Agent Product Page · AI Autonomous Media Optimisation Engine
SENTINEL — AI Autonomous Media Optimisation Engine
EVERY POUND.
EARNING ITS PLACE.
SENTINEL watches your paid spend the way a good media buyer would — hourly, across platforms — and ties every pound to a booked call rather than a click. Its budget engine answers one question continuously: where should the next £1,000 go?
SENTINEL value proposition
SENTINEL watches paid spend hourly across platforms, rotates approved creative when fatigue appears, and reports cost per booked call rather than cost per click.
PROBLEMS IT SOLVES
Spend drifting for days before anyone notices
Reporting on clicks that never became conversations
Creative fatigue spotted only in the monthly review
Audiences built on guesswork instead of closed-won data
No line between ad spend and booked revenue
AUTONOMOUS BUDGET OPTIMISATION
Paid media does not fail in monthly reports — it fails hourly, in small amounts, while nobody is watching. SENTINEL's budget engine reallocates spend continuously, at the pace platforms actually move, inside bounds you set and own.
“Where should the next £1,000 go?”
Media platforms
- Google Ads
- Meta Ads
- LinkedIn Ads
- TikTok Ads
Measurement
- GA4 and Google Search Console
- Server-side conversion tracking
- Multi-touch and platform-reported attribution, reconciled
- Holdout and geo tests for incrementality
Business systems
- HubSpot, Salesforce or Pipedrive
- Closed-won revenue and margin data
- Booking and qualification events from ARIA
- CORTEX forecasts and briefings
Optimised for
- Qualified pipeline generated per pound
- Customer acquisition cost
- Revenue attributed to spend
- Contribution profit after media cost
- ROAS on closed revenue, not platform-reported conversions
Not optimised for
- Impressions
- Clicks and CTR as an end in themselves
- Platform-reported conversions taken at face value
- Lead volume without qualification
Platform-reported numbers are used as signals, never as the scoreboard. The scoreboard is CRM revenue.
BUDGET ALLOCATION
Continuous reallocation across platforms, campaigns and ad sets, scored on marginal qualified pipeline per pound and bounded by your caps.
AUDIENCE DISCOVERY
Builds and tests audiences from closed-won profiles and first-party CRM segments, retiring those that stop producing qualified pipeline.
CREATIVE TESTING
Structured tests across approved variants, with sample-size and duration rules so a winner is called on evidence rather than an early lead.
CREATIVE FATIGUE DETECTION
Frequency, CTR decay, CPM drift and reach saturation tracked per variant, triggering rotation before performance collapses.
CAMPAIGN ANOMALY DETECTION
Hourly checks for spend spikes, tracking breaks, sudden CPA drift, disapprovals and delivery collapse, with alerts carrying the evidence.
LANDING-PAGE TESTING
Proposes and measures page variants against booked calls and qualified pipeline, not bounce rate. Publishing a variant requires approval.
CONVERSION TRACKING
Server-side event tracking with consent respected, so bookings and revenue are matched to spend rather than inferred from the browser.
ATTRIBUTION
Reconciles platform-reported, GA4 and CRM-sourced attribution into one number per campaign, with the disagreement between sources shown rather than hidden.
INCREMENTALITY ANALYSIS
Holdout and geo tests estimate what spend actually caused, separating incremental demand from demand you would have captured anyway.
Performance
- Cost per booked call per campaign, ad set and creative
- Marginal return — what the last £100 in each campaign actually produced
- Conversion rate and lead quality by audience segment
- Pacing vs plan, per platform
Fatigue & saturation
- Frequency and CTR decay per creative variant
- Audience saturation — shrinking reach against rising CPM
- Days since creative rotation per ad set
Business context
- Pipeline capacity — no point buying leads a full sales calendar cannot take
- Closed-won value by source, from CRM attribution
- Seasonal and daypart performance patterns
- CORTEX forecasts for the campaigns affected
01
SCORE
Every campaign is scored on marginal return per pound — not historical averages, but what the most recent spend actually produced.
02
COMPARE
Candidate moves are ranked: scale the winner, hold, rotate creative, or pull spend from a fatigued ad set. Each carries expected impact and risk.
