Agent Product Page · AI Revenue Intelligence Engine
CORTEX — AI Revenue Intelligence Engine
DESCRIBE. DIAGNOSE.
PREDICT. PRESCRIBE.
The intelligence layer a revenue leader would build if they had the time: one engine watching every revenue system at once, telling you what happened, why, what comes next, and exactly what to do about it — with the action already prepared for your approval.
CORTEX value proposition
CORTEX watches every revenue system you run and answers four questions on every material movement — what happened, why, what is likely next, and what to do — with evidence, a confidence score, expected impact and risk attached, and the action prepared for your approval.
PROBLEMS IT SOLVES
Dashboards that show the number but never the reason behind it
Three systems, three different versions of the same number
Findings with no owner, no deadline and no next action
No sense of what a problem costs if it is left for another month
Deals going quiet and nobody spotting it until the pipeline review
Insight arriving as a report to read rather than a decision to make
FOUR LAYERS OF INTELLIGENCE
A dashboard stops at the first layer. CORTEX runs all four continuously across every revenue system you connect — describing what happened, diagnosing why, predicting what comes next, and prescribing the action, with the action itself prepared and waiting on your decision.
Analysed continuously
Revenue & pipeline
- Revenue and closed-won
- Pipeline value and coverage
- Conversion rates by stage
- Churn and retention
- Customer acquisition cost
Sales & CRM
- CRM record and stage changes
- Sales activity and response times
- Lead quality and scoring
- Deal ageing and stall risk
- Owner and team performance
Demand & marketing
- Advertising performance and ROAS
- Campaign creative and audience fatigue
- Email engagement and deliverability
- Website and GA4 analytics
- Customer behaviour and journey signals
- Reconciles CRM, ad, email, analytics and billing data into one version of each number
- Separates material movement from normal week-to-week variance
- Ranks findings by revenue materiality, so the briefing stays short
- Isolates the change to a channel, campaign, segment, stage, source or owner
- Lines it up against what changed in the system — budgets, creative, audiences, routing, sequences, team
- Names a most probable cause with a confidence score, lists the evidence, and says so when the cause is unclear
- Projects the current trend forward across 14, 30, 60 and 90 days
- Expresses the outcome in pipeline, revenue, CAC and ROAS terms — not percentages alone
- Publishes the assumptions, the range and the confidence, and labels every figure as a projection
- Turns each finding into one specific action rather than a menu of options
- States expected impact, risk and how the result will be measured
- Prepares the executable action, assigns an owner and a deadline, and holds it at the approval gate
01 · Describe
What happened?
You get: A short, ranked list of what actually changed — the metric, the size of the move, the window, and the records behind it.
02 · Diagnose
Why did it happen?
You get: A named probable cause, a confidence score, the supporting evidence, and the alternative explanations it ruled out.
03 · Predict
What is likely to happen next?
You get: A labelled forecast range for the do-nothing path, with assumptions and confidence stated on the face of it.
04 · Prescribe
What action should the business take?
You get: A ranked recommendation carrying evidence, confidence, expected impact, risk and a prepared action — approve, reject, override or escalate.
ONE FINDING, END TO END
Illustrative example using representative figures — not a client result. On your account every figure would come from your own connected systems.
- Describe
- Qualified leads decreased 17% this week.
- Cause
- LinkedIn campaign audience fatigue.
- Evidence
- CTR down 23% week on week on the same creative set
- Frequency up 31% against the same audience
- Cost per lead up 18%
- Every other source flat over the same window
- Prediction
- If unchanged, qualified pipeline is expected to decline a further 9–14% over the next 14 days (modelled projection).
- Recommendation
- Refresh creative set B and move 15% of budget to audience segment C.
- Confidence
- High — 0.86. Four independent signals agree and no competing cause is supported.
- Expected impact
- Recovery of most of the lost qualified-lead volume within 10–14 days, with CPL returning towards its prior range.
- Risk
- Low. The budget shift is reversible within one cycle, segment C has existing performance history, and the change is capped at 15% of campaign spend.
- Action
- SENTINEL has the creative rotation and budget shift prepared. Request approval — nothing moves until a human approves it.
