Learn Excel formulas by writing them
New: a Modern Excel track — FILTER, SORT, UNIQUE and LET, the functions most tutorials predate.
You can follow any Excel tutorial and still freeze on a blank cell. Writing formulas is a different skill from watching someone write them — this trains that one.
- exercises
- 90
- tracks
- 11
- free to play
- 34
Total the sales in B6
Type a formula, press Enter| Select all cells | A | B | C |
|---|---|---|---|
| 1 | Product | Sales | |
| 2 | Cordless drill | 1240 | |
| 3 | Extension lead | 385 | |
| 4 | Safety goggles | 156 | |
| 5 | Work gloves | 227 | |
| 6 | Total |
Click B6 and start with an = sign. This is the real thing — same grid, same engine as every exercise.
One real problem, one missing formula
Every exercise drops you into a working spreadsheet with the answer taken out — a total that isn't there, a lookup returning #N/A. You write the fix, not just read about it.
Browse all exercises- 01
Pick something to fix
Every exercise is a small job with the answer missing: a total that is not there, a lookup returning #N/A, a column that will not sort.
11 tracks · 90 exercises
- 02
Type the formula
Into a real grid, with the keys you already use: F2 to edit, arrow keys to build a reference, Tab to accept a function name.
=SUM(B2:B5)
- 03
Find out exactly what broke
Press Check. You get the cell that is wrong and the reason it is wrong — a range one row short, a missing $, text that only looks like a number.
B6 · one row short
Practise as much as you like without an account. Sign in and your progress follows you between devices, your streak survives a new laptop, and you show up on the leaderboard.
Today's challenge
A new exercise every morning, plus every day that came before it — a back catalogue that grows by one a day, whether or not you were here for it.
Solve it before midnight UTC to keep your streak going.
Solve itFor the sheet in front of you, not the one in the exercise
The exercises teach you the functions. Formula AI is for the Tuesday afternoon when the file on your screen looks nothing like an exercise.
- Describe it, get the formula — you write it in your own words, a formula you can paste comes back
- Paste a formula, get the explanation — including the one a colleague left behind three jobs ago
- Ask anything Excel — why a lookup misses, which function to reach for, how to lay a sheet out
Formula AI is part of Pro. The interface is open to everyone, so you can see exactly what you would be paying for before you pay for it.
You type
Flag every invoice more than 30 days old as Overdue
Formula AI turns it into
=IF(TODAY()-B2>30,"Overdue","")
Or the other way round
=INDEX(B:B,MATCH(A2,C:C,0))
Finds A2 in column C and returns the value on that row from column B — a lookup that reads leftwards, which VLOOKUP cannot do.
Why this and not a video
Same hour, same functions, two different results. Here's the difference, stated plainly.
| What you can actually do | Formula Exercises | YouTube Tutorials | Traditional Online Courses |
|---|---|---|---|
| Type formulas into a working grid | Yes | No | Sometimes |
| Find out immediately whether you got it right | Yes | No | Sometimes |
| Be told what kind of mistake you made | Yes | No | No |
| Pick up where you left off | Yes | No | Yes |
| Works without installing Excel | Yes | Yes | Yes |
| Costs nothing to start | Yes | Yes | No |
What practising earns you
Not because points matter, but because starting tomorrow's session does. XP, levels, streaks and badges all exist to make the next one easier to begin.
XP for every solve
10 for a Beginner exercise, 20 for Intermediate, 35 for Advanced, and up to 10 more for a first-try answer with no hints.
Levels
XP compounds into levels, and the first few arrive in one sitting — the encouragement is front-loaded on purpose.
Streaks
One exercise a day keeps it alive. Miss a day and it goes back to zero, which is the only threat this site will ever make.
Achievements
There are 9: finishing a track, finishing one without a single wrong answer, practising thirty days straight.
And then there is everyone else
Two global boards, five windows each. Today's board is empty every morning, so somebody who started last year has no head start on you today.
- XP earned
- Counted from the exercises you actually solved and priced by difficulty — not from a number your browser reports.
- Longest streak
- Consecutive days of practice: live on today's board, best-inside-the-window on the rest.
TodayThis weekThis monthThis yearAll time
Players appear under a stable nickname, never an email. Solves count from the moment you are signed in.
