Four checks, in this order. Most of the time it is the first.
1. Check the period
Month to date is not last month. Year to date in January is three weeks. A figure compared against a different window will differ by exactly the amount you would expect it to.
This accounts for more "wrong numbers" than the other three together, and it is the fastest to rule out.
2. Check which sources are connected
A revenue figure from the bank alone counts money that arrived. The same figure with your accounting software connected counts work that was invoiced. Both are correct and they are not the same number.
Sources shows what each connected system reports, side by side, with the differences named.
3. Check what is uncategorised
Uncategorised money is not missing from your totals — it is missing from the breakdown. So a total can be right while the category you are looking at is understated.
If a category looks low, look at the review queue before anything else. See unreconciled transactions.
4. Open the records behind it
Every figure can be traced to what it was built from. See the evidence chain.
This is where it resolves either way. Either the records do not support the number, or — far more often — they do, and the input was incomplete: a job never costed, a transaction never categorised, a connection that stopped a fortnight ago.
If a figure has disappeared
A number that was there last month and is not now is usually data rather than the business. Anything the platform cannot support well enough is held back rather than shown with a caveat — see how confident a number is.
When your books and your bank disagree
That gap is expected and has its own reading. See books versus bank variance — a small one is timing, a growing one is something recorded in one place and not the other.
Still wrong
Tell us, with the figure, the period, and where you are looking. See getting help.