Improve Business Reporting Accuracy
You open the dashboard on Monday and revenue for last month reads $212,400. Your bookkeeper's report says $208,950. The sales lead insists the real figure is higher than both. Nobody is lying, and nobody is exactly wrong. You just spent twenty minutes of a leadership meeting arguing about a number, and the decision you came to make never got made.
If you have stopped trusting your dashboard, you are probably right to be careful, and also probably not facing a software bug. In my experience, inaccurate reporting comes from a small set of causes that repeat across businesses of every size. This article walks through them in the order I check them, with a trust checklist at the end that you can use this week.
Start with the definition, not the chart
The most common reason two reports disagree is that they answer two different questions with the same word. "Revenue" is a good example. It can mean invoices issued, payments received, revenue recognized under accounting rules, or bookings signed. All are legitimate, and none are interchangeable. If the dashboard counts invoices issued and your bookkeeper counts cash received, they will differ every month by the amount of timing between those two events.
So before you touch any data, write a one paragraph definition for each number you care about. A usable definition answers these questions.
- What exactly is counted, in plain words?
- What is the source system or table it comes from?
- What is included and excluded, such as refunds, taxes, internal test orders and cancelled orders?
- Which date drives it, such as order date, invoice date, payment date or ship date?
- Who owns the definition and can approve a change?
A short metric specification with those five answers, kept in a shared document, settles an astonishing share of arguments. It also protects you when someone leaves, because the meaning of "active customer" no longer lives in one person's head.
Same word, different rule
Watch for these frequent collisions.
| Term | Variant A | Variant B |
|---|---|---|
| Revenue | Invoiced | Cash received |
| Customer | Anyone who has ever bought | Bought in the last 12 months |
| Active user | Logged in this month | Performed a billable action |
| Margin | Before labor | After labor and overhead |
| New lead | Any form fill | A form fill that passed qualification |
If two people use the same term and you cannot say which variant each means, you have found a problem worth fixing before anything technical.
Grain, the idea that explains many wrong totals
Grain is the level of detail that one row in a table represents. One row per order, one row per order line, one row per customer per month. Mixing grains is how totals get silently inflated.
Here is a simple trap. You have a table of orders, one row per order, with a total. You have a second table of order lines, one row per product on the order. You join them to see revenue by product category, and then sum the order total column. An order with four lines now appears four times, so its total is counted four times. The report looks plausible and is wrong.
The defenses are straightforward.
- Know the grain of every table you use, and write it down. "One row per order line" is a sentence you should be able to say.
- Before you join, ask whether the join multiplies rows. A join from one order to many lines does.
- Sum measures at the grain where they belong. Order totals belong at the order level. Line amounts belong at the line level.
- Compare counts before and after each join. If the row count changed unexpectedly, stop and look.
Reconcile to the source
A reconciliation is a check that two independent sources agree, within a tolerance you have chosen. It is the single most powerful trust building habit in reporting, and the one most often skipped.
You pick a total that exists somewhere authoritative and compare. Revenue in the dashboard against the payment processor's payout reports. Inventory in the system against a physical count. Payroll in the report against the payroll provider's register. Bank deposits against recorded receipts.
A reconciliation has four parts.
- The two sources. One is your report, and one is an independent source of truth.
- The scope. The same period, the same entity, the same currency, the same definition.
- The tolerance. An exact match for money counted in cents, or a small percentage where timing blurs things.
- The explanation of differences. A list of known reasons, such as pending payouts, refunds in transit and fees.
A worked example (illustrative)
Suppose your dashboard shows $212,400 in revenue for October and your payment processor shows $208,950 in settled payments. The gap is $3,450. These numbers are invented to show the process.
You investigate and find two explanations.
- Two invoices for $2,100 and $850 were issued on October 30 and paid on November 2, so they are in the dashboard but not in October's settled payments. That accounts for $2,950.
- One customer was refunded $500 on October 29 and the refund is recorded in the processor but not yet in the dashboard. That accounts for $500.
$2,950 plus $500 equals $3,450. The gap is fully explained, and the two reports were each correct about different things. You have also learned something useful. The dashboard does not net out refunds promptly, which is a real defect you can now fix. Without the reconciliation, you would have had an argument and no learning.
If the gap had not been explained, you would have a list of unexplained items to chase, which is exactly what you want.
Time zones and the moving edge of "today"
Time handling produces some of the most baffling errors, because everything looks fine until you examine the boundary.
Systems usually store timestamps in UTC, a single universal reference. Your business lives in a local time zone. An order placed at nine thirty at night in Puerto Rico on March 31 is already April 1 in UTC. If a report buckets orders by the UTC date, that order lands in the wrong month, and your month-end revenue shifts by whatever you sell in the last four hours of the day.
