Skip to content
Africa Tech FestivalMeet us in Cape Town, 17–19 Nov.

Data Migration

Clean data before anything moves.

Profiles every source system, matches duplicate records and scores readiness, so you know what to fix before cut-over.

Used by
Migration leads, data owners and test teams
You receive
A readiness score, golden records, rules and reconciliation checks
Proof
Telco data migration: 239 tables profiled with no manual profiling

The problem

Data quality eats the migration budget.

The work centres on records that describe the same thing but differ across systems.

Most

of a migration budget goes on manual data quality work

6+ months

of spreadsheet-driven reconciliation across teams

Weeks

of analyst time lost to matching records by hand on each migration

Where the fragmentation comes from

Same subscriber, different identity

One subscriber exists across CRMs under different identifiers, statuses and account links.

Same product, divergent codes

One product appears under different codes, prices and bundle definitions per system.

State changes with no audit trail

Uncontrolled entry points let offerings change state with no record of who changed what.

Contracts on obsolete pricing

Live contracts reference rate plans that no longer exist in any current catalogue.

Integration gaps between platforms

No system is the agreed master, so reconciliation has no reference point.

Analyst effort restarts each time

Each migration starts the matching work from scratch.

Most migration effort goes into deciding which record is correct. Data Migration proposes those decisions for a person to approve.

How it works

From source profile to approved change.

  1. Discover

    Profiles every table, column and relationship in the source estate.

  2. Match

    Resolves entities and aligns catalogues and pricing across systems.

  3. Reconcile

    Checks that the target matches the source, run after run.

  4. Approve

    A person approves each change the model proposes.

Privacy

The AI only sees masked values, never real customer data. It runs inside your network.

Used by

Migration leads, data owners and test teams

Modules

Six modules score readiness.

Five modules run on their own, each with its own score and review screen, so you can start with one and add the next. Golden Record Assembly runs last.

  1. Entity Resolution

    Uniqueness

    Matches and deduplicates subscriber records across source CRMs using fuzzy matching and AI reasoning.

    Score weight 0.25

  2. Catalogue Alignment

    Cross-system consistency

    Reconciles product codes, pricing tiers, offering bundles and hierarchies across divergent catalogues.

    Score weight 0.20

  3. Pricing Reconciliation

    Contract integrity

    Aligns rate plans, discounts, contract terms and billing arrangements across every source.

    Score weight 0.20

  4. Anomaly Detection

    Business rule compliance

    Finds data that violates business logic regardless of source: negative prices, orphaned accounts, impossible dates.

    Score weight 0.15

  5. Semantic Enrichment

    Metadata completeness

    Classifies columns with business-level types where the source systems carry no documentation at all.

    Score weight 0.10

  6. Golden Record Assembly

    Master data accuracy

    Builds the authoritative version of each entity from the other modules’ corrections, so it runs last.

    Score weight 0.10

Discovery

Map the estate first.

Three passes build a catalogue of every table, column and inferred relationship, then render it as a graph to navigate.

  1. Quick scan

    Row counts, column lists and table schemas.

  2. Full profile

    Distinct values, null counts and the most common values for each column.

  3. Deep analysis

    Seven AI checks infer business meaning, personal data, relationships and anomalies.

Seven AI checks

  1. Business meaning
  2. Personal data
  3. Hidden relationships
  4. Field mapping
  5. Anomaly explanation
  6. Quality advice
  7. Readiness

For hidden relationships and field mapping, the AI checks its own guesses by running a query against your data.

  • Subscriber to Account
  • Subscriber to Contract
  • Subscriber to Product
  • Subscriber to Address
  • Account to Rate plan
  • Contract to Product
  • Contract to Address
  • Product to Rate plan
  • Product to Bundle
  • Bundle to Invoice
  • Rate plan to Invoice
Inferred relationships, not declared foreign keys.

On a telco data migration

239

tables profiled

75

relationship edges

0

manual profiling hours

Governance

People decide what the system may change.

Every AI action comes with a confidence score and its sources. You set how much the system may do without approval.

Deterministic

Fixed rules

Seven rule-based checks cover structure, fill rate, format, cross-reference, formula, duplicates and distribution. No model cost where no judgement is needed.

Judgement

AI enrichment

Business meaning, personal data, hidden relationships and anomaly explanation, using the best model for each task.

Authority

Human governance

A three-level autonomy ladder, an immutable decision ledger and a confidence score on every output.

The autonomy ladder

  1. Read-only

    Autonomous

    Discovery, profiling, business meaning and anomaly detection.

    Your role: review and validate findings.

  2. Proposal

    Human-gated

    Entity resolution, field mappings, corrective fixes and catalogue alignment.

    Your role: approve, reject or modify each proposal.

  3. Execution

    Human-budgeted

    Merges above the confidence threshold, and writes and runs fix scripts.

    Your role: set thresholds and audit the results.

Immutable decision ledger

Readiness

Readiness you can measure.

Each module adds a weighted score. The total is a go/no-go readiness score you can track each sprint.

72%Conditional

After catalogue alignment

Quality score history

RunEntitiesCataloguePricingOverall
Baseline62%45%51%53%
After entity resolution87%45%51%61%
After catalogue alignment87%78%51%72%

Verdict bands

  • ReadyReady for migration.At or above 80%
  • ConditionalProceed with documented risks.50% to 79%
  • Not readyAdditional cleanup required.Below 50%

Artefacts delivered

  • Golden records with per-field source attribution
  • Versioned rules for which value wins
  • Reconciliation checks with run history and trend charts
  • Transformation and validation SQL
  • Immutable decision ledger
  • Quality score history per module
Proof

Validated on a telco data migration: 239 tables profiled, 1,666 tests passing and 13 of 13 reconciliation checks passed.

See EcoGenie on your own architecture.

Bring a dataset you plan to move. We’ll show you its readiness score.