KENYA · GUIDE 0590 · EDITION 4.2
Data cleanup service in Kenya: Operations and Quality Checks
Deliver the agreed result consistently
Clean a bounded dataset using agreed rules for names, dates, duplicates, missing values, and exceptions, preserving originals for comparison and approved recovery.
The operating sequence for this offer is:
- Preserve an untouched original and agree field rules, duplicate criteria and privacy access before cleaning a representative sample for client approval.
- Apply approved transformations to a copy, log changed records and route uncertain matches or missing values to the client instead of inventing replacements.
- Reconcile record counts and exceptions, deliver cleaned data with the change log and verify that rollback to the original remains possible.
Put acceptance into the job record
Original records remain recoverable and uncertain matches are flagged.
Have the owner review ambiguous matches and sample corrected records, checking that legitimate distinctions survive rather than treating every similar entry as duplication.
Inspect at the point where a correction is still possible, then again at handover. Record rejected or incomplete output separately from accepted output. A completed task is not automatically an accepted task, and an invoice does not establish that the customer received the right result.
Measure real capacity
Time preparation, the main operation, transport, cleaning, approvals and correction as applicable. The slowest stage determines how much you can promise. Where two orders need the same person, space or equipment at the same time, resolve the conflict before confirming both deadlines. An additional enquiry should not erase a commitment already made.
| Operating checkpoint | Evidence for this business |
|---|---|
| Before starting | Clean a bounded dataset using agreed rules for names, dates, duplicates, missing values, and exceptions, preserving originals for comparison and approved recovery. |
| During delivery | Apply approved transformations to a copy, log changed records and route uncertain matches or missing values to the client instead of inventing replacements. |
| At acceptance | Have the owner review ambiguous matches and sample corrected records, checking that legitimate distinctions survive rather than treating every similar entry as duplication. |
| At the pilot review | Keep original copies, transformation rules, changed-record counts, ambiguous cases, owner decisions, validation findings, and the delivered mapping between old and cleaned values. |
Keep a correction and escalation rule
When inputs, access, specifications or customer instructions change, pause the affected work and confirm the effect on time and cost. Do not quietly lower the acceptance standard to avoid a delayed delivery. Keep rework caused by your error distinct from a customer-requested change of scope and agree the remedy clearly.
Cleanup damages the dataset when two different customers sharing a name are merged or an empty field is replaced with an invented assumption.
Retain the minimum records needed to investigate a complaint without disclosing unrelated personal information. Use measured delivery and actual acceptance as the basis for expanding capacity, rather than the maximum theoretical output of a tool or production line.
Operating control points for this exact offer
| Sequence | Work method | Release checkpoint |
|---|---|---|
| 1 | Define allowed fields, category dictionary and duplicate criteria with owner. | A person's preferred name is not changed merely to match a guessed naming convention. |
| 2 | Store an unchanged client-controlled source and assign source-row IDs. | Every output row can be traced back even if sorting changes. |
| 3 | Trim agreed spaces and map only approved category variants; keep identifiers/text types. | Do not convert all phone numbers to numbers or infer country codes without evidence. |
| 4 | Show exact-email/code matches and contradictory fields to owner. | Two people can share a reception email; candidate matching is not proof of identity. |
| 5 | Report 500 input rows, retained rows, owner-approved merges and unresolved cases, with changes logged. | No record vanishes without a documented rule/approval; reopen export through intended import route. |
Do not substitute a tick for an observation: record the count, measurement, approved sample or evidence reference used at each checkpoint. Mark unresolved work as held and identify who can approve the recheck.
Diagnose a failure before repeating it
| Observed symptom to investigate | Check the cause | Proposed response |
|---|---|---|
| Phone loses initial 0 | Check data type/import conversion. | Restore from unchanged source as text and retest export. |
| Two different people merged | Inspect duplicate rule and conflicting fields. | Undo merge and require owner adjudication. |
| Category mapping changes meaning | Compare dictionary and original value. | Return ambiguous values to unresolved list rather than forcing a category. |
These are diagnostic prompts, not proof of the cause in a real incident. Separate suspect output, record any correction and repeat the relevant acceptance check. Link rejected units, rework time, replacement materials, customer credits and collection delays to the actual job result.