To investigate pipeline count inflation, compare the reported row count with a distinct count of the deal identifier. Metabase defines these as separate measures.
Key takeaways
When the deal count in a pipeline coverage review looks high, the number often reflects how the data was joined, grouped, sequenced, or displayed rather than the actual number of unique opportunities. Investigating these six specific behaviors helps separate the reported row count from a distinct count of deal identifiers.
- Write down whether the deal count uses rows or distinct deal identifiers.
- Inspect joins for repeated matches and retained unmatched rows.
- Check the
observedanddropnasettings and your pandas version. - Verify a report showing exactly 10,000 rows with a distinct query or export.
Pipeline count inflation: distinguish rows from unique deals
The first step in tracing a high pipeline total is identifying whether the report counts physical rows or unique deal identifiers. According to Metabase, the "Count of rows" measure calculates the total number of rows in the table after any filters have been applied. This metric reflects the volume of data records present in the query result set. In contrast, the "Number of distinct values of…" measure calculates the number of unique values in all the cells of a single column. If your pipeline report uses the row count measure, every record in the filtered dataset contributes to the total, regardless of whether multiple records reference the same deal.
A discrepancy between these two measures often signals join expansion or duplicate entries in the underlying data. When a query joins a deals table with an activities or line items table, the resulting dataset may contain multiple rows for a single deal identifier. The row count measure captures this expanded volume, while the distinct value measure isolates the unique deal identifiers. For a coverage review, the distinct count of the deal identifier typically represents the actual number of opportunities, whereas the row count may reflect the number of associated data points.
For one coverage report, write down whether the displayed deal count counts rows or distinct deal identifiers, then inspect the joins feeding that measure. This documentation helps the number presented in the review align with the intended definition of a pipeline deal. If you need to reconcile these totals across different dashboard views, see Reconcile sales dashboard totals with 4 checks.
Joining records: repeated matches and retained unmatched rows
When a pipeline query joins two tables, the resulting row count can exceed the number of unique deal identifiers. The PostgreSQL documentation describes how different join types handle row matching and retention. A cross join produces a row for every possible combination of rows from the first table and the second table, creating a Cartesian product where the joined table contains all columns from both sources PostgreSQL.
For matching joins, the documentation states that for each row in the first table, the joined table includes a row for each row in the second table that satisfies the join condition with the first row PostgreSQL. A pair of rows matches if the ON expression evaluates to true, meaning multiple valid matches for a single deal identifier will generate multiple output rows. If a deal record in the primary table matches several activity records in the secondary table, the query returns one row for each match, repeating the deal identifier in the result set.
Left joins introduce a different counting behavior by retaining unmatched rows from the first table. The documentation notes that for each row in the first table that does not satisfy the join condition with any row in the second table, a joined row is added with null values in the columns of the second table PostgreSQL. This means deals without matching activity records still appear in the output, but they do not multiply the count; they simply appear once with null values for the unmatched side. When investigating count inflation, distinguish between rows multiplied by multiple matches and rows retained despite no match.
Inspect the join logic feeding your coverage measure to determine which behavior applies. If the query uses a cross join or a matching join with multiple valid pairs, the row count will reflect the number of matches rather than the number of unique deals. If it uses a left join, the count includes deals with no matches but does not duplicate them for non-matching cases. For how activity records link to deals, see Activity linking: 8 rules for deals and visibility.
Grouping records: categorical defaults and missing values
When a pipeline query groups deals by a categorical field, the number of resulting groups depends on how the tool handles values that are not present in the current dataset. In pandas, the observed setting controls which categorical values appear in the group output. According to pandas.DataFrame.groupby — pandas 3.0.6 documentation, if observed is True, the output shows only observed values for categorical groupers. If the setting is False, the output shows all values for categorical groupers. This distinction matters for count inflation: a False setting can produce group rows for categories that have no deals in the current slice, inflating the number of groups compared to the number of actual deals.
The default behavior for this setting changed in a specific version. The documentation notes that in version 3.0.0, the default value for observed is now True. If you are using a version earlier than 3.0.0, the default may differ, and your group count might include unobserved categories unless you explicitly set observed to True. Check your specific version’s documentation to confirm which behavior applies to your environment.
Missing values introduce a separate grouping behavior. The dropna setting determines how NA values are treated during grouping. According to the same pandas.DataFrame.groupby — pandas 3.0.6 documentation, if dropna is False, NA values will also be treated as the key in groups. This means deals with missing categorical values form their own group rather than being excluded from the count. If you are investigating why a group count seems higher than expected, check whether dropna is set to False. If it is, the missing values contribute to the total group count as a distinct key.
