An auditable decision engine, step by step.

A walkthrough of the sources, contracts, confidence states, automation and safeguards that connect a fixture to a decision and a measurable campaign.

Open the live product

Traceability first. Automation second.

Back to commercial case

Design principles

Four rules before any model.

Canonical fixture
Calendar, signal, audience, campaign and outcome share one fixtureId.
Explicit states
Published, inferred, pending, demo and private data never collapse together.
Last valid observation
A source failure is logged without erasing the last correct observation.
Human in the loop
The engine proposes; authorised owners approve offer, assets and budget.

Logical architecture

Five layers joined by keys, not assumptions.

A future CRM connection enters at the source layer through a versioned contract; it does not require rebuilding the layers above.

  1. Decision & campaignToday · Calendar · CampaignPlan · scorecard
  2. Decision modelsTerritory · demand · audience · matchweek
  3. Contracts & statesfixtureId · campaignId · source · confidence
  4. Normalised datacalendar · attendance · signals · journeys
  5. Sourcesclub · WSL · weather · transport · events · calendar

Build phases

Each increment answers a new question.

  1. V1.0

    Territory opportunity

    940 LSOAs scored through family potential, girls' football networks, access and competition context.

    WHERE to acquire
  2. V1.1

    Live fixture context

    Canonical calendar, results, weather, travel friction and sourced material-change signals.

    WHEN to act
  3. V1.2

    Demand history

    An auditable 2025/26 attendance ledger, venue split and benchmark layer with source-state awareness.

    WHAT normal looks like
  4. V1.3

    Audience impact

    Owned-channel signals are connected while broadcast and search remain separate evidence states that may be pending or unavailable.

    WHY attention moved
  5. V1.3b

    CRM readiness

    Versioned ticket-level contract, synthetic Brighton rehearsal and time-bound post-match closure states.

    HOW conversion will close
  6. V1.4

    Signal-to-campaign

    CampaignPlan contract, reusable playbooks, activation briefs, UTM joins, approvals and guardrails.

    WHAT to execute
  7. V1.5

    Search demand observatory

    Matched GB/Spain comparison sets, an intent ladder, activation links and evidence gates for travel products.

    IF attention becomes demand
  8. V1.6

    Experience demand validation

    Three configurable concepts, a non-personal intent event and launch gates that keep registration and sale blocked until approved.

    WHICH product merits a pilot
  9. V1.7

    Mobility partnership layer

    Three fixture pilots, corridor scoring, aggregate privacy thresholds and a shuttle economics simulator that never presents assumptions as quotes.

    WHETHER mobility merits partner review
  10. V1.8

    Experiment measurement layer

    Allowlisted anonymous events, test/production separation, provider-aware delivery and aggregate fixture cohorts with a minimum evidence threshold.

    WHEN a scenario becomes evidence
  11. V1.9

    Partner commercial pack

    Four candidate-specific narratives join pilots, evidence states, value exchange, KPIs and approval gates without implying outreach or endorsement.

    HOW to frame an exploratory conversation
  12. V2.0

    Decision validation

    Time-stamped engine hypotheses can now be compared with later observable club action, dimension by dimension and without claiming causation.

    WHETHER hypotheses remain relevant in reality
  13. V2.1

    Decision-first product architecture

    Today, Calendar, Territories, Access, Experience, Measurement and Partnerships were reorganised around operator questions instead of feature inventory.

    WHAT deserves attention now
  14. V2.2

    Decision reliability

    Technical source health is mapped to the business decisions it supports, qualifies or blocks.

    WHICH decisions can be trusted today

Sources & cadence

Public sources already form an operating layer, with visible limits.

Automation checks and preserves comparable observations. When a source fails, it exposes the failure and retains the last valid value.

Daily

Fixtures & results

Official source, schedule changes and post-match closure.

Daily / windows

Weather & travel

Matchweek refresh without treating a forecast as certainty.

Weekly + T±

Events & attention

Competing events, broadcast picks and news. Social measurement stays with Blinkfire or the club’s social tool.

Post-match

Attendance & benchmark

Official observation, visible sample and comparable median.

Trust contract

The interface says what is, and is not, known.

