Techbanq

Data · migration

A migration can move every row and still lose what the data meant

The count matches, the rows all arrive, the new system loads clean — and a field that meant "signed" in the old system now means "sent", a customer history has collapsed into its latest state, and a link the old database enforced is gone. None of that shows in a row count. A migration is not judged on how much data moved but on whether its meaning, its history and its relationships survived the move, which is a different and harder problem than transfer.

Enterprise delivery credentials

  • ISO

    ISO/IEC 27001

    ISO/IEC 27001 certified

Certifications & registrations

The clean-looking disaster

The migration that finished on time and corrupted the meaning of a field

We do not have an office in Saudi Arabia

We serve organisations there from our Riyadh office and our engineering hubs in Lahore and Karachi, travelling for the parts of an engagement that need to happen in person.

A migration that goes badly rarely looks like it at first. The project finishes, the record counts line up, the new system is live, and everyone moves on. The damage surfaces weeks later, in ones and twos: an invoice generated against a status that used to mean one thing and now means another, a customer whose earlier orders have vanished because only their current record came across, a report that will not balance because a relationship the old database quietly enforced was dropped on the way. By then the old system may be gone, the context that would explain the gap with it, and what looked like a clean migration is a slow leak of trust in the new one.

The reason this happens is that a row count is the one thing a migration can prove easily and the least thing that matters. Data carries meaning that lives outside the values themselves — in what a status code stands for, in the history behind a current figure, in the relationships between records that a schema enforces and a copy does not. Moving the values without carrying that meaning is not a migration; it is a transcription that happens to compile. The honest work is to understand what the data actually means in the old system before moving it, map that meaning deliberately into the new one, and prove the result matches the source in substance and not merely in count.

How data is lost

Three ways a migration loses data while every row seems to arrive

Each of these passes a row count and fails the business. They are the losses that do not announce themselves, which is exactly why the checking has to look past the count.

  1. 01

    A field that meant one thing now means another

    The old system used a status code whose meaning lived in a lookup or in people's heads. Copied as a raw value into a new system that reads it differently, the record is intact and wrong at the same time.

  2. 02

    A history flattened into its latest state

    The source held how a record changed over time; the migration took only the current snapshot. The row is there, but the past that explained it — the earlier orders, the prior addresses, the audit trail — did not come with it.

  3. 03

    A relationship the old system enforced and the new one forgot

    The old database guaranteed that every order pointed to a real customer. Migrated as flat tables without that constraint, orphaned records slip in unnoticed and the reports that depend on the link quietly stop balancing.

The method

How data is moved with its history and meaning carried across intact

The order is deliberate. Most of the risk is retired before anything is loaded, by understanding the source and proving the mapping on a copy rather than discovering the losses in production.

STAGE 01

Profile the source before touching it

We examine what the old data actually contains — the real values in each field, the codes in use, the gaps and the duplicates — because a mapping built on what the schema claims rather than what the data holds fails on the exceptions. A true picture of the source, not the documented one.

STAGE 02

Map each field to its new meaning

We decide deliberately what each value becomes in the target — how a status code translates, how history is preserved, how a relationship is reconstructed — rather than copying column to column and hoping the meaning survives. A mapping that carries meaning, not just data.

STAGE 03

Move a copy and reconcile it

We migrate into a copy of the target and check the result against the source in substance — totals, distributions, sampled records, the relationships that should hold — so the losses are found in a rehearsal, not after go-live. Proof the migration preserved substance, not only count.

STAGE 04

Cut over with a way back

We plan the switch so the old system remains available to compare against and to fall back to until the new one has been trusted with real use, rather than betting everything on an irreversible cutover night. A cutover the business can reverse if it has to.

Our role

What we carry across and what your data owners verify

  1. 01

    Understanding before moving

    We profile the source and surface what the data really means and where it is inconsistent, so the move is designed against reality rather than against a schema diagram that has drifted from it.

  2. 02

    A move that proves itself

    We reconcile the migrated data against the source in substance and show you the result, so acceptance rests on evidence the meaning survived, not on the fact that the counts happen to match.

  3. 03

    Your owners as the final check

    Your people know what a record is supposed to mean; we give them a reconciliation they can read and sample, because the person who owns the data is the one who can confirm it arrived intact.

What the data is supposed to mean, and which discrepancies are acceptable, are judgements only your data owners can make; we give them a profiled source, a deliberate mapping and a reconciliation they can verify, so signing off the migration is reading evidence rather than trusting a promise.

Common questions

What buyers ask

How do we know the migrated data actually matches the old system?+

By reconciling substance, not by comparing counts, and by doing it before the old system is retired rather than after. A matching row count proves almost nothing — the same number of records can arrive with their meaning changed, their history dropped or their relationships broken. Real verification checks the things a count cannot see: control totals on key figures, the distribution of values in important fields, a sample of individual records compared field by field against the source, and the relationships that should still hold in the target. We run that reconciliation against a copy first, so any discrepancy is found and explained in a rehearsal, and we keep the old system available to compare against through cutover. The point is that you accept the migration by reading evidence, not by trusting that clean-looking output is correct.

Can we move to the new system without a weekend of downtime?+

Often yes, and the answer depends on how much data there is and whether the business can tolerate a brief freeze on changes. For a smaller dataset, a planned window — a quiet evening rather than a whole weekend — is usually enough to migrate and reconcile with the old system as a fallback. For a large or constantly changing dataset, we migrate the bulk while the old system stays live, then sync only what changed in a short final cutover, so the freeze is minutes rather than days. What we will not do is skip the reconciliation to save time, because a fast cutover onto unverified data is how the clean-looking disaster happens. We plan the downtime around proving the data is right, not the other way round.

What happens to years of history the old system holds?+

That is a decision to make deliberately, because history is the part a migration most often loses by accident. Some systems only store a current state, and if your old one recorded how records changed over time — past statuses, superseded values, an audit trail — that history will not survive a copy of the current snapshot unless it is migrated on purpose. We work out with you what history the business genuinely needs: sometimes it must be carried into the new system in full, sometimes it is enough to preserve it in an accessible archive that can be queried but is not live, and sometimes a defined cut-off is acceptable. What matters is that the choice is made knowingly rather than discovered months later when someone asks for a record that no longer exists. We will not claim to preserve history the source never actually recorded, but whatever the old system holds, we make its fate a decision, not an accident.

Audit the source first

A migration plan that names what the source data really contains, how each field maps to its new meaning, and how the result is reconciled before the old system is retired.

Start the conversation