Data & customer operations · November 2026

Synthetic / simulated practice project

Customer records,
ready for review.

A customer data cleaning exercise for a fictional online store. I used Python and spreadsheet outputs to standardize records, control duplicates, and document the details that still need confirmation.

Record accuracy Duplicate control Exception reporting

Source rows
500
Unique customer IDs
470
Duplicate rows excluded
30
Review queue entries
77

01 / Project brief

Prepare a consistent
customer-level handoff.

The source mixes name casing, phone layouts, date formats and city spelling. Some records repeat; others lack contact details. The task was to produce one row per customer ID and an audit trail for uncertain fields.

My contribution

AI-assisted synthetic dataset design, deterministic cleaning rules, spreadsheet output and a documented validation pass. I checked the source CSVs against the workbook and reconciled each headline figure before preparing this case study.

PythonCSVExcel-compatible workbookData validationException reporting

Self-directed practice. No client engagement, real customer records, or measured business growth.

Inspect the Python cleaning rules

02 / Verified reporting output

A dashboard with
traceable totals.

Calculated from the supplied CSVs and reconciled against the original workbook. All figures describe generated practice data.

Customer IDs by city

470 retained records across 10 generated city categories.

  • Jakarta89
  • Malang82
  • Surabaya70
  • Yogyakarta44
  • Bandung37
  • Semarang37
  • Bogor35
  • Denpasar31
  • Makassar28
  • Medan17

Total: 470 IDs. Bar lengths are relative to Jakarta (89). This is not real market research.

What the 77 review entries contain

Count of review entries by issue type.

  • Excluded duplicate rows30
  • Missing emails25
  • Missing addresses16
  • Dates needing confirmation6

30 + 25 + 16 + 6 = 77 entries
The queue touches 72 distinct IDs. After excluding duplicate entries, 47 unresolved field issues remain across 46 retained IDs. An ID can have more than one issue.

464 of 470 registration dates populated
Six dates remain blank for confirmation. Missing email and address values also stay blank.

03 / Before and after

Show the decision behind the edit.

Selected field values from the actual source and clean CSVs. Personal contact values and names are omitted from this public view.

Original values → retained output → decision
Record / fieldBeforeAfterDecision
C0001 / City␠yogyakarta␠YogyakartaTrim spaces and use the city lookup.
C0001 / Registration date18-01-20252025-01-18Convert the known DD-MM-YYYY format to ISO.
C0011 / Registration date03/04/2026Blank; review queuedQuarantine the generator's deliberately ambiguous marker.
C0024 / Total_Spent␠1829952␠1829952Trim whitespace; preserve the integer value.

␠ represents a source space. Phone normalization changes layout to +62 format; it does not prove that a number is valid or belongs to a person.

04 / Methodology

Keep the source. Make each rule explicit.

  1. STEP 01

    Inspect and preserve

    Keep the original 500-row CSV separate. Identify blanks, repeated IDs and the generator's source conventions.

  2. STEP 02

    Normalize known fields

    Trim whitespace, normalize names and email casing, format phones, map city variants, and parse dates under the known conventions.

  3. STEP 03

    Retain one row per ID

    Keep the first observed source row for each Customer_ID. Exclude later rows without summing their Total_Spent values.

  4. STEP 04

    Log and reconcile

    Leave uncertain fields blank, record the original CSV row number, and reconcile the clean export, queue, dashboard and workbook.

Date policy and its limits

The generator uses YYYY-MM-DD, DD-MM-YYYY and MM/DD/YYYY. Its literal 03/04/2026 marker is quarantined. Other slash dates follow the generator's known MM/DD convention, including values that would be ambiguous without that context. This rule is specific to this exercise; a production import must confirm the source locale before parsing.

Field definitions and handling
FieldMeaning / handling
Customer_IDStable synthetic record key. “Unique customer” means one distinct ID, not a verified real person.
Full_Name, Email, Phone_Number, AddressFictional contact fields; formatting changes do not verify identity, deliverability, ownership or consent. Values omitted from public downloads.
CityGenerated location category; trim, title case and map the known Surabya spelling to Surabaya.
Registration_Date, Last_Purchase_DateGenerator dates normalized to YYYY-MM-DD under the documented source convention; unresolved marker dates remain blank.
Total_SpentInvented customer-level lifetime spend, in unspecified synthetic units. Retained total: 1,197,969,255. This is not revenue, growth, or a business outcome.

05 / Quality control

Reconcile the handoff,
then explain the gaps.

  • Reproduce all 470 cleaned rows and all 77 queue entries from the untouched source.
  • Confirm 500 − 30 = 470 and verify one row per Customer_ID.
  • Match workbook clean data, review queue, 60-row source sample, dashboard counts and city totals to the CSVs.
  • Check retained date order and duplicate spend conflicts. Neither check found a conflict in this dataset.

What still needs human review

Missing contacts and the six quarantined dates stay unresolved. Duplicate exclusions are audit events, not pending blank fields. First-observed-row retention is reproducible for this generated source; a real system may require a verified timestamp or source-priority rule.

The original workbook is a static snapshot with no formulas. Its matching totals do not imply that it refreshes when someone changes a cell. Names repeat across IDs, so name alone is not a reliable key.

Lessons learned

Define the counting unit before building a dashboard. State the date convention. Keep formatting separate from verification, and preserve the source row reference when handing an exception to another operator.

06 / Relevance to remote operations

Practical skills for the next handoff.

Demonstrated tasks and role relevance
RoleWhat this exercise demonstrates
Remote Data EntryConsistent formatting, missing-field checks and accurate record handling.
Customer OperationsContact-data hygiene, exception queues and clear escalation notes.
Sales SupportCustomer-level deduplication and reporting without double-counting repeated records.
Virtual AssistantSpreadsheet organization, documented procedures and a reviewable handoff.

07 / Evidence to inspect

Take a closer look.

The revised PDF includes the verified figures and limitations. The CSV contains all 470 retained IDs, city/date/spend fields and missing-field indicators. Names, emails, phone numbers and street addresses are omitted. The review queue retains issue-level traceability without contact details.

The original workbook and unredacted files remain local audit inputs. Generated phone numbers can resemble real numbers; omitting contact values avoids publishing them.