From a raw spreadsheet to a finished analysis

You open a spreadsheet someone sent you. Forty thousand rows, columns you did not name, a header that may or may not be a header and a question you are supposed to answer by end of day. The temptation is to jump straight to a chart. Skip that. The work that makes an analysis trustworthy happens before the first chart, and it is the work most people rush.

Here is the full journey from a raw file to a finished, grounded analysis. Most of it is unglamorous. All of it matters, because a clean number presented with no context is worse than no number at all.

First, understand what the data actually is

Before you compute anything, work out the shape of the file. This is the step people skip and regret. Ask four questions of every dataset.

• What is the grain? What does one row represent. One transaction, one customer, one customer per month, one invoice line. If you do not know the grain, every total you produce is suspect.
• What are the columns, really? A column named "value" could be revenue, units, a forecast or a code. Read a sample of actual rows, not just the header.
• What are the units? Pounds or thousands of pounds. Percentages stored as 0.12 or as 12. A single units mistake will be off by orders of magnitude.
• What period does it cover? A full year, a partial month, a rolling window. A figure compared across mismatched periods is meaningless.

Spend ten minutes here and you save an hour of rework. Skip it and you will present a number you cannot stand behind when someone asks where it came from.

Clean it and sanity-check it

Raw data is dirty. Assume it until proven otherwise. The cleaning pass is not optional polish, it is the difference between a figure you trust and a figure that quietly lies.

1. 1.Find the blanks. A column that is 30% empty cannot carry an average without a decision about what the gaps mean.
2. 2.Find the duplicates. Repeated rows inflate every total. A duplicate order doubles revenue without warning.
3. 3.Find the mixed types. Numbers stored as text will not sum. A stray label in a number column breaks the math silently.
4. 4.Check that totals foot. Add the parts and see if they match the stated whole. If they do not, something is missing, double-counted or in the wrong units, and you need to know which before you go further.

If your totals do not foot, stop. Do not present a single figure until you know why. A number that does not reconcile to its own source is the fastest way to lose a room.

Decide the question before you compute

An analysis without a question is just a pile of numbers. Decide what you are actually answering. Is spend up or down and why. Which segment is driving the change. Where is the risk concentrated. Are we on track against the plan.

The question dictates everything downstream: which figures matter, which comparison gives them meaning and which chart tells the story. Compute for the question, not for the sake of filling a slide. A focused analysis that answers one question beats a sprawling one that answers none.

Compute the figures that matter

Now the math. Keep it to the figures that move the answer, not every metric the data could produce.

• Totals and subtotals, by the dimensions that matter to your question (by region, by product, by month).
• Growth and change. Period over period, year over year. The direction and the size of the move.
• Ratios. Margin, conversion, cost per unit, spend per customer. A ratio often says more than a raw total because it normalises for size.
• Concentration. What share sits in the top few. If 80% of revenue comes from three accounts, that is the headline, not the grand total.

Compute these exactly. Rounding early, or eyeballing a sum across thousands of rows, is where errors creep in. The figures are the foundation. If they are wrong, the narrative built on them is wrong too.

Benchmark so a number means something

This is the step that separates an analysis from a data dump. A number alone tells the reader nothing. Revenue of 4.2 million. Is that good. You cannot say without a comparison.

Anchor every important figure against at least one of these:

• The prior period. Up 11% on last quarter tells a story. The raw figure does not.
• A peer or segment. This region versus the others. This product versus the category average.
• A plan or target. Ahead or behind, and by how much.
• A baseline or expectation. What would normal look like, and how far is this from it.

Benchmarking turns a fact into a finding. It is also where most decks fall short, because computing a comparison takes more effort than copying a total, so people skip it and present raw numbers that the reader has to interpret for themselves.

Turn it into a narrative with the right chart for each point

Only now do you build the story. Each point needs a chart that fits what it shows, and a clear so-what.

• Change over time: a line or column chart. Trends read instantly.
• Composition: a stacked bar or a proportion chart. What makes up the whole.
• Comparison across categories: a bar chart, sorted, so the ranking is obvious.
• Concentration: show the top few against the long tail, not a flat list.

Every chart needs a sentence that says what it means and what to do about it. "Spend rose 11% year over year, driven almost entirely by one category" is a finding. A chart with no so-what is a fact waiting for someone else to interpret it. That someone should be you.

The grunt work versus the thinking

Look back over those steps. Understanding the grain, cleaning, computing, benchmarking, that is the grunt work. It is mechanical, repeatable and necessary, and it eats the hours. The thinking is the question you chose and the so-what you drew. That is where your judgment lives, and it is the part that gets squeezed because the grunt work ran long.

That is the wrong allocation. A computer should do the mechanical work exactly and fast, and leave the judgment to you. Watch for the traps that sink the grunt work, because a flaw here flows straight into the conclusion:

• Summing an ID column. Order numbers and account codes are labels, not quantities. Add them and you get a nonsense figure that looks plausible.
• Trusting headers that are really data. A title row or a stray label read as a header shifts every column and names your figures after a value.
• Presenting raw numbers with no comparison. Without a baseline the reader cannot tell good from bad, so the analysis does no work.
• Rounding away the truth. 63.7 million rounded to 64 hides a real difference. Keep the precision the figure carries.

How Nexlyr AI does the heavy lifting

This is exactly the journey Nexlyr AI was built to carry. You give it the spreadsheet and a short brief. It reads the file and works out the structure on its own: the grain, the columns, the units, the period, including the case where the first row is data and not a header.

Then it computes. Not by guessing, which is where most AI tools go wrong. Nexlyr AI runs a real analysis engine: it writes SQL and runs it against your actual data, so the totals, growth, ratios and concentration are calculated from your source, exactly. It benchmarks by default, comparing against prior periods and across segments, so the figures arrive already meaning something. And it grounds every fact and figure in your own data. If the source does not support a number, the number is not shown. It will not invent one to fill a gap.

Operational trust comes from one rule: every figure traces back to your own data. You can hand the result to a client or a colleague knowing the numbers are computed from the source, not made up to look right.

What you get back is a grounded first draft: a fully editable analysis with the right chart for each point and a clear so-what, in a branded PowerPoint deck you can present as is or refine. Where standard frameworks fit the material, it uses them, never forced. And once the draft exists, a "Think further" pass reviews the finished work like an analyst, raising the questions, risks and what-ifs you would want a sharp colleague to flag before you walk into the meeting.

The grunt work is done, exactly and in minutes. Your time goes where it should: to the judgment only you can bring.

Keep exploring

• Budget variance analysis: what actually moved versus plan
• Writing an investment memo that stands up to scrutiny
• Spreadsheet to presentation