The same rows.
Different answers.

Start with a decision you need to make. Change what counts and see how the result moves, then check the calculation against the records.

Three worked examples with fictional data.

Sales · May 5–8, 2025

Which region brought in the most revenue?

For a sales review, start with completed orders. Each row is one order and each total is in US dollars. A canceled or pending order contributes nothing to completed revenue.

Decision. Start the review with East, which contributed $360 of the $740 in completed revenue. Track pending orders in a separate pipeline figure.

Order value by region

$740 from 5 completed orders

  • East$360
  • West$265
  • Central$115

East leads. The $120 canceled order and the $90 pending order are out of scope.

$950 across all 7 submitted orders

  • East$480
  • West$355
  • Central$115

Including every submitted order adds $210 to the total. East still leads.

Inspect 7 orders
OrderPlacedRegionStatusUSD
1042May 5EastCompleted210
1043May 5WestCompleted170
1044May 6CentralCompleted115
1045May 6EastCanceled120
1046May 7EastCompleted150
1047May 8WestCompleted95
1048May 8WestPending90
Run the calculation

Download the orders fixture and load it in SQLite. This query calculates both views for the same dates.

SELECT region,
       SUM(CASE WHEN status = 'completed' THEN total_usd ELSE 0 END)
         AS completed_usd,
       SUM(total_usd) AS submitted_usd
FROM orders
WHERE ordered_on >= '2025-05-05' AND ordered_on < '2025-05-09'
GROUP BY region
ORDER BY completed_usd DESC;
See the prompt
Use DewPoint. Build a sales review dashboard showing completed revenue and order counts by region from May 5 through May 8, 2025. Use USD and exclude canceled and pending orders.

Customer support · May 8, 2025 UTC

Which team needs help with first replies?

Each row is one ticket opened on May 8. At the May 9 midnight UTC snapshot, every ticket has had at least an hour to receive a first reply. Count replies sent within 60 minutes, then divide by the tickets that had a chance to meet that target.

Decision. Review Account’s queue first. Two of its four eligible tickets received no reply by the snapshot. The rate among answered tickets hides that backlog.

First reply within 60 minutes

All eligible tickets

  • Account50% 2 of 4
  • Billing75% 3 of 4

Billing leads. Unanswered tickets count in the denominator once their hour has passed.

Answered tickets only

  • Account100% 2 of 2
  • Billing75% 3 of 4

Account appears to lead when its two unanswered tickets drop out of the denominator.

Inspect 8 tickets
TicketTeamOpened UTCMinutes to first reply
201Account09:0021
202Account09:1553
203Account11:00No reply
204Account11:30No reply
205Billing09:009
206Billing10:0022
207Billing14:0058
208Billing15:0095
Run the calculation

Download the support and fulfillment fixtures and load them in SQLite. The one-hour cutoff ensures each ticket in the report had time to meet the target.

WITH cohort AS (
  SELECT team,
         CASE WHEN datetime(opened_at, first_reply_minutes || ' minutes')
                   <= '2025-05-09 00:00:00' THEN first_reply_minutes END AS reply_at_cutoff
  FROM tickets
  WHERE opened_at >= '2025-05-08 00:00:00'
    AND opened_at <= '2025-05-08 23:00:00'
), counts AS (
  SELECT team, COUNT(*) AS eligible,
         COUNT(reply_at_cutoff) AS answered,
         SUM(CASE WHEN reply_at_cutoff <= 60 THEN 1 ELSE 0 END) AS in_time
  FROM cohort
  GROUP BY team
)
SELECT team, in_time, eligible, answered,
       ROUND(100.0 * in_time / NULLIF(eligible, 0), 1) AS coverage_pct,
       ROUND(100.0 * in_time / NULLIF(answered, 0), 1) AS answered_pct
FROM counts
ORDER BY team;
See the prompt
Use DewPoint. Compare first-reply coverage by support team for tickets opened May 8, 2025 UTC. Use the May 9 midnight UTC snapshot. Count replies within 60 minutes and include unanswered tickets once their reply window has elapsed. Show the numerator and denominator beside the rate.

Fulfillment · Start of May 9, 2025

Which warehouse needs attention today?

Each row is one shipment. At the start of May 9, review orders with a promised ship date from May 5 through May 8. Count shipments that left after their promised date and overdue orders still waiting to leave.

Decision. Start with Warehouse A’s two overdue orders. Warehouse B has a recorded late shipment, which matters for a retrospective but is already out the door.

Share that missed the ship date

Includes overdue orders still open

  • Warehouse A50% 2 of 4
  • Warehouse B33.3% 1 of 3

Warehouse A has the higher missed-deadline rate. The May 10 shipment is outside this window.

Shipped orders only

  • Warehouse A0% 0 of 2
  • Warehouse B33.3% 1 of 3

Warehouse B has the higher rate when the report only covers orders that already shipped.

Inspect 8 shipments
ShipmentWarehouseDueShipped
301AMay 5May 5
302AMay 6May 6
303AMay 7Open
304AMay 8Open
305AMay 10Open
306BMay 6May 7
307BMay 7May 7
308BMay 8May 8
Run the calculation

Download the support and fulfillment fixtures and load them in SQLite. A shipment dated after the snapshot counts as open at the time of the review.

WITH period AS (
  SELECT warehouse, due_on,
         CASE WHEN shipped_on < '2025-05-09' THEN shipped_on END AS as_of_shipped
  FROM shipments
  WHERE due_on >= '2025-05-05' AND due_on < '2025-05-09'
), counts AS (
  SELECT warehouse, COUNT(*) AS due,
         COUNT(as_of_shipped) AS shipped,
         SUM(CASE WHEN as_of_shipped > due_on THEN 1 ELSE 0 END) AS late,
         SUM(CASE WHEN as_of_shipped IS NULL THEN 1 ELSE 0 END) AS overdue_open
  FROM period
  GROUP BY warehouse
)
SELECT warehouse, late, overdue_open, due, shipped,
       ROUND(100.0 * (late + overdue_open) / NULLIF(due, 0), 1) AS missed_pct,
       ROUND(100.0 * late / NULLIF(shipped, 0), 1) AS shipped_late_pct
FROM counts
ORDER BY warehouse;
See the prompt
Use DewPoint. Show missed ship dates by warehouse as of the start of May 9, 2025. Use orders due May 5 through May 8. Include both orders that shipped late and orders still open after their due date. Keep the two counts visible.

Before you publish a number

Ask what one row represents, which records qualify, and how the result can be checked independently. Save the dashboard once those answers match the decision you are making.