03
BOUND
The move is checked against your caps, CPA thresholds and pre-authorised action types. Inside the bounds it can execute; outside them it cannot.
04
ACT OR PROPOSE
In-bounds moves execute and are logged with before/after values. Out-of-bounds moves arrive as a proposal with the full evidence — nothing moves until you approve.
05
VERIFY
The next cycle measures whether the move did what it predicted. Moves that miss are reversed or flagged, and the allocator's own accuracy is tracked.
A WORKED ALLOCATION DECISION
Illustrative example with simulated figures. Not a claimed result.
“Move £2,000 from Campaign A to Campaign C.” Campaign A is spending £6,000 a month at £610 per qualified opportunity; Campaign C is spending £2,500 at £226 — 2.7× more qualified pipeline per £1 over the last 60 days of CRM-attributed data.
MOVE £2,000 A → C
Recommended
Campaign C produces 2.7× higher qualified pipeline per £1, its frequency is 1.8 with reach still expanding, and a geo holdout shows the majority of its conversions are incremental.
MOVE THE FULL £6,000 A → C
Rejected
Beyond roughly £4,500 a month, Campaign C's audience saturates and modelled marginal cost per opportunity rises sharply. Scaling past that point buys volume, not pipeline.
HOLD AND ROTATE CAMPAIGN A CREATIVE
Held in reserve
Campaign A's decline tracks creative fatigue rather than audience quality, so a refreshed variant is queued and measured before deciding whether the remaining £4,000 stays.
- Decision
- £2,000 per month moves from Campaign A to Campaign C. Campaign A retains £4,000 pending a creative refresh test.
- Confidence
- High — 60 days of CRM-attributed opportunities across both campaigns, supported by a geo holdout
- Expected impact
- Modelled 12–16 additional qualified opportunities per month and a 9–14% reduction in blended CAC. Modelled, not guaranteed.
- Risk
- Low to moderate — reversible within one cycle; the main risk is Campaign C saturating faster than modelled, which the next cycle would detect.
- Action
- Under Recommend, it waits for you. Under Approve, it is queued for one click. Under Auto-execute, it runs inside your caps and is logged with before/after values. The rejected options and their reasoning are logged either way.
WHAT EVERY RECOMMENDATION CARRIES
- Current allocation
- What each affected campaign is spending now, with its current cost per qualified opportunity.
- Recommended allocation
- The exact proposed split, in currency, with the source and destination named.
- Expected outcome
- Modelled qualified pipeline, CAC and revenue effect over a stated window.
- Confidence
- How strong the underlying evidence is: volume of attributed conversions, test duration and how far the signal sits from noise.
- Evidence
- The specific metrics and records behind the call, linked back to the source system.
- Risk
- What could go wrong, how reversible the move is, and the worst realistic outcome.
RECOMMEND
SENTINEL proposes the move with full evidence and takes no action until someone acts on it.
Human, every time
APPROVE
The move is prepared and queued; one click executes it, and it expires if left unactioned.
Human approves, SENTINEL executes
AUTO-EXECUTE
Pre-authorised action types inside your caps run automatically and are logged with before/after values.
You define the action types and bounds up front
THE MEDIA COMMAND CENTRE
One view across every connected platform, where each major budget decision is shown with the reasoning that produced it — including the options SENTINEL rejected.
Allocation map
Live spend by platform, campaign and audience, against qualified pipeline produced.
Decision log
Every allocation change with its evidence, confidence, expected impact, risk and who or what approved it.
Rejected options
The moves SENTINEL considered and declined, and why — so the reasoning is auditable, not just the outcome.
Creative board
Variant performance, fatigue state and days since rotation, per ad set.
Anomaly feed
Spend, tracking and delivery anomalies with severity and the action taken or proposed.
Incrementality
Running holdout and geo tests, with measured lift and the confidence interval around it.
Prediction accuracy
Whether previous recommendations delivered what they forecast — the engine graded against its own predictions.
Approvals queue
Proposals awaiting a human, each with the full decision record attached.
Figures shown in the interface reflect your own connected accounts. Examples on this page are simulated.