Merge validation offers a way to verify the structure of your data before grouping. According to pandas.merge — pandas 3.0.6 documentation, if a merge type is specified, the function checks if the merge is of that specified type.
To investigate this aspect of count inflation, start by identifying the categorical fields used in your pipeline grouping. Check the observed setting to see if unobserved categories are being included. Then check the dropna setting to see if missing values are forming their own groups. Finally, verify that any joins feeding into the groupby operation use the expected merge type. These checks isolate whether the group count reflects actual deals or includes structural artifacts from categorical defaults and missing value handling.
Reading totals: sequence gaps and displayed row limits
When a pipeline total looks high, check whether the number reflects actual records or a display limit. Metabase documents that by default, the platform limits the number of aggregated rows displayed to 10,000 rows. That limit applies to the rows shown, not to the count inside an aggregate. A report showing exactly 10,000 rows may be at this limit. Verify the actual record count through a distinct query or export to know if deals are missing from the view or simply hidden by the limit.
Sequence numbers in CRM tools can also mislead about record counts. Airtable explains that Auto Number fields guarantee uniqueness, not contiguity. If you delete a record, the numbers are not re-flowed, so you will see gaps in the sequence, such as 1, 2, 4, 5. You should not use the last number in the sequence to estimate how many deals remain in the pipeline. The sequence is a unique identifier, not a counter of active items.
If a continuous sequence is required for reporting, Airtable notes that Auto Number alone will not provide it. To get a truly gap-free sequence, you would use a formula or an automation that recalculates position. The trade-off is that such position numbers can shift when records are added or removed. This makes them unreliable for tracking historical record counts. For pipeline coverage reviews, rely on distinct deal identifiers rather than sequence numbers or display limits. Record whether your count is a row count or a distinct identifier count, and note any display ceilings applied by your reporting tool.
Count behavior reference table by publisher
The table below maps each documented count behavior to its source publisher. Each row cites the specific rule that affects how a pipeline deal count is calculated or displayed.
| Publisher | Documented Count Behavior | Source Qualifier |
|---|---|---|
| PostgreSQL | Matching joins return each qualifying match. | "match if the ON expression evaluates to true" |
| PostgreSQL | Left join retains unmatched left rows. | "a joined row is added with null values in columns of T2" |
| pandas | The default value for the observed parameter changed in version 3.0.0. | "Changed in version 3.0.0: The default value is now True." |
| Metabase | Aggregated rows displayed are limited to 10,000 by default. | "By default, Metabase will limit the number of aggregated rows displayed to 10,000 rows." |
Keep the count definition and merge validation on file
This distinction is the first line of your investigation file. When you trace repeated deal identifiers, you are checking if the query structure multiplied records. The count definition must be explicit so that any future reviewer can reproduce the same number from the same data state.
Record the specific merge type used in your pipeline query. If you used a left join to retain unmatched rows, that setting is part of your count definition. If you used a matching join to return only qualifying pairs, that is a different definition.
Your documentation should list the exact join type and the validation status. This prevents confusion when the deal count changes between reporting periods. A change in the join type can alter the row count without any change in the underlying deal records.
Next step: today, write the count definition and join type for your current coverage report in your review notes.
FAQ about the deal count in a coverage review
How can I investigate pipeline count inflation?
Compare the reported row count with a distinct count of the deal identifier. Then inspect the joins feeding that measure, grouping settings, sequence gaps, and display limits. This approach separates documented behaviors from actual data errors.
Does the categorical grouping default apply to my pandas version?
For pandas 3.0.0 and later, the default value of observed is True; earlier versions may differ. This setting controls whether only observed values are shown for categorical groupers, so verifying your version is essential before interpreting group counts.
What is the difference between counting rows and counting distinct deal identifiers?
Counting rows provides the total number of rows in the table after any filters have been applied. Counting distinct values provides the number of unique values in all the cells of a single column. These two metrics measure different aspects of the data structure.
Can an Airtable Auto Number tell me how many deals remain?
Not reliably: the Auto Number field guarantees uniqueness, not contiguity. If you delete a record, the numbers are not re-flowed, so you will see gaps in the sequence.
What does the Metabase aggregated row display limit measure?
By default, Metabase limits the number of aggregated rows displayed to 10,000 rows. This limit applies to the visual output of the query results. It does not prove that underlying records were deleted or excluded from the database.