EXECUTIVE DAILY BRIEFING
Every morning, by email or Slack · Founder, revenue lead, sales and marketing owners
- What moved yesterday, ranked by revenue materiality
- Named cause and confidence for each material movement
- 14-day projection for anything left unaddressed, labelled as a projection
- Today's recommended actions with owner, expected impact and risk
- Actions prepared and waiting on your approval
- Data-quality or reconciliation issues, flagged rather than smoothed over
WEEKLY STRATEGIC BRIEFING
Weekly, with a monthly roll-up · Leadership and board reporting
- Revenue, pipeline, CAC, ROAS and churn against target and prior period
- Which channels and segments actually produced revenue, deduplicated
- Lead-quality and conversion trends by source and stage
- What last week's approved actions actually delivered against their expected impact
- Forecast for the next 30, 60 and 90 days with assumptions and confidence stated
- The two or three decisions that matter most this week, with the trade-offs
DESCRIBE and DIAGNOSE come from your own connected systems and cite the source records behind every figure. PREDICT is an explicitly labelled model output with a confidence score and stated assumptions — never presented as an actual or a guarantee. PRESCRIBE always states expected impact and risk, and CORTEX itself is read-only: an approved action is executed by NEXUS, SENTINEL, HERALD or NOVA under your approval policy, and executes automatically only for action types you have explicitly pre-authorised.
INPUTS, LOGIC, OUTPUTS
Inputs
- CRM deals, stages and activity history
- Ad platform spend and performance
- GA4 and Search Console data
- Booking events and closed revenue from billing
- Change history — routing rules, budgets, sequences, creative and team changes
- Agreed scoring rules, thresholds and attribution model
Processing Logic
- Pull and reconcile all sources so one booking is counted once
- Separate material movement from normal variance
- Trace each material movement to its most probable cause and state the confidence
- Score leads, deals and campaigns, and attribute revenue by channel and agent
- Model the 30/60/90-day effect of leaving each finding alone, labelled as a projection
- Convert each finding into one recommended action with an owner, effect and deadline
- Prepare the executable action and hold it at the human approval gate
- Measure what happened after each approved action and feed the result back
Outputs
- A daily intelligence briefing answering all five questions
- A named probable cause per finding, with confidence and cited evidence
- Labelled 30/60/90-day projections of the cost of inaction
- A ranked action list with owner, expected effect and deadline
- Prepared agent actions held at the approval gate
- Lead, deal and campaign scores, and spend-to-revenue attribution
CORTEX specification
Connects To
- HubSpot, Salesforce or Pipedrive
- GA4 and Google Search Console
- Meta, Google and LinkedIn ad accounts
- Stripe / invoicing for closed revenue
- Slack and email for briefings
- Every other Xiilio agent as a data source
What Triggers It
- Scheduled daily and weekly reporting runs
- A deal changing stage, or going quiet past a threshold
- A campaign breaching a performance threshold
- A closed-won or closed-lost outcome
- A report requested on demand
Decisions It Makes
- What score each lead, deal and campaign should carry, and why
- Which movement is a real signal and which is normal variance
- What the most probable cause of a change is, and how confident it is
- Which channel gets credit for a booked call
- Which recommended action has the highest expected value
- What the cost of doing nothing looks like over the next 30, 60 and 90 days
- What to escalate as urgent today rather than report weekly
Acts Autonomously On
- Reconcile, score and attribute data
- Diagnose causes and generate projections, labelled as projections
- Generate and send the daily briefing
- Raise risk alerts to owners
- Prepare recommended actions and hold them at the approval gate
- Update ICP weighting signals for NOVA
GUARDRAILS, IN WRITING
An autonomous agent is only useful if its limits are explicit. Here is what CORTEX will never do, and where a human always signs off.
What It Cannot Do
- Change deal values, stages or outcomes in the CRM
- Present modelled estimates as verified results
- Report on data sources you have not connected
- 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
- Scoring rules and thresholds are agreed with you before going live
- Attribution model and reporting definitions are signed off
- CORTEX reports and recommends — humans decide and act
- Changes to scoring logic are versioned and approved
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 source fails to reconcile, the briefing ships with that section marked incomplete rather than estimating the gap
Escalation To Humans
- CORTEX reports and recommends — humans decide and act
- Material anomalies are escalated the same day, not held for the weekly view
- Data-quality problems are escalated rather than smoothed over
- Scoring-logic changes are versioned and require approval
EXAMPLE OUTPUT
Daily intelligence briefing — a representative example of what CORTEX writes back into your systems.
period : yesterday · all sources reconciled
what : qualified-lead→meeting conversion down 14% week on week
where : isolated to inbound web form leads, all other sources flat
why : median response time 6 min → 4 hr since the 8 Sep routing change (confidence: high)
evidence : 212 CRM records, booking events, NEXUS run history — all cited
do nothing : projected −£18k pipeline over 30 days (modelled, not an actual)
recommendation : revert the routing rule, reinstate the 10-minute response SLA
xiilio action : NEXUS revert prepared and held — awaiting your approval
owner / by when : operations lead · today
note : actuals come from your systems; projections are labelled as projectionsEXAMPLE DASHBOARD
CORTEX — intelligence dashboard. Every figure below is a simulated demo value, not a reported client result.
