Internal Data

Data-Audit Report: Buhi Supply Co. B Y D U R C H D E N W A L D G LO B A L

This report outlines the findings by Durchdenwald Global (DG) from a database audit for Buhi Supply Co. In this audit, DG assessed the integrity and usability of Buhi’s data without making any changes or corrections to it.

Trustworthy data is a critical element in making data-driven business decisions with confidence. Therefore, we recommend that Buhi address errors and issues in its data to the fullest extent possible.

ERRORS AND POTENTIAL ISSUES FOUND

DG’s audit uncovered the following issues and potential problems in the data.

MISSING VALUES

Missing values are blank or null values in cells where an actual value is expected.

We found 126,061 instances of missing values in the database.

Examples

These examples represent the most common patterns of missing values seen in Buhi’s data:

• The image_type field in the campaigns table has 80 instances of missing values.

• The employee_surveys table has 32 total missing values in the job_satisfaction, mgr_relationship, and coworker_relationship fields. If those values correspond to optional responses in the survey, their absence may not be detrimental, since having some blank optional responses would be expected.

PAGE 1

Recommendation

We advise Buhi to review each missing value in its database and determine if its absence could impair decision-making. If a missing value is likely problematic, it should be addressed either by deleting the entire record or interpolating the value using known data.

OUTLIERS

Outliers are data points that differ noticeably from the other equivalent values in the data. They represent truthful, valid data. However, their inclusion in an analysis can greatly skew the results, producing inaccurate outcomes.

We found 3,754 instances of outliers in the database.

Examples

These examples represent the most common patterns of outliers seen in Buhi’s data:

• Most order_total_revenue values in the orders table ranged from $35 to $4,500. However, in 1,820 instances, they were $150,000 or more.

• Most impressions values in the campaigns table ranged from zero to 1,600. However, in 86 instances, they were 2,000,000 or more.

Recommendation

We advise Buhi to consider the impact of outliers on its data analysis. If it’s significant, rows containing outliers should either be removed from the database or filtered out in queries.

IMPOSSIBLE VALUES

Impossible values don’t belong in the range of acceptable values. Unlike outliers, impossible values cannot possibly be truthful values of the data they represent.

We found 1,548 instances of impossible values in the database.

PAGE 2

Examples

These examples represent the most common patterns of impossible values seen in Buhi’s data:

• Values in the feature_score field in the product_feature_survey table ranged between 1 and 5. However, in 63 instances, the value is not in the 1-5 range. It’s unclear how this could have happened with software that allows a maximum response value of 5.

• In 312 instances, the age value in the cust_segmentation_survey table is a negative number; in one instance, it’s 128.

Recommendation

We advise Buhi to remove impossible values from its database and then treat them as missing values.

INCONSISTENT VALUES

Inconsistent values are those that may be incorrect based on the data around them. They aren’t necessarily impossible, but their validity is questionable.

We found 41,222 instances of inconsistent values in the database.

Examples

These examples represent the most common patterns of inconsistent values seen in Buhi’s data:

• In 1,177 instances, the bag_count value in the cust_segmentation_survey table is greater than 100. Not an inconceivable number of bags for a customer to own, but it does warrant confirmation.

• The role of one employee in the human_resources table is listed as “software engineer.” However, the salary_wage value for that record is 21,000 — which seems low for the position.

PAGE 3

Recommendation

We advise Buhi to investigate all database values flagged as inconsistent. If a value can be corroborated by data in other tables or data sources, the flag can be removed. Otherwise, it may be necessary to delete the value from the database and then treat it as a missing value.

ERRONEOUS FORMATTING

Erroneously formatted values arise when data is entered with an incorrect format; the value may be accurate, but it may not be usable in its present form in the database.

We found 4,261 instances of erroneous formatting.

Examples

These examples represent the most common patterns of erroneous formatting seen in Buhi’s data:

• In the order_line_item table, 400 quantity values are spelled out instead of being numeric values; for example, “six,” “eight,” “twenty,” and “twenty-seven.”

• In the expenses table, 628 expense_amount values are spelled out, either with the word “dollar” (“eighty-seven dollars”) or without (“two hundred and twenty”).

Recommendation

We advise Buhi to convert all instances of erroneously formatted values to their proper format. If a value is too ambiguous to determine, we advise removing it from the database and then treating it as a missing value.

DUPLICATE RECORDS

Duplicate records can occur when information from the same row is repeated more than once by mistake. A form may be submitted twice due to a technical mishap; that would be a duplicate record.

PAGE 4

However, it’s entirely possible a customer may purchase the same item for the same price at two different times; that wouldn’t be a duplicate record, but two genuine records of two separate transactions.

We found 27,049 possible instances of duplicate records in the database.

Examples

These examples represent the most common patterns of duplicate records seen in Buhi’s data:

• In the delivery table, there are five records containing the same delivery_id value. Since the ID is uniquely generated, it’s almost certain these are accidental duplicates.

• There are three records in the cust_segment_survey table containing exactly the same information in all fields except survey_id. It’s unlikely that three different 26-year-olds named Caoimhe Banasiewicz in Wyoming, United States, completed the survey on the same date and with all the same responses.

Recommendation

We advise Buhi to identify and investigate any records that appear to be duplicated. If a record is determined to be a duplicate, it should be removed, leaving a single instance of the record in the database.

PAGE 5