BUSI 3014 Week 4 Data Quality Audit Example

Auditing a Year of Job Records Before Measuring First-Visit Wait: 2,412 Rows Checked for Completeness, Accuracy and Timeliness, With a Count for Every Problem and Its Likely Effect on the Saturday Route Question

Student Name

Walden University

BUSI 3014: Data Driven Decision Making Workplace Certificate

Instructor Name

Month Day, Year

Auditing a Year of Job Records Before Measuring First-Visit Wait: 2,412 Rows Checked for Completeness, Accuracy and Timeliness, With a Count for Every Problem and Its Likely Effect on the Saturday Route Question

The File and the Standard It Is Held To

The file was pulled from the scheduling and job software at Kestrel Pest Solutions, the Knoxville composite used across this course. It holds 2,412 rows, one for every new residential request logged in 2027, with the logged timestamp, the request type, the customer-date flag, the scheduled date, the first completed visit date, the status and the dispatcher's free-text notes.

The audit asks one question of each problem: would it change first-visit wait as defined last week, and in which direction? Wang and Strong (1996) found that the people who use data judge its quality by whether it fits their purpose, not by abstract tidiness. A misspelled street name is untidy but harmless here; a missing visit date is not.

The three dimensions are defined in a sentence each. Completeness asks whether the fields the measure needs are filled. Accuracy asks whether the values that are filled are correct. Timeliness asks whether values were recorded when the event happened.

Findings

Table 1

Data Problems Found in the 2027 Request Export

Kestrel Pest Solutions and its records are composite. Shares are of all 2,412 rows.

Completeness

Eighty-eight requests have no logged timestamp. For 70 of them, the dispatcher's notes record the date of the call, which can stand in for the missing timestamp; the remaining 18 cannot be measured and are dropped. Twenty-three requests are marked completed but have no visit date. These are almost all from two weeks in June when a technician's tablet failed and visits were closed out later from paper. They are held aside rather than guessed.

Accuracy

Forty-one rows are duplicates created when a job was rebooked and the dispatcher entered a new request instead of moving the old one. Keeping both would count the same customer's wait twice, and the second row starts the clock late, so the duplicates are removed and the original row kept.

Seven visits are dated before their requests. On inspection these are typing errors in the year or month. With no way to tell the right date, they are dropped.

Twelve commercial accounts were coded as residential. Their visits follow contract schedules, so they are removed under last week's exclusion.

Sixty-one requests have notes such as "wants after the 20th" or "call back after vacation" but no customer-date flag. These customers chose to wait, and counting them would inflate the share of long waits. They are reclassified as customer-date requests and excluded.

Timeliness

The subtlest problem is timing. In 312 requests, nearly one in eight, the dispatcher's notes record a call date at least one business day before the logged timestamp. Most fall on Mondays, when weekend voicemails are entered, and in May and June, when the dispatcher was handling the most calls. For these rows, the measured wait starts late and understates the real wait by one to two days.

This bias is not random. It is heaviest in the peak months, which are exactly the months the Saturday route question is about. Leaving it alone would make peak-month waits look better than they were and tilt the answer against the route. Redman (1998) argued that poor data quality quietly distorts decisions in ordinary businesses; this is a concrete case of that.

The audit therefore recommends adjusting the start point for these rows to the call date in the notes, and reporting the measure both ways so the owner can see how much the adjustment matters.

What Survives Cleaning

After removing duplicates, impossible dates, unrecoverable blanks, visits with no date and miscoded commercial accounts, and after applying the definition's exclusions, including 214 requests already flagged for a customer-chosen date, the 61 reclassified ones and 96 requests canceled before any visit was scheduled, 1,940 requests remain usable for first-visit wait. The 96 cancellations are kept in a separate table for the cancellation rate. They matter because a canceled request is a wait that never ended; Kaplan and Meier (1958) developed estimates of time to an event that use such incomplete observations instead of discarding them, an approach worth applying if the cancellations turn out to be customers who gave up. The audit also suggests two process fixes: a rule that rebooked jobs are moved, not re-entered, and a voicemail timestamp field so the start point no longer depends on the notes.

References

Kaplan, E. L., & Meier, P. (1958). Nonparametric estimation from incomplete observations. Journal of the American Statistical Association, 53(282), 457-481. https://doi.org/10.1080/01621459.1958.10501452

Redman, T. C. (1998). The impact of poor data quality on the typical enterprise. Communications of the ACM, 41(2), 79-82. https://doi.org/10.1145/269012.269025

Wang, R. Y., & Strong, D. M. (1996). Beyond accuracy: What data quality means to data consumers. Journal of Management Information Systems, 12(4), 5-33. https://doi.org/10.1080/07421222.1996.11518099