Sign in

Blog · AI and self-service analytics

Association and cause: what the company's own data can say, and the sentence that says it

Why every effect measured from a company's own history, samples and prescriptions, onboarding milestones and renewals, first-touch latency and first-year revenue, delayed installs and non-renewal, is an association and not a cause, the selection that makes it so, the comparison group that removes most of the noise, the sentence that reports it honestly, associated with rather than caused, why it is still enough to direct effort, and the two claims that are never made from it.

The short answerA company's own data shows that accounts touched in the first two weeks earn more, that onboardings which connect an integration by day fourteen renew more, that sampled accounts prescribe more. Each is an association, because the accounts were not assigned at random: reps touch the promising ones, success managers push the engaged ones, reps sample the responsive ones. A comparison group of similar unassigned accounts removes most of the noise; it does not remove the selection. The honest sentence is associated with, stated on the line, and it is still enough to decide where the next hour goes, because the association is the best evidence the company has. Two claims are never made from it: that the effort caused the result, and that the effect size would hold if the effort were applied to everyone.

The company's own data says that accounts touched early do better. It does not say the early touch made them do better, because the reps chose which accounts to touch early. Every effect measured from observational data on this site is an association, and the wording says so. This guide sets out why, what the comparison group does, the honest sentence, and why it is still enough to act on.

The pattern

Effort Outcome Selection
First touch within 14 days Higher first-year revenue Reps touch the accounts they expect to buy
Integration connected by day 14 Higher first renewal Engaged customers connect integrations
Sample dropped Higher prescriptions Reps sample the accounts likely to respond
Contract offered in year one off warranty Higher attach Service sellers offer where they expect a yes
Delivery inside the window Steadier order frequency Routes are sequenced around the accounts that complain

Each is real, useful, and not a cause.

The comparison group

Effect = change at accounts that got the effort − change at similar accounts that did not, same period

Similar means the same tier, segment and access state. The comparison group carries the season, the market and the product change. What is left after subtracting it is associated with the effort. The choice of who got the effort remains.

The sentence

"Accounts that reached the milestone by day 14 renewed at 91 percent; those that did not, at 52 percent. The difference is associated with the milestone, not shown to be caused by it; accounts that reach it early may be the ones that were going to renew."

On the line. Every time.

Why it is still enough

Decision What the association supports
Where the success team's hours go The milestone with the largest separation
Which territories get samples The ones where the sampled accounts moved net of comparison
Whether to call about warranty expiry The attach difference, offered against not offered
Whether to re-sequence a route The order frequency drop after out-of-window deliveries

None of these requires proof of cause. Each requires the best evidence available, and the association is it.

Two claims never made

That the effort caused the result. The sentence says associated.

That the effect would hold for everyone. Applying the effort to the accounts that did not get it is applying it to the ones the reps chose not to touch. The effect on them is unknown, and the report says the association was measured on the accounts that got it.

Where it goes wrong

Reported as cause. "Samples increased prescriptions by 2.7."

No comparison group. The market credited to the effort.

Effect size applied to the whole base. A forecast built on the responsive accounts' response.

Association dismissed as nothing. The best evidence the company has, unused.

Every effect, the honest sentence

Covirage reports every measured effect with a comparison group and the words associated with, and never restates one as a cause. The pull-through guide and the time to value guide show the sentence in use on two desks.

Questions people ask

If it is only an association, why act on it?

Because the alternative is acting on nothing, or on a benchmark from someone else's company. An association from the company's own accounts, net of a comparison group, is the strongest evidence available short of an experiment, and it says which milestone, which touch and which territory to spend the hours on. Acting on it is right; claiming it is a cause is not.

What does the comparison group do?

It carries the other reasons the outcome moved: the season, the market, a product change. Similar accounts that did not get the effort in the same weeks moved for those reasons too, and subtracting their change leaves what can be associated with the effort. It removes noise; it does not remove the choice of who got the effort.

Can a company ever measure cause from its own data?

With a holdout: a random set of accounts deliberately not given the effort for a period. Most companies will not run one on their sales effort, and the report does not pretend they did. Where a holdout exists, the report says so, and the sentence changes.