What the engine can never do
- Total budget caps are hard bounds in code — the allocator can never exceed them
- CPA thresholds pause spend automatically; restarting needs a human
- Only moves you have pre-authorised execute without approval; everything else is a proposal
- New campaigns, channels or creative always require sign-off
- Every allocation is logged with the signals, scores, options considered and the decision taken
The engine optimises for expected revenue per pound, not spend volume, impressions or clicks. Its allocation accuracy is itself a tracked KPI — you can audit whether its predictions held.
INPUTS, LOGIC, OUTPUTS
Inputs
- Google, Meta, LinkedIn and TikTok ad account data
- First-party CRM and closed-won profiles
- Your approved creative library
- Budget caps, CPA thresholds and pacing plans
- Booking and revenue events from the CRM
Processing Logic
- Build and refresh audiences from first-party data
- Pull spend, pacing, CAC and frequency hourly across platforms
- Detect fatigue and CPA drift with the supporting evidence
- Reallocate budget within your approved caps
- Match booked calls back to campaigns for true CPA
- Propose anything outside caps for human approval
Outputs
- Refreshed audience segments pushed to each platform
- Creative rotations from the approved library
- Budget reallocations inside your caps, with before/after values
- Cost per booked call by campaign and channel
- Approval proposals for out-of-cap changes
SENTINEL specification
Connects To
- Google Ads, Meta Ads, LinkedIn Ads, TikTok Ads
- GA4 and Google Search Console
- HubSpot, Salesforce or Pipedrive (for closed-loop revenue)
- Your creative asset library
- Slack or email for spend alerts
- CORTEX (reporting) and NEXUS (workflow)
What Triggers It
- Hourly and daily performance checks
- Spend pacing above or below plan
- Cost per acquisition breaching a threshold you set
- Creative fatigue signals (frequency, CTR decay)
- A new audience segment produced by NOVA or the CRM
Decisions It Makes
- Which audiences to build, refresh or retire
- Which approved creative to rotate in when fatigue appears
- How to reallocate budget within your approved caps
- Which campaigns to pause when CPA breaches thresholds
- What to escalate to a human before acting
Acts Autonomously On
- Refresh and retire audiences
- Rotate approved creative on fatigue
- Shift budget between campaigns within caps
- Pause campaigns breaching CPA thresholds
- Raise spend alerts
GUARDRAILS, IN WRITING
An autonomous agent is only useful if its limits are explicit. Here is what SENTINEL will never do, and where a human always signs off.
What It Cannot Do
- Raise total budget beyond the cap you set
- Publish creative or claims that have not been approved
- Launch a new campaign or channel without human sign-off
- Target protected categories or breach platform ad policies
- Sign contracts, agree discounts or commit your business to anything
- Invent facts, case outcomes, guarantees or performance claims
- Give legal, medical or regulated financial advice
- Act outside the systems and data scopes you explicitly grant it
Human Approval Points
- Total budget caps and CPA thresholds are set and owned by you
- All creative and ad copy is approved before it can run
- New campaigns, channels and audience types require sign-off
- Budget shifts beyond the approved range are proposed, not executed
WHEN THINGS GO WRONG
Failure Handling
- Every run is idempotent — a retry can never duplicate a record or a send
- Transient failures retry with backoff; persistent failures stop the run rather than guess
- A failed step never silently continues — the workflow halts in a known state
- Failures raise an alert to the owning human with the run ID and the exact failing step
- Errors, retries and halts are written to the same audit trail as successful actions
- If a platform API fails mid-change, SENTINEL reconciles the actual platform state before acting again — never re-applying a change blind
Escalation To Humans
- Any increase to total budget is a proposal, never an action
- New campaigns, channels or audience types need sign-off
- Unapproved creative can never be published
- Sustained CPA breaches escalate to the owning human with the evidence
EXAMPLE OUTPUT
Hourly spend check — a representative example of what SENTINEL writes back into your systems.
window : 13:00–14:00 UTC · 3 platforms
pacing : 96% of daily plan — within tolerance
cpa : £64 vs £75 threshold ✓
fatigue : ad set 'Litigation-A' frequency 4.2 → rotate
action : approved creative v3 swapped in ✓
budget : £120 shifted from 'Retarget-B' (within £250 cap) ✓
escalation : proposal to raise daily cap → awaiting human approval
attribution : 6 booked calls matched to spend this weekEXAMPLE DASHBOARD
SENTINEL — acquisition dashboard. Every figure below is a simulated demo value, not a reported client result.
