Somewhere in there is the answer to your question. How do you get it out, in a way you can stand behind?

Hand it to the black box. shortcut 1 “ChatGPT, what does this say?” Fast and fluent, but you cannot see what it leaned on, so you cannot defend the answer.

Read it all yourself. shortcut 2 Thorough, but it does not scale. Three hundred transcripts, and you still have to show your synthesis really reflects what people said.

There is a better way.

AI does the patient clerical work. You make the judgements.

This becomes training → confidence → started a business, keeping the exact quote and the source on every link.

Do that across all your interviews and the links join up into a map: your theory of change, built from what people actually said.

Maybe X really did influence Y maybe, maybe not.

Twenty people saying so is not proof. It is evidence you can now weigh. Turning that evidence into a conclusion is your job, and we come back to it later.

People say “X made Y happen”. They rarely say how strongly. So we do not invent it.

Coding what people actually said is simpler, and that is exactly why AI can do it.

The clerk’s job: find every causal claim, attach a quote. Tireless, exhaustive, cheap.

Your job: decide the question, check the work, judge what it all means.

It codes every claim across hundreds of documents, and each one traces back to the sentence it came from. So you capture everything cheaply and quickly.

Every later choice follows from that one sentence: the data you gather, the labels you use, the maps you make.

Pick questions the method can serve.

The golden rule: test on a small, varied sample, see exactly where the output is wrong or thin, fix the instruction, then scale up.

And one rule you never break: every link needs a quote. Without it you cannot show your working.

You ask a question by filtering. Women only. Everything downstream of training. Stack the filters and you answer a bigger question. The same data gives very different maps, each one just the result of a different question.

“What did the cash transfer lead to?” Filter to that factor, look downstream. Out comes a map of every reported consequence, with counts.

“Do women and men tell different stories?” Split the same map by group. The links each group stresses light up differently.

A wheat farmer says more cash let them buy more seed.

So cash grants lead to more seed?

No.

Two people, two stories, stitched into one that nobody told.

The AI can draft a source-by-source commentary on each pathway, doing only what a patient reader could. Treat it as a first draft and edit it.

No team could hand-code that and stay fresh. Every link still carries the quote it came from.

Rather than trade impressions, we put the question to the data and traced it through the 43 sources that raised it. Every account pointed the other way: the support helped communities resist backlash, none named it as the cause.

This is not statistical proof. It is a disciplined way to assemble evidence, weigh it transparently, and reach a conclusion you can stand behind.

If you want to go from a stack of interviews to a theory of change you can defend, come and try it with us at app.causalmap.app.