Most of your day is not analysis.

It is cleaning the export, reconciling two systems that disagree, and rebuilding the same deck. This course hands that part to a model and teaches you the checks that keep a confident, plausible, wrong number from reaching anyone.

See what's inside

3
Modules: 10
Practical lessons: Free
No card required: 1-to-1
A coach reviews your work

The thinking gets whatever time is left

The extract arrives with dates in three formats. Two systems report the same figure four percent apart and someone has to work out which one is lying. The monthly deck gets rebuilt from the same source it was built from last month. Then at four o'clock somebody asks whether you could just pull one quick number. The part you were actually hired for — deciding what the question is and what the answer means — gets whatever is left over.

The usual AI advice lands badly here. In the demo, someone pastes a spreadsheet in and a chart comes out. Your data is not that. A column called revenue_adj means something specific in your company, and nothing at all to a model that has never seen your finance team's conventions. The query it writes runs. It returns rows. The total looks about right. That is precisely the problem: an answer you cannot verify at a glance, delivered in the same confident tone as one you can.

This course is built around that gap. Where a model genuinely earns its place in an analysis, what it should never be trusted with, and the small set of checks that catch a wrong number before you put your name on it.

What you will be able to do

Turn a vague request into a question that actually has an answer
Draft a query or a formula from a plain-English description, then read it back and know whether it does what you asked
Understand a spreadsheet or query you inherited from someone who has left, in minutes rather than an afternoon
Clean and reshape a messy export without silently dropping or duplicating rows
Run the specific checks that catch a plausible wrong number: row counts, totals that must reconcile, a spot check you do by hand

What you'll cover

1
From spreadsheet to question
Most bad AI analysis has already failed before the model is opened — on a vague request and a dataset the model cannot see. This module fixes the framing, which is the cheapest place to fix anything.
2
Doing the work
Summarising, categorising, writing the query, drafting the narrative — and the verification pass that turns any of it into something you can put your name on.
3
Capstone
One real dataset from your own work, taken from question to written insight — with a verification trail, and reviewed by a coach.

This is for you if

You spend most of your week in a spreadsheet, a BI tool, or a query editor
You can write a basic formula or a simple SELECT, and you do not need to be able to do more than that
You are answerable for the numbers you hand over, and being wrong is expensive
You have pasted data into an assistant, got an answer back, and had no good way to decide whether to trust it
You are not a data engineer and have no plans to become one
You already write a decent prompt, or you have finished a prompting course here

Why this course is different

Built around being wrong

A model producing a wrong number is not like a model producing a wrong sentence. It is fluent, it is plausible, and nobody downstream can see it. So most of this course is the unglamorous half: the checks you run every time, and the two or three that would have caught the mistake.

Your data, your questions

The exercises run on an extract from your own work rather than a tidy sample set. An early training covers what to strip out before anything leaves your machine, so you are working on something real without pasting something you should not.

Spreadsheets and queries, not notebooks

Everything happens in the tools you already have open. Nothing to install and no pipeline to maintain — the one training that writes a per-row labelling pass accepts pseudocode, because what gets reviewed is the label set and the check, not whether it runs. If your team has standardised on a BI tool, the method works there too.

Questions people ask before joining

Do I need to know SQL?

Enough to read a simple SELECT is plenty, and if you only work in spreadsheets you can do almost all of it there. This is not a SQL course — one training reads a generated query, and reading it back is the skill, not writing it. It is about describing what you want precisely, then checking that what came back is what you asked for.

What does it cost?

$399, one payment, for the whole course including the coach-reviewed exercises. There is nothing else to buy inside it and no subscription attached.

Can I use my company's real data?

One training is about exactly that question. You will learn what should never be pasted into an unapproved tool and how to do the same analysis on a redacted or synthetic sample. Read your own organisation's policy first — this course cannot do that for you, and it does not override it.

What tools do I need?

An assistant you already have access to — ChatGPT, Claude, Copilot and Gemini all work — plus a spreadsheet, plus whatever query editor or BI tool you use at work. Nothing new to buy.

Will it do the analysis for me?

No, and any course promising that is selling you something. A model drafts, explains and reshapes. Deciding what the question is, whether the answer is right, and what it means for the business stays with you. That division of labour is the actual subject of the course.

How much time does it take?

Bring one question you have been putting off.

Pick the analysis that keeps getting pushed to next week because the data is a mess. By the end of the course it is answered, checked, and written up in a form somebody will read.