Spend managed
£8,240
last 30 days · illustrative
Cost per booked call
£64
vs £75 threshold
Booked calls attributed
129
matched to CRM
Creative rotations
14
fatigue-triggered
Budget shifts
9
all within approved caps
Awaiting approval
2
out-of-cap proposals
KPIs SENTINEL IS HELD TO
Cost per booked call
Spend ÷ appointments actually booked
Pacing accuracy
Actual spend vs planned spend
Creative fatigue response time
Hours from fatigue signal to rotation
Spend-to-revenue mapping
Share of spend matched to CRM outcomes
Wasted spend prevented
Spend paused on campaigns breaching thresholds
Technical Architecture
HOW SENTINEL CONNECTS
SENTINEL reads from your existing systems, applies its own governed processing, and writes results back. Nothing runs outside these lanes.
Upstream · sources
- Meta / Google / LinkedIn Ads APIs
- GA4 and Search Console
- CRM first-party and closed-won data
- Approved creative library
SENTINEL · processing
- Audience builder
- Hourly pacing and CPA monitor
- Fatigue detector
- Budget allocator (cap-bounded)
- Closed-loop attribution matcher
Downstream · destinations
- Ad platform campaign changes
- CRM revenue attribution
- CORTEX reporting
- Slack spend alerts and approval proposals
- The budget allocator is hard-bounded by your caps in code, not by prompt instruction
- Only assets present in the approved creative library can be published
SECURITY & AUDIT LOGS
Security
- Data encrypted in transit (TLS 1.2+) and at rest
- Least-privilege access — the agent only sees the records its job needs
- Row-level security on every customer record in the Xiilio platform
- UK/EU data handling with GDPR rights (access, export, deletion) honoured
- No customer data used to train third-party foundation models
Logged On Every Action
- Timestamp (UTC) and agent identity
- Trigger that started the run
- Every system read and write, with the record ID touched
- The decision taken and the rule or score behind it
- Human approvals, overrides and rejections
- Every budget change, creative rotation and campaign pause, with before/after values
Full detail on encryption, data handling and your GDPR rights is on the Trust & Security page.
How a broker would use SENTINEL
Spend is fine on Monday and quietly wrong by Thursday, and nobody notices until the monthly report. SENTINEL checks hourly, rotates approved creative when frequency climbs, and moves budget within a cap the broker set themselves. Anything bigger than that cap is a proposal sitting in Slack, not a change already made.
This scenario is written to show how SENTINEL is deployed. It is not a named client, a delivered engagement or a reported result.
Live Demo
SEE SENTINEL RUNNING
Choose 'Improve paid advertising' in the homepage builder to see SENTINEL's stage. Values shown are simulated.
See paid acquisition in the system builderSENTINEL FAQs
SENTINEL — QUESTIONS WE GET ASKED
- Who writes the ads?
- All creative and ad copy is approved by you before it can run. SENTINEL decides which approved variant to rotate in, based on fatigue and performance signals — it does not publish unapproved claims.
- How is performance measured?
- Cost per booked call, tracked end to end by matching bookings back to spend, rather than cost per click. Results are passed to CORTEX for reporting.
- Which platforms are supported?
- Google Ads, Meta, LinkedIn and TikTok, alongside GA4 and Search Console for measurement, with CRM and revenue data closing the loop.
- What does SENTINEL optimise for?
- Qualified pipeline, customer acquisition cost, revenue, profit after media cost and ROAS on closed revenue. Impressions, clicks and platform-reported conversions are treated as signals, not as the scoreboard.
- Can it move budget on its own?
- Only if you set that action type to auto-execute, and only inside your caps. The default is Recommend, where nothing moves until a human acts. Approve sits in between: the move is prepared and queued for one click.