Analysis: from twelve sources to a conclusion anyone can check
Three interviews, an eighty-page report and a spreadsheet become a note with a path from every claim back to its evidence.
Anna is the analyst in a company of sixty people. On Monday the board will decide whether to open a pilot in a second city this year, and by Friday she has to write the note they will decide on. Her material: three customer interviews she ran herself, an eighty-page market report from a vendor, a comparison spreadsheet from sales, and two emails from two directors who have already made up their minds in opposite directions.
She has written notes like this before. She knows the risk. The conclusion comes out confident, somebody asks “where does this number come from”, and she spends twenty minutes digging through files while the room waits.
She starts from Monday, not from the report
The first thing Anna writes is not a finding. It is the question the board will actually be answering and the three things that would change their mind. That takes half an hour and saves her two days, because now every source gets read against those three things instead of against her curiosity.
The eighty-page report goes in as it is. She does not summarise it. She writes one line under it: vendor, 2025, optimistic by design. When she later searches for the market size figure, the search finds page 41 inside the PDF, and the number has its source attached forever.
Two directors, two branches
Anna does not file the two emails as noise: they are two readings of the same data, and she gives each one a branch: demand is real, demand is an artefact of how sales counted leads. Each branch argues only from what is in it. By Wednesday evening she can see which findings survive in both readings and which live in only one.
The ones that survive both go into the note as facts. The ones that do not go in as assumptions, each with a sentence about what would confirm or kill it. The conclusion is shorter than usual and much harder to argue with.
Monday, the question gets asked
Somebody does ask where the number comes from. Anna does not open a file. The director has the notebook in the view role, clicks the paragraph, lands on the report note, and reads the line about the vendor being optimistic by design. The discussion moves on to whether that matters, which is the right discussion.
The board approves a smaller pilot than either director wanted. The open questions stay in the notebook. In March, when the pilot has three months of data, the next note starts from them.
Someone has to decide, and you have to write the note they will decide on. The material is scattered: interview notes in one file, a market report as PDF, a comparison spreadsheet, two conflicting opinions in email. Below is how an analysis notebook is laid out in Kartoteq and how to get from a pile of sources to a conclusion that survives a sceptical reader.
How the notebook is laid out
Start from the decision, not from the data
In the goal write which decision the note has to support: “Should we open a pilot in a second city this year?” The interview then asks the useful things: who decides, by when, what would change their mind, what counts as enough evidence. Write the criteria down in the first section while they are fresh: demand, cost of entry, risk of losing the current team’s focus.
AI proposes sections. For analysis the usual set works: sources, comparison, conclusion and open questions. Add hypotheses as a note at the top: three sentences, each of which can turn out to be wrong.
This first section is what you re-read before writing the conclusion. If the conclusion does not answer the question written there, the note is not finished.
One source, one note, one line about trust
Put every source into the Sources section as a separate note: the PDF report, the interview notes, the spreadsheet, the two emails as pasted fragments. Under each, write one line: who made it, when, and how much you trust it. “Vendor report, 2025, optimistic by design” is more useful than a tag.
The text of the PDF and the DOCX files is searchable. When you need the page where the report quotes market size, search for the number, not for the file name. Interview notes get a quote block for what the person actually said and a paragraph below for what you make of it; the two must not blend.
Then ask AI across the section: what do the sources agree on, where do they contradict each other, what does nobody cover. The answer cites specific notes, and you can see at once if a source was left out of the analysis.
Keep rival explanations in separate branches
When the data allows two readings, do not pick one in your head. Open two branches in the Comparison section: “demand is real” and “demand is an artefact of the survey”. Each gets its own discussion and its own sources, and AI argues each one only from what is in that branch.
Compare by the criteria from the first section. Mark with “Keep” the findings that hold in both readings: those go into the conclusion as facts. What holds in one branch only goes in as an assumption, with a sentence on what would confirm or kill it.
The consolidated note shows next to each paragraph which branch and which source it rests on. If two kept fragments contradict each other, AI points at the pair instead of writing an average.
Give the reader the trail, not just the verdict
The result document is the note itself. Write the conclusion in three paragraphs: what we recommend, what it rests on, what we do not know. AI proposes edits with a source for each; accept those that make a claim more precise, reject those that make it sound more certain than the data.
Share the notebook with the decision-maker in the view role. They can open any paragraph, follow it to the source note and read the interview quote themselves. Reviewers leave comments right on the paragraph they doubt; you answer there, and the thread stays with the note.
Open questions stay in their own section after the decision. When the pilot runs, the next analysis starts from them, in the same notebook, with the old sources still in place.
What is left when the project ends
A notebook you can open a year later and still understand: the sources, the discussions and the result, with the reason next to every decision. If the next project is similar, you start from here instead of from an empty page.
Where to go next
AI that reads only your notebookStart with one notebook.
The Personal plan is free: notes, files and search. AI and shared notebooks can be added later, when a project needs them.
Create account