Evidence states
StateMeaningExample
public-verifiedObserved in a cited public source.Fixture, published attendance
public-inferredA planning inference, never a known customer fact.Territory or player-led audience
pending-sourceExpected but not publicly available yet.TV audience, missing attendance
source-unavailableA capture failed without overwriting or inventing a value.Rate-limited Google Trends baseline
validation-conceptA proposition can be tested but is not on sale.Matchday, VIP and international concepts
modelled-scenarioA transparent planning assumption, not measured demand or a supplier quote.Corridor score and shuttle economics
provider-not-configuredEvents validate locally but are not stored or counted.Experience and mobility telemetry
prospecting-draftA non-binding narrative for review, not outreach or a relationship.Mobility and travel partner packs
demoSynthetic data that proves calculations only.Brighton ticketing rehearsal
requires-instrumentationMeasurable after UTMs or event tracking are deployed.Landing-page intent
requires-accessNeeds an authorised club system.Purchase, scan, no-show, repeat

Data contracts

The future integration already has a shape.

Even without real CRM access, the product defines exactly what it needs, how it validates and which calculations it unlocks.

fixture_id → campaign_id → ticket_id → scan → repeat_90d

CRM / ticketing

One ticket per row

fixture → campaign → product → price → purchase → scan → consent-safe geography → repeat

Open JSON schema ↗

Campaign plan

One governed campaign

fixture → signals → audiences → playbooks → activations → UTMs → approvals → measurement

Open JSON schema ↗

Search demand

One reproducible comparison

market → window → comparison set → term → dated index → activation → decision gate

Open JSON schema ↗

Experience demand

One non-personal intent signal

concept → fixture → origin → party → price → needs → gates → pilot decision

Open JSON schema ↗

Mobility partnership

One governed aggregate scenario

fixture → corridor → modelled score → aggregate threshold → quote → approval

Open JSON schema ↗

Experiment measurement

One allowlisted aggregate event

event → validate → deliver → deduplicate → cohort → threshold → review

Open JSON schema ↗

Partner commercial pack

One traceable non-binding pack

candidate → fixture → evidence → value exchange → KPIs → gates → review

Open JSON schema ↗

Pilot readiness

One provisional governed decision

pack → score → owners → blockers → timeline → hold / go

Open JSON schema ↗

Signal to campaign

Brighton exercises the full path without pretending demo data is real.

  1. SignalsOpener, kickoff, attention and weather.
  2. BriefAudience, proposition, offer, assets and channels.
  3. GovernanceOwner, four approvals and no automatic spend.
  4. AttributionUTMs and six campaign IDs exercised in demo.
  5. ClosureT+1, T+7, T+30, T+60 and T+90.

Validation & guardrails

The system blocks conclusions the data cannot support.

  • validate:campaigns
  • validate:crm-demo
  • validate:public-signals
  • validate:search-demand
  • validate:experience-demand
  • validate:mobility-partnership
  • validate:experiment-measurement
  • validate:partner-commercial-pack
  • validate:pilot-readiness
  • validate:source-health
  • typecheck
  • Vercel build
  • Fixture and campaign IDs must resolve to one canonical match.
  • Missing public observations remain pending; they never become zero.
  • A campaign cannot be ready while an approval gate is pending.
  • Repository data cannot mark a campaign live or authorise spend.
  • UTM campaign keys must equal the canonical fixture key.
  • Purchase, scan and repeat cannot be claimed without authorised data.
  • Hospitality gates cannot pass without measured downstream intent.
  • A validation concept cannot collect personal data, take deposits or claim inventory.
  • A mobility scenario cannot become a pilot without aggregated measured intent and an operator quote.
  • Addresses, postcodes and individual movements are prohibited from the partnership contract.
  • Test events cannot contribute to a production cohort or partner-facing threshold.
  • Unknown event properties are rejected before any provider delivery.
  • A partner pack must expose verified, modelled, unmeasured and partner-dependent evidence together.
  • Repository data cannot imply candidate contact, endorsement, inventory, quote or approval.
  • Observed alignment between an engine hypothesis and later club action never implies causation, access or influence.

Without CRM access

Separate available value from unlockable value.

Available now

  • Territory prioritisation
  • Fixture and matchweek context
  • Attendance history; current benchmark only when its source is available
  • Audience signals and campaigns; search only when a valid capture exists

Requires access

  • Purchase conversion
  • Scan and no-show
  • Real acquisition cost
  • Cohort repeat behaviour

Implementation

Simple to operate. Ready to grow.

  • Next.js
  • TypeScript
  • Versioned JSON
  • JSON Schema
  • GitHub Actions
  • Vercel
  • Public APIs
  • Responsive UI

AVELA · London City technical case studyIndependent demo · Data and limits visible by design.

Technical Case Study | AVELA