The path, start to finish
11 tracks, each ending in something you can use at work. Difficulty ramps inside a track, not across the catalogue.
- 01Free
Foundations
Add up, average and count anything without touching a calculator.
9 exercises
- 02Free
Logic & Decisions
Make your spreadsheet decide things for you with IF.
9 exercises
- 03Free
Text & Cleanup
Turn messy exported data into something you can actually use.
9 exercises
- 04Free
Everyday Maths
Conditional totals, rounding rules and the operators nobody teaches.
7 exercises
- 05Pro
Lookups
Pull a value out of a 10,000-row table in one formula. The job-interview skill.
7 exercises
- 06Pro
Advanced Logic
Replace towers of nested IFs with formulas you can still read next month.
5 exercises
- 07Pro
Statistics
Describe a dataset properly — spread, not just averages.
11 exercises
- 08Pro
Dates & Time
Deadlines, working days and durations — the stuff that silently breaks reports.
10 exercises
- 09Pro
Finance
Loan payments, present value and depreciation — the analyst toolkit.
8 exercises
- 10Pro
Power Tools
The formulas that make colleagues ask how you did that.
6 exercises
- 11Pro
Modern Excel
Dynamic arrays: one formula that fills a whole range and resizes itself.
9 exercises
When one grid is not enough
Guides for the things that take more than an exercise to explain — the index is laid out as a workbook you can filter and sort.
- Data Model
The Data Model: The Board Pack Beat Target by £3,900.75 and Every Region In It Missed by More Than £28,000
Fourteen orders, six customers, three regions and a half-year pack whose bottom line read £63,900.75 of sales against a £60,000.00 target — £3,900.75 ahead, 106.50% of plan, correct to the penny. Above that line the same pivot said North was £44,305.40 behind, South £43,198.85 behind and East £28,595.00 behind, because the Targets table had been added to the Data Model and never joined to anything. A measure that no filter can reach returns its unfiltered value, so every region row showed the whole £60,000.00 target, and the grand total was right precisely because the grand total is the one cell where 'no filter reached me' is the true answer. Nothing errored, no cell was blank, and the three line variances add to −£116,099.25 against a total row of +£3,900.75 — a gap of exactly £120,000.00, which is the target counted two extra times. This guide covers what the Data Model actually is and why it is not a set of worksheets; how filters travel from the one side of a relationship to the many side and nowhere else; why the missing join produced a repeated column rather than an error; the Regions dimension that fixes it and why Excel refuses to join Customers[Region] to Targets[Region] directly; the relationship rules Excel enforces and the four it does not; implicit versus explicit measures, DIVIDE and BLANK; COUNT against COUNTROWS against DISTINCTCOUNT; the SUMIFS and XLOOKUP version of the same report, which cannot fail this way because it cannot stay silent about the join; four checks; and twelve traps.
24 min read
- Unpivot
Unpivot: The Q3 Number Reconciled to the Grand Total, and Both of Them Were June to August
Eight products, six months across the top, and a quarterly pack that read Q2 £59,043.80, Q3 £72,198.05 and total £131,241.85 — three figures that add up to each other exactly. Q3 was really £83,495.15, the 4% commission pool paid £2,887.92 instead of £3,339.81, and September, the largest month of the year at £30,367.10, was in none of the three. The sales export had been pasted over the same block every month since April and arrived in September with March in front of it, so every column moved one to the right; =SUM(E2:G9) kept pointing at the same cells and those cells were now June, July and August. Nothing errored, because a pasted value does not move a reference the way an inserted column does, and the reconciliation passed because all three totals are positional over one block. This guide covers the fix end to end: why a column heading is data and a cross-tab therefore stores one variable in the geometry of the sheet; the flat 48-row table the same numbers should live in; unpivoting by hand with INDEX, QUOTIENT and MOD in any version of Excel; the Microsoft 365 one-liner with LET, SEQUENCE, TOCOL and HSTACK, and why broadcasting makes the three columns line up by construction; Power Query's Unpivot Other Columns and the one menu word that decides whether a new month is absorbed or dropped; SUMIFS keyed on a name instead of a range; two-way INDEX/MATCH for when you must stay on the grid, and the HLOOKUP typed index that returns 910.60 where 12,085.75 belongs; one SUMIFS that rebuilds the whole cross-tab from the flat table; blanks, zeros, totals rows and text month names; four checks; and twelve traps.
