From Spreadsheet to Answer: A Step-by-Step Walkthrough

This walkthrough follows one question from a raw file to a decision. The dataset is a marketing export with 2,400 rows: customer segment, email frequency, ad spend, prior purchase history, a seasonality measure, and revenue over the following 30 days.

The question: does moving customers from two emails a week to four actually increase revenue — and if so, for whom?

Step 1: Upload the file

Start a new analysis and upload the CSV. CausoAI reads it immediately: summary figures for every column, a check for missing values, and a guess at what each column represents.

For this file it works out that email frequency is the thing being changed, that 30-day revenue is the outcome, and that prior purchases, customer segment and seasonality are the context that needs holding constant. Customer ID gets set aside as an identifier rather than something to analyse.

You can override any of it. Here the guesses are right, so we continue.

Step 2: Check the diagram

CausoAI drafts a causal diagram from the data. This is the one step that needs you, so it is worth slowing down for. The draft here shows four things:

• —Prior purchases influences email frequency — engaged customers get put on heavier sends
• —Prior purchases also influences revenue, independently
• —Seasonality influences revenue
• —Email frequency influences revenue — the arrow we are trying to measure

Read each of those aloud and they hold up. Prior purchases matters most: it sits upstream of both the treatment and the outcome, which is exactly the shape that ruins naive comparisons. Because it is in the file, the analysis can hold it constant.

Step 3: See how far the result can be trusted

Once the diagram is saved, CausoAI scores whether the data can support the conclusion. This file comes back at 84 out of 100, which sits in the high-confidence band. The breakdown is more useful than the total:

• —Enough data: 22/25 — good size, little missing. Small penalty because 8% of customers spent exactly nothing
• —Question answerable: 25/25 — nothing important is being held constant that should not be, and the timing makes sense
• —Method can run: 18/20 — the split is close to even at 45/55, which is healthy
• —Groups comparable: 16/20 — fine across most of the base, thin for customers with ten or more prior purchases
• —Result robust: 3/10 — a moderately strong missing factor could soften the finding

That last line is the one to act on, and it points at the same gap we spotted in the diagram: acquisition channel. If that column exists somewhere, add it and re-run before treating this as final.

Step 4: Get the answer

Run the analysis. CausoAI picks the appropriate method for this shape of data on its own and takes under a minute. The headline result:

• —Moving from two to four emails a week is worth about $34.20 per customer per month
• —The honest range is $22.80 to $45.60 — the effect is real, but treat $34 as a midpoint, not a promise
• —That is roughly an 18% lift in 30-day revenue
• —Prior purchase history did most of the distorting work, which is why holding it constant mattered so much

For comparison, simply averaging the two groups without holding anything constant suggests a far larger effect. Almost all of that gap is prior purchases: the heavy-email group was already the group that buys.

Who it actually works on

The single average hides the useful part. Splitting by segment shows the effect is concentrated rather than spread:

• —Customers with one to three prior purchases: about $51 — this is where the whole effect lives
• —Customers with none: about $3, and the range crosses zero, so effectively nothing
• —Customers with four or more: about $28

That changes the action. Sending four emails a week to the whole list is wasteful, and mildly annoying to a group it does nothing for. Sending them to the one-to-three segment is where the money is.

Step 5: Try it before you do it

The simulator answers the follow-up question a marketing lead will always ask: what would it be worth if we actually did this? Set the change — move the one-to-three segment from two sends to four — and it projects the revenue impact across that group, with the same honest range attached.

This is also where you can test whether more is better. Pushing to six sends a week extrapolates beyond what the data has seen, and the simulator widens the range accordingly rather than quietly making something up.

Step 6: The written version

Finally CausoAI writes the analysis up: what was found, what was assumed, how confident to be, and what to do next. That last part matters most in practice, because it is the version you forward to someone who was not involved in running it.

For this dataset the recommendation is straightforward: increase frequency for the one-to-three purchase segment, leave the zero-purchase segment alone, and get acquisition channel into the file before committing budget on the back of it.