Skip to content

Analysis engine

Subscription detection

Transform raw statements into subscription intelligence: recurrence, amounts, deadlines and alerts — with traceability, user control and honesty about the limits of the source data.

Detection is the basis of tocancel: without an initial reliable mapping, any termination or negotiation is based on partial perceptions. We designed the engine as an enrichment chain rather than a black box: each subscription offered can be explained (rule, counterpart, reason for recurrence) and adjusted by you if the reality on the ground differs.

theoretical sources
banks, neo-banks, manual exports
4
large cross signals
recurrence, amount, consideration, context
100%
human control possible
merger, split, exclusion by position
€0
bankauto payment
we are not your bank

Processing pipeline

Simplified diagram: in production, queues, validations and guardrails complete these steps to absorb import peaks.

Subscription detection pipeline diagramSourcesStandardizationDetectionAlertsActions

What we mean by “subscription detected”

This is not a universal legal definition: it is a position classified as a recurring commitment based on available signals — always revisable.

The same direct debit may be a subscription, an association membership, a split payment or a wording error. Our rules seek to isolate what behaves like a commitment that is renewed (same amount at stable intervals, known counterpart, wording consistent with a digital service or telecommunications). When ambiguity remains strong, we favor the status “to be validated” rather than a categorical affirmation – precision is better than the illusion of completeness.

  • Stable monthly direct debits from an identifiable provider (streaming, cloud, online press).
  • Trials or first months at a reduced rate followed by a move to full price when the reason appears on the wording or history.
  • Annual or quarterly formulas with predictable expiry date.
  • Recurring “add-on” options attached to a main service (additional storage, additional seats).
  • Telecom subscriptions or boxes whose wording sometimes aggregates several components — detected as a grouped line with context note.
  • “Onboard” insurance or tacitly renewed contracts when renewal signals are visible in the flow.
  • Classic B2C services hidden behind a generic title (“INTERNET CB PAYMENT”) with few clues — placed in “low trust”.
  • Borderline cases: recurring donations, internal amortization, rentals — reported rather than imposed as leisure subscriptions.

Six levers produced to secure the quality of cartography

Coverage & standardization

Gathering by counterparty, hidden IBAN or aggregation, harmonization of noisy labels. We reduce “Netflix” vs “NFLX” duplicates when the signal is sufficient.

Immediate budget reading

Amount per period, monthly equivalent, trend over 3 to 12 months, contribution to total household subscriptions.

Contextualized alerts

End of trial, double withdrawal, increase at renewal, imminent annual change – each alert cites the data that triggers it.

Preservation of evidence

Useful historical snapshot for a complaint or termination: what was seen, when, with what level of confidence.

Multi-account scenarios

Personal + joint + professional account: filter by source without mixing legal frameworks or internal expense reports.

Resilience to partial imports

Error recovery, incremental import, reminder if a date range is missing — avoid temporal blind spots.

« A good detection is not the one that fills the table the fastest: it is the one that warns you when it is in doubt. »

Product principles tocancelInternal documentation

False positives, false negatives: how we deal with them

Transparency on what algorithms cannot guarantee.

A false positive makes you believe that a service is still active when it is no longer; a false negative leaves an actual subscription off the radar. Both cost time or money. We document trust reasons (high, medium, low) and provide suggested groupings that you can reject with one click. On merchants who invoice under various legal entities, detection can fragment the same offer: your manual merger teaches the engine for the coming months.

  1. Ingest

    Secure connection or file: we log the date range covered and the parser version used.

  2. Classify

    Application of business rules + heuristic scores + correspondence with a merchant repository when it exists.

  3. Consolidate

    Careful automatic merge; otherwise suggestion with summary “why we thought it was the same thing”.

  4. To present

    List view, due date calendar view, household view — same data, three reading angles.

  5. Adjust

    Your corrections feed into the local trust profile (not shared with other users without legal basis).

  6. Reassess

    With each new import, recalculation of alerts and projections without losing the history of previous versions.

Typical sources and compromises (indicative)

SourceBenefitsUsual limits
PSD2 banking aggregationLess manual entry, fresher feedsDepends on connector and bank; some merchants absent
CSV/PDF exportFull user control, excluding API dependencyVariable quality; OCR sometimes necessary; export delays
Multi-accountsConsolidated view of the homeRisk of double counting without merger discipline
Short historyRapid detection of the existingLess visible annual or seasonal patterns
Screen and cybersecurity concept

Sensitive data: scope and purposes

We limit storage to what is strictly necessary for detection, alerts and preparation of your procedures. Third-party login credentials are processed through industry-standard tokenization and rotation mechanisms. You can revoke a connection at any time or request the export and then deletion of your file subject to legal obligations (billing, ongoing disputes). Technical logs allow you to audit who accessed what on the support team side — with internal traceability.

  • Systematic in-transit encryption
  • Subcontractors listed in the privacy policy
  • No resale of transactional data to advertising brokers

“Audit” mode

For accountants or family mediators: structured export of classification reasons and triggered alerts.

Service not found?

Submission of a raw label to enrich our repository — prioritized manual processing.

Fiscal calendar

Annotation of potentially deductible items depending on your situation — indication, not tax advice.

Multi-currency

Indicative conversion with course date — useful for cross-border workers and expatriates.

Generally within a few minutes for standard aggregators. Very large histories or complex PDF formats may require queuing or user validation.

As an extension, detection directly feeds the “Manage & optimize” modules (budget, tags, simulations) and the “Termination” paths (evidence, chronology). We recommend that you complete the digital view with a six-monthly manual pass: contracts with handwritten signatures or direct debits from secondary accounts often escape automatic synchronizations.

Moving from statement to decision

Learn how to budget, tag and simulate cuts once your map is established.

Manage & optimize

Back to home