Analytics for small companies: sales and customer numbers, read before you bet on them.
Rows in the file541,909
Clean sale lines522,568
The source file as a ruler: 541,909 rows, of which 522,568 are clean sale lines. The 19,341 rows set aside are magnified: 5,268 exact duplicates, 9,251 cancellations netted, 2,327 postage and fees, 2,495 zero or negative.
A customer tells you how they found you, and it never reaches the system.
Two reports, two different totals for the same month.
Three channels, and only a guess at which one carries the other two.
The data you already hold, made usable before anyone analyses it.
Pick this when you know what to do with the data, and its state is in the way.
One place to look each week, updated without anyone rebuilding it.
Pick this when the same report is rebuilt every month, or the team works from different numbers.
A real effect, or a coincidence you’re about to spend money on?
Pick this when a decision is expensive. Did the change work, or was that the season?
How the business really runs, and where the data comes from.
What’s captured, missing or inconsistent. This sets the ceiling.
What the numbers say, checked against how you operate.
The few things worth acting on now.
Clear options, with what each would take. The decision stays yours.
A UK retailer’s transactions, cleaned as shown at the top. Public data: check us.
Of the 522,568 clean sale lines, 24,754 fall in December 2011, a partial month, and are left out: 497,814 are in the chart’s window.
| Month | Net, £ | Share |
|---|---|---|
| Dec 2010 | £758,167 | 8.1% |
| Jan 2011 | £578,914 | 6.2% |
| Feb 2011 | £499,531 | 5.4% |
| Mar 2011 | £679,338 | 7.3% |
| Apr 2011 | £482,154 | 5.2% |
| May 2011 | £731,088 | 7.8% |
| Jun 2011, 1 flagged line, £38,970, kept | £723,970 | 7.8% |
| Jul 2011 | £676,845 | 7.3% |
| Aug 2011 | £701,382 | 7.5% |
| Sep 2011 | £1,011,365 | 10.8% |
| Oct 2011 | £1,061,718 | 11.4% |
| Nov 2011 | £1,427,125 | 15.3% |
| Twelve months | £9,331,596 | 100.0% |
The flagged June line: 60 × £649.50, re-ordered as one set five minutes later. Kept: the retailer’s question, not ours to delete. Two other lines pass the test: one cancelled in January, one outside the window.
Same twelve months
customers, about 1 in 18, brought half of identified customers’ revenue.
The largest order: 74,215 storage jars, cancelled 16 minutes later. Left in, January reads 13% high.
of net revenue has no customer ID (123,628 sale lines). Flagged, not deleted.
From a real conversation
Reporting gaps and missing history were the symptom, not the thing to fix.
We expected dashboard work. The problems sat a level below, in how information was captured. Shared with permission.
So: a worked example on public data, not a client list. We price accordingly.
.describe(
In Python, .describe() is the first thing you run on an unfamiliar dataset: rows, averages, extremes, gaps. You look first. The bracket stays open: what goes inside is your business.
I’m Wafa’a Al-Hayek, and I run .describe( from Gaza.
I studied business administration, joined a software bootcamp in 2022, and moved to data analytics in 2023: analytics, monitoring and evaluation, reporting. Most data work goes wrong on the question, not the technique.
The team is me, for now; a software engineer and a data analyst join as work grows. You’ll always know who’s on your project.
A number doesn’t have to be wrong to mislead you.
We show our method, name what we excluded, and say when a finding is weaker than you hoped. You should be able to check us.
From data to decision
من البيانات إلى القرار