23 min read
- Random Sampling
Random Sampling: The Audit Sample Was Four Invoices, and It Was a Different Four Every Time the File Was Opened
Fourteen supplier invoices worth £40,231.00, a spot check of four drawn at random, and a working paper that says "no exceptions found on £11,166.80 of spend, 27.76% coverage". The workbook was reopened a week later to answer a reviewer's question and the four highlighted rows were a different four, worth £7,228.40 — 17.97% coverage — because the sample lived in a column of =RAND() and RAND redraws on every recalculation, every open, every keystroke anywhere in the file. Nothing errored, nothing went red, and the coverage figure in the summary cell obligingly recalculated to match the new sample, so the working paper now describes a test that was never performed. This guide covers drawing a sample you can defend: what volatile really means for RAND, RANDBETWEEN and RANDARRAY; the F9-and-paste-values step that turns a draw into a record, and the ISFORMULA check that proves it happened; why RANDBETWEEN picking row numbers hands you the same invoice twice 37.46% of the time; the RAND plus RANK.EQ construction that samples without replacement; SORTBY with RANDARRAY for the Microsoft 365 one-liner; the 49.45% of four-invoice samples that contain neither invoice over £8,000; stratified and value-weighted sampling for populations where five rows carry 86.71% of the money; systematic sampling and the random start that decides the whole result; why Excel has no seed and what to record instead; four checks; and twelve traps.
21 min read
Start with the free tracks
Four complete tracks, thirty exercises, no card and no account. If it turns out not to be for you, you have lost nothing.
Questions people ask before they start
How can I practise Excel formulas online for free?
Open the first track and start typing. There are 90 exercises here, the first four tracks are free to solve, and nothing asks for a card or an email before you find out whether you like it. Each exercise gives you a grid with real data in it, you write a real formula, and it is checked by running it — not by matching the text you typed.
How is this different from an Excel video course?
A video shows you a formula being written. This makes you write it. That sounds like a small difference and it is the whole difference: watching VLOOKUP is not the same skill as producing VLOOKUP with a deadline on you and somebody else's messy export in front of you. Ten minutes of typing beats an hour of watching.
Which Excel formulas should I learn first?
SUM, then IF, then COUNTIF and SUMIF, then VLOOKUP or XLOOKUP, then INDEX/MATCH. That is the order the tracks here run in, because each one leans on the last: SUMIF makes sense once IF does, and INDEX/MATCH only makes sense once you have hit VLOOKUP's limits yourself.
What happens when my formula is wrong?
You get the specific cell, the value your formula produced, and the value it should have produced — not the word "incorrect". Reading that gap is most of what learning formulas actually is. Every valid route to the right answer is accepted, so =B2+B3 and =SUM(B2:B3) both pass.
Can I use this to prepare for an Excel job interview or assessment?
Yes, and it lines up closely with what those tests check. Employer skills tests and interview exercises hand you a sheet of tasks built on IF, VLOOKUP, COUNTIF, absolute references and INDEX/MATCH — the same formulas drilled here. Most of them also score your method rather than only your number, so typing the answer by hand scores zero even when it is right.
Do I need to know Excel already to start?
No. Exercise one is a single addition, and the difficulty climbs from there one new idea at a time. You need no finance or analytics background — if you can open a web page you can start, and by the end of the free tracks you are writing SUMIF and COUNTIF.
Which Excel version do these exercises work with?
The formulas transfer to Excel 2016, 2019, 2021, Microsoft 365 and Excel for the web, and to Google Sheets for most of what is drilled here. The exceptions are the newest additions — XLOOKUP, FILTER, SEQUENCE and the other dynamic arrays need 365 or 2021 — and the reference pages say so on the functions where it matters.
How long does it take to get good at Excel formulas?
For the everyday set — totals, conditions, lookups, cleaning up text — a couple of weeks at ten focused minutes a day is realistic, and that is the honest version rather than a marketing number. Short deliberate practice is what builds recall; one long weekend session mostly builds the illusion of it.
I only use Excel at work. Is that enough to get fluent?
Usually not, and it is the most common trap. Using Excel at work teaches you reactively — you learn a function the day a task forces you to, then never touch it again. The fix is short deliberate practice away from a deadline, which is why the exercises here take five minutes each and need no setup and no data of your own.