Puerto Rico uses Atlantic Standard Time all year and does not observe daylight saving. That is convenient locally, but it means the offset between your business and any mainland tool that does observe daylight saving changes twice a year. Systems that assume a fixed offset will quietly break by an hour each spring and fall.
The rules I follow are these.
- Store timestamps in UTC, and convert to the business time zone only when you group by day, week or month.
- Decide which time zone defines "a day" for the business, and write it into the metric specification.
- Test the boundary. Look at the orders in the last and first hours of a month, and confirm they land where you expect.
- When you compare two systems, check that both use the same time zone for the cutoff.
Late-arriving data and restated numbers
Numbers change after you first see them. A payment is reversed. A vendor invoice arrives ten days after the month closed. A sales rep logs a deal two days late. A system that was down catches up and backfills. This is called late-arriving data, and it is normal. The accuracy problem comes from pretending it does not happen.
You have three honest options.
- Close the books. Pick a date after which a period is frozen, and report later changes in the period when they occur. Accountants do this for good reason.
- Restate. Allow past periods to change, and make the dashboard show when a figure was last updated. The number for October may differ on November 5 and November 20, and that is acceptable if people can see it.
- Show a provisional label. Mark recent periods as incomplete. A week-to-date bar that looks low because the data has not all arrived yet causes false alarms.
Whichever you choose, tell people. A short note on the dashboard, such as "figures for the last 5 days may change," prevents a surprising amount of distrust.
Freshness is part of accuracy
A figure that was correct at 6 AM and is shown at 4 PM may be wrong for the reader's purpose. Display the last refresh time on every dashboard. If a data feed fails, the dashboard should say so loudly instead of quietly showing yesterday's numbers as if they were current. This matters most after storms or outages in Puerto Rico, when a source system may stop updating for days and a dashboard that does not announce it will mislead you at the moment you most need it.
Other places accuracy leaks out
- Duplicates. The same customer, order or payment exists twice, usually from a retry or a manual entry on top of an import. Decide on a unique key for each record type, and check for repeats.
- Missing values. Blank fields are dropped from sums and averages without warning. A column of discounts with blanks can overstate the average price.
- Manual overrides. Someone types a number over a formula or edits a record to "fix" a report. Track manual adjustments openly, in a log, instead of letting them hide.
- Currency and units. Pounds and kilograms, dollars and a foreign currency, hours and minutes. Label units on every figure.
- Test and internal data. Demo orders and staff purchases that were never meant to be counted. Flag them so they are excluded consistently.
- Changed definitions. Someone updated a rule in the middle of the year, and trend lines now compare unlike things. Version your definitions and note the change date.
A trust checklist
Use this as a monthly review. If you can answer yes to every line for your top five numbers, you can start to relax.
| Question | Yes or no |
|---|---|
| Does each key metric have a written definition with an owner? | |
| Is the source system for each metric named? | |
| Is the grain of each underlying table documented? | |
| Do totals match an independent source within the stated tolerance? | |
| Are differences between the two sources explained in writing? | |
| Is the business time zone defined and tested at month boundaries? | |
| Is the last refresh time visible on every dashboard? | |
| Are late-arriving changes handled by a stated rule, close, restate or provisional? | |
| Are duplicates checked for with a unique key? | |
| Are test, internal and cancelled records excluded consistently? | |
| Are manual adjustments logged with who and why? | |
| Is there a named person who can explain every figure when asked? |
What to ask your vendor or analyst
If someone else builds or maintains your reports, these questions reveal a lot.
- Can you show me how this number is calculated, from source to screen?
- Which time zone does the report use to define a day?
- How do you handle refunds, cancellations and corrections after the period has closed?
- How often does this refresh, and what happens when a source fails?
- Can I see the reconciliation to the source system?
- What changed since last month's report, and why?
A good answer is specific and usually comes with a document. A vague answer, or a reply that "the system handles that," means the work has not been done.
What you can do on Monday morning
- Pick the three numbers you use to make real decisions.
- Write a five line specification for each, covering what is counted, source, inclusions and exclusions, driving date and owner.
- Choose one and reconcile it to an independent source for last month. Write down every difference and its cause.
- Add the last refresh time to the dashboard if it is not already there.
- Tell your team which numbers are provisional and which are final.
You will probably find at least one real error in the first hour, and that is a good sign. It means you are now looking. If you would like help setting up the reconciliations or documenting your definitions, Ozcorp InfraOps does this kind of work for small teams, with InfraOps Base starting at $500 per month and vendor and license charges billed separately. Either way, the habit matters more than the tool.
Discuss your infrastructure requirements
Review the current service offer or schedule a conversation about the scope you need.
Stay in touch
Follow Ozcorp
Leave your details for occasional notes about new articles and Ozcorp updates.