ! Payouts above threshold
• ! Excluding a partner
• ! Changing bonus structure
• ! Approving a new partner product category

Live metrics (checked hourly)

• TACOS per ASIN
• Feed suppressions (count/reasons)
• Sponsored ads ROAS
• Ranking trend per keyword
• Review anomalies
• Budget utilisation per campaign

Automated A/B testing

A/B tests run automatically: listing variants (title, main image, A+) are tested per product; conversion data decides which version gets scaled.

Self-learning status

TACOS and feed-suppression data per ASIN feed the learning loop: the orchestrator recognises which listing changes lift organic rankings and which bids eat margin – and from that it automatically updates the prioritisation rules for sponsored ads and the feed-maintenance strategy.

Actions that need your approval

• ! Bid above daily limit
• ! Listing change on top-50 ASINs
• ! Feed change with price/stock impact

Live metrics (checked hourly)

• Reach & engagement per channel
• Follower growth
• Response times (comments/DMs)
• Trending formats (top 3)
• Posting coverage per market
• Escalations (open)

Automated A/B testing

A/B tests run automatically: formats, posting times and hooks are variant-tested; engagement data decides the default strategy.

Self-learning status

Engagement and response-time data per channel flow into the learning base: the orchestrator learns which tone, which formats and which times work per market – and plans the next content calendar from that instead of relying on fixed frequency plans.

Actions that need your approval

• ! Posts with prices or promises
• ! Replies to critical topics/feedback
• ! Influencer collaborations
• ! Reposting third-party content

Live metrics (checked hourly)

• Qualified leads per week (MQL→SQL)
• Lead score distribution & drift
• Workflow error rates
• Journeys (active/paused)
• Data field completeness
• Sales handoff response time

Automated A/B testing

A/B tests run automatically: subject lines, timing and content variants of journeys are tested in continuous operation – the scoring agent uses the winners for segment assignment.

Self-learning status

The scoring calibration runs continuously: from every won or lost deal, the orchestrator learns which behavioural signals predict real purchase intent – score weights, trigger timing and segment boundaries update themselves, tighter to the reality of your sales every week.

Actions that need your approval

• ! Journey changes affecting > 10% of contacts
• ! New segments with commercial promises
• ! Customer/CRM data into new workflows
• ! Changes to lead-scoring rules

Live metrics (checked hourly)

• Pipeline value per source
• Lead-to-deal rate
• Sync status per integration
• Field completeness (top 15 fields)
• Stalled deals (> 14 days)
• Handoff response time

Automated A/B testing

A/B tests run automatically: incentives, reply timing and handoff wording are variant-tested – response and deal data decide the default rule.

Self-learning status

Every pipeline movement – from lead to deal, from deal to revenue – trains the CRM model: the orchestrator recognises which sources, segments and fields predicted quality and recalibrates scoring and routing rules on its own instead of waiting for manual review cycles.

Actions that need your approval

• ! Routing rule changes sales ownership
• ! Scoring rule with big impact
• ! Bulk contact cleanup
• ! New integration connection

Live metrics (checked hourly)

• Inbox placement / domain reputation
• Bounce & spam complaint rate
• Send volume vs. limit
• Conversion per journey type
• Unsubscribe rate per send
• Revenue per email (attributed)

Automated A/B testing

A/B tests run automatically: subject lines, preheaders, send time and content blocks are tested continuously per journey – winners are adopted without anyone having to compare reports.

Self-learning status

Bounce, open and conversion data per journey and domain setup feed the learning base: the orchestrator recognises which segments, subject lines and send windows protect reputation and lift conversion – and plans the next send wave automatically from that instead of relying on static list logic.

Actions that need your approval

• ! Send to > 20% of the list
• ! Sending from a new domain/IP
• ! Win-back to inactive contacts
• ! Price changes in emails

Live metrics (checked hourly)

• Event validation (error rate per event)
• Consent consistency (CMP vs. server-side)
• Attribution drift per model
• Tag errors (PageSpeed/console)
• Estimated data loss
• Open validation fixes

Automated A/B testing

A/B tests run automatically: attribution models and event designs are compared in variants before being standardised – the report shows the difference, not the hope.

Self-learning status

Tag-level validation and consent deviations become the learning base: the orchestrator recognises where data is leaking and which event definitions are unstable – and automatically prioritises the fix queue by data-loss impact instead of waiting for manual audits.

Actions that need your approval

• ! Changes to event definitions (schema)
• ! Attribution model switch
• ! New domains/tags in live tracking
• ! Deletion of tracking data

Live metrics (checked hourly)

• Revenue vs. forecast (per channel)
• Pipeline coverage
• Attribution deviation per source
• Open data fixes
• Report freshness (hours)
• KPI changes (open)

Automated A/B testing

A/B tests run automatically: KPI definitions and dashboard logic are validated against history before rollout – only reproducible models make it into standard reporting.

Self-learning status

Deviations between forecast and reality per channel become training data: the orchestrator learns which channels are over- or underestimated and which attribution models track reality more closely – and adjusts KPI definitions and the forecast model itself before they land in the report.

Actions that need your approval

• ! Changing a KPI definition
• ! Report data to external recipients
• ! Swapping the forecast model
• ! Extending access to finance data