Findings today
5
material only · illustrative
Causes identified
4 / 5
one marked cause unclear
Actions recommended
5
owner and deadline attached
Awaiting your approval
2
prepared, not executed
Projected cost of inaction
£18,400
30 days · modelled projection
Attribution coverage
94%
booked calls matched to source
KPIs CORTEX IS HELD TO
Diagnosis rate
Findings with a named probable cause, not just a number
Time to cause
Hours between a movement starting and its cause being identified
Recommendation adoption
Share of recommended actions your team approves
Post-action effect
Measured result of approved actions against the expected effect
Projection accuracy
How close the do-nothing projection was to what actually happened
Attribution coverage
Share of booked calls matched to a source
Risk detection lead time
Days between a deal stalling and the alert
Technical Architecture
HOW CORTEX CONNECTS
CORTEX reads from your existing systems, applies its own governed processing, and writes results back. Nothing runs outside these lanes.
Upstream · sources
- CRM deals and activity
- Ad platform spend
- GA4 / Search Console
- Billing and invoicing
- System change history (rules, budgets, sequences)
- Every other Xiilio agent as a data source
CORTEX · processing
- Ingestion + reconciliation layer
- Deduplication and identity matching
- Anomaly detection vs normal variance
- Causal diagnosis engine with confidence scoring
- Projection model for the cost of inaction
- Recommendation ranker and action composer
Downstream · destinations
- Daily briefing (email / Slack)
- Risk alerts to record owners
- Prepared actions → approval gate
- Approved actions → NEXUS, SENTINEL, HERALD, NOVA
- Attribution reporting in the Command Centre
- Read-only against your CRM — CORTEX cannot change deal values, stages or outcomes
- CORTEX never executes its own recommendations; an approved action is carried out by another agent
- Every reported figure cites the source records behind it, and projections are labelled as projections
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
- Scoring rule version and the source records behind every figure reported
Full detail on encryption, data handling and your GDPR rights is on the Trust & Security page.
How a professional services firm would use CORTEX
Meetings booked drop and nobody can agree why. Overnight, CORTEX reconciles the data, isolates the drop to one lead source, traces it to a routing change made nine days earlier, projects what the drop costs over the next 30 days if nothing changes, recommends the exact revert, and holds that revert ready for a partner to approve. One decision replaces a week of arguing about whose spreadsheet is right.
This scenario is written to show how CORTEX is deployed. It is not a named client, a delivered engagement or a reported result.
Live Demo
SEE CORTEX RUNNING
CORTEX's intelligence stage appears in the homepage system builder, and its findings appear in the Command Centre operations feed. Metrics shown there are simulated demo values.
See the intelligence layerCORTEX FAQs
CORTEX — QUESTIONS WE GET ASKED
- How is this different from a dashboard?
- A dashboard answers one question: what happened. CORTEX runs four layers — describe, diagnose, predict, prescribe — across CRM, pipeline, sales activity, ads, email, analytics, CAC, ROAS, churn and revenue at once. The fourth layer arrives as a prepared action carrying evidence, confidence, expected impact and risk, waiting on your decision — not a chart.
- Does CORTEX change anything in our CRM?
- No. It reads, diagnoses, scores, attributes and recommends. It cannot change deal values, stages or outcomes, and it never executes its own recommendations — an approved action is carried out by another agent under your approval policy.
- How reliable is the “why”?
- Every diagnosis carries a confidence level and the evidence behind it, including the records and the system changes it lined up against. Where the cause is genuinely unclear, CORTEX says so rather than inventing one.
- Are the predictions real numbers?
- No — forecasts are modelled projections, always labelled as such, published as a range with their assumptions and confidence shown. Reported actuals come from your own connected systems and cite the source records. Nothing modelled is ever presented as a result.
- How is attribution decided?
- Against an attribution model and reporting definitions agreed with you before go-live. Records are reconciled first so one booked call isn't counted several times across channels.