Baltic CVM Lab Competitive pricing and retention decision intelligence · Latvia

Baltic CVM Lab

Competitive pricing index and retention decision layer · Latvia · public price lists and a synthetic base · no operator data

30
Offers tracked
3 operators, published prices
16
Personas
priceable baskets
6
Ahead
of 6 priced
9
Unserved
no qualifying offer
35
Tests green
one per shipped defect
1.2%
Base churn
monthly, in published band

Build state against the concept

The concept specifies nine views. 1.5 of 9 exist in some form. This is shown rather than hidden — a menu that promises screens which do not exist is worse than a short menu.

§ViewNote Status
12.1Market PressureMatrix and index live. No gap heatmap, change feed, competitor timeline or promo calendar.partial
12.2Portfolio & ExposureBase composition, contract-expiry exposure curve, margin at risk by segment.not built
12.3Decision CockpitReplaced by the break-even surface. No ranking toggle or live sliders.partial
12.4Customer 360 / Agent panelOne subscriber: risk, uplift, reason codes, guardrails, override capture.not built
12.5Experiments & KnowledgeHoldout status, findings ledger with decay, hypothesis queue.not built
12.6Semantic Layer & LineageFeature catalogue with owner, lawful basis, quality SLO, git commit. The concept calls this the most strategically important screen.not built
12.7Vendor PlaneTwo adapters compared under one policy. No live swap control.partial
12.8GovernanceAI Act evidence pack, searchable decision log, fairness monitors.not built
12.9MCP surfaceNatural-language querying of the semantic layer.not built

Proof-of-concept modules

ModuleNoteStatus
A · External signal3 operators in 1 market. Concept scope is 9 operators across 3 markets, with movement detection.partial
B · Simulated estateCalibrated and CI-gated — but emitted as ONE CLEAN TABLE. The concept explicitly warns that this assumes away the real problem: the estate should arrive as separate systems with conflicting definitions.partial
C · Vendor adapterTwo interchangeable adapters, one deliberately worse.built
D · Sovereign corePolicy engine, score interception and uplift built. Semantic layer, evidence store and knowledge base are not.partial
E · Surfaces3 of 9 concept views, plus overview and method pages.partial

Current position — Latvia

PersonaReference operator Best competitorPosition
Value Seeker€16.99lmt €30.98ahead €13.99
Digital Native€30.89lmt €30.98ahead €0.09
Cross-border Commuter€30.89lmt €30.98ahead €0.09
Voice-first Senior€16.99lmt €30.98ahead €13.99
Student / Youth€30.89lmt €30.98ahead €0.09
Prepaid Occasional€16.99lmt €30.98ahead €13.99

Compared on standard rates, not promotional headlines. 9 personas are unserved by every operator — multi-line family baskets, which no extractor captures yet. Reported as a finding rather than filled in.

Three findings worth the click

Advertised prices rank differently from standard rates. Tele2's unlimited plan promotes below its own 25 GB plan, so an index built on headlines reports a position that reverses when the promotion ends. Market pressure
The conventional retention offer cannot break even at any targeting quality. 18% × 6 months needs ~17pp of uplift; at 1.2% monthly churn the ceiling is ~5pp. An earlier claim of a €44k gain from better targeting was withdrawn after calibration. Retention economics
Let the platform rank and it contacts people your own rules forbid. 61–79% of each vendor's top list breaches an operator suppression rule. Vendor plane

Largest known gaps

Semantic layer and lineage — the concept's §12.6, which it calls the most strategically important screen, does not exist. Feature definitions live in code and YAML but are not surfaced with owner, lawful basis or quality SLO.
The estate is one clean table. The concept explicitly warns against this: a proof of concept with a single tidy table assumes away the reconciliation problem that the semantic plane exists to solve.
No evidence store. Decisions are produced but not persisted append-only, so the decision log — the asset the concept says cannot be regenerated once lost — does not yet exist.
Persona weights are uncalibrated. Informed estimates, not measured against SPRK, RRT or TTJA statistics.