How to Review a Data Deck Before You Send It (2026 Checklist)
A useful review pass has three layers, and the order matters. First, do the numbers agree with the source. Second, does every chart say what it appears to say. Third, does each slide state a conclusion rather than a topic. Formatting comes last, and it is where most people start.
The order matters because the layers are not equally expensive to fix. A wrong number found after sending costs a correction email. A misaligned logo costs nothing.
This guide covers what to gather first, the three-pass manual review, and where the routine breaks down when the deck is rebuilt every month.
What you need before you start
You need the source file the deck was built from, not a copy of the summary. Reviewing a deck against another summary confirms only that you copied consistently.
You also need the definitions you used. Date range, currency, which records were excluded and why. Without them, the review cannot tell a deliberate exclusion from an accident.
Two decisions come first.
Who the reader is. An executive deck and a working-session deck fail differently. The first fails by being too detailed, the second by being too thin to interrogate.
What decision it supports. A data deck that supports no decision is a report, and reviewing it against decision criteria will produce pointless changes.
One habit saves the most time. Keep the review checklist in the deck file itself, so the next person reviewing a data deck starts from the same list.
How to do it manually
Option 1: Reconcile the numbers three ways
Every figure in a deck usually appears in three places: on a chart, in a table, and in prose. Check all three against the source and against each other.
The prose is where errors hide. A chart gets regenerated when the data updates, and a sentence written two weeks ago does not. The text then quietly describes an earlier version of reality.
Then check the totals. Rebuild each headline number from the rows beneath it rather than trusting the cell it came from. A filtered range or a stale reference produces a plausible figure with nothing behind it.
This pass finds the expensive errors. It is also the pass people skip when time is short, which is exactly backwards.
Option 2: Audit every chart for honesty
Read each chart as a sceptic would. Does the axis start where it should, are both axes labelled with units, and does the title describe what the chart actually shows?
Three checks catch most problems. Truncated axes on bar charts overstate differences, because bar length encodes the value. Unlabelled dual axes let a reader infer a relationship the data does not support. Missing sample sizes make a percentage from twelve responses look like a percentage from twelve hundred.
Also check that the chart type matches the claim. A line chart asserts that time is the relevant dimension, and using one for unordered categories asserts something untrue.
The ceiling here is that this pass verifies presentation rather than substance. A perfectly labelled chart can still be the wrong chart for the argument.
Option 3: Check the narrative and the appendix split
Read only the slide titles, in order, and nothing else. They should form an argument. If they read as a list of topics, the deck has no spine and the reader will supply their own.
Then check what earns a place. For each slide, ask what changes if you remove it. Anything that survives that question stays; the rest belongs in an appendix where it can be produced on request.
Finish with the last slide. A data deck that ends on a chart hands the conclusion back to the audience. Name the recommendation, the owner and what you need.
The shared ceiling. All three passes are manual and they take an hour on a deck of any size. Because they are expensive, they get shortened under deadline, and the pass that gets cut is almost always the first one.
Where the manual route slows down
The first review is thorough. The fourth monthly rebuild is not.
The reason is structural rather than lazy. Each rebuild changes the numbers but not the prose, the axis limits or the slide titles. The review therefore has to re-verify things that look unchanged. Re-verifying something that appears identical is the least motivating work there is.
Degradation follows a predictable path. Month two drops the three-way reconciliation. Month four stops rebuilding totals from rows. By month six the review is a proofread, and a proofread catches typos rather than wrong numbers.
The error that eventually escapes is rarely dramatic. It is one figure that stopped updating, quoted confidently for two quarters.
How to review a data deck with Powerdrill Bloom
Step 1: Upload the source data
Upload the file the deck was built from. Powerdrill Bloom profiles the columns on arrival, so exclusions, blank blocks and type mismatches are visible before you check a single slide.
Step 2: Ask the reconciliation questions in natural language
Ask for the headline figures rebuilt from the underlying rows, broken out the same way the deck breaks them out. Then compare that output against what the slides say.
Ask the honesty questions in the same pass. Ask which comparisons rest on small samples, and which movements sit inside normal variation rather than outside it.
Step 3: Export the chart, report, or deck
Take out corrected charts, a short written list of figures that did not reconcile, or replacement slides.
Why this beats a proofread every month
| Manual review | Powerdrill Bloom | |
|---|---|---|
| Rebuilding totals from rows | By hand, each figure | Ask for all of them at once |
| Prose that lags the chart | Read and compare manually | Compare stated figures against recomputed ones |
| Small-sample comparisons | Spot them by eye | Ask which rest on thin samples |
| Next month | Repeat the whole pass | Upload the new export |
The first row decides whether the review survives. When rebuilding every total costs one request, you do it every month. When it costs an hour, you do it once and then trust the deck.
Common mistakes
Starting with formatting. Fonts and alignment are visible, so they attract attention first. They are also the cheapest errors to leave in.
Reviewing against the previous deck. Comparing to last month's numbers checks consistency, not correctness. Two consecutive decks can be wrong the same way.
Trusting a chart because it regenerated. A refreshed chart with a stale caption is the most common form of quiet error. Read the words next to every updated figure.
Leaving the axis on automatic. Automatic ranges change when the data changes, so the same chart can tell a different story next month. Set the limits deliberately, then state what you set.
Sending without reading the titles alone. The title-only read takes ninety seconds and exposes a missing argument faster than any other check.
Reviewing your own data deck immediately after building it. You will read what you meant rather than what is there. Leave an hour, or hand it to someone else.
Treating the appendix as a dumping ground. An appendix nobody can navigate is the same as no appendix. Order it to match the deck.
Conclusion
Reviewing is a habit rather than an event. A written checklist is what keeps that habit alive when the deadline is short. Keep it in the file, not in your head.
Review in order of cost: reconcile the numbers, audit the charts, then check the narrative. Formatting is last because it is the cheapest thing to get wrong.
What breaks the practice is the monthly rebuild, where everything looks unchanged and quietly is not. If that is where your reviews degrade, try Powerdrill Bloom on the source file behind the deck. See also our guides to building a monthly business review deck and analyzing data before converting it to PowerPoint. The Excel to PPT and AI graph maker pages cover the build side.
Frequently asked questions
What should I check first when reviewing a data deck?
Reconcile the numbers against the source, including the figures written in prose. Wrong numbers are the most expensive error to ship, and text is where outdated figures usually survive.
How long should a deck review take?
Budget an hour for a deck that supports a real decision. Most of it goes on rebuilding headline figures from the underlying rows rather than on reading slides.
How do I know if a chart is misleading?
Check whether the axis starts at zero for bar charts, whether both axes carry units, and whether sample sizes are stated. Those three checks catch the majority of misleading charts.
Should every slide have a conclusion in the title?
Yes for a decision deck. Read the titles in sequence with nothing else visible; if they do not form an argument, the reader will construct their own from the charts.
What belongs in the appendix rather than the deck?
Anything whose removal would not change the conclusion. Keep it ordered to match the main sequence so you can find a slide while someone is waiting.