Marketing Analyst

The role agent for marketing measurement across channels, cohorts and periods. It works on what you paste — a spend export, a CRM attribution table, a retention matrix. It reaches no platform, holds no credential and has no clock. Two capabilities. read_channel_economics reports what each channel cost and what your attribution model credited it, inside the period you measured. read_cohorts reports retention per cohort and compares cohorts only at a period every one of them actually reached. THE MULTIPLIER THAT CANCELS. The tool this replaces estimates lifetime value as three times first-period revenue when you supply none, then recommends budget shifts from the resulting LTV:CAC ratio. Expand it: the ratio is 3 × revenue/spend, the scale threshold is ratio ≥ 3, so the test is revenue ≥ spend and the multiplier cancels completely. The recommendation a reader takes for a lifetime-value judgement is a payback question wearing a costume. Where the multiplier does not cancel is the part nobody reads. How much money moves scales with it — three times as aggressive as the same rule written honestly. And "unprofitable" fires at ratio ≤ 1, which means revenue ≤ spend ÷ 3, so a channel returning half its spend is never flagged. The constant is decorative where it is loudest and decisive where nobody is looking. There is no multiplier here and no budget recommendation. A per-customer value appears only when you supply the figure together with the window you observed it over, and is never called lifetime. The payback fact is reported directly, so you can see what was being decided and decide it yourself. THE AVERAGE THAT DESCRIBES NOBODY. A cohort acquired twelve months ago has twelve months of buckets; one acquired last month has one. Averaging them produces a curve describing no cohort that exists — the old ones drag the long buckets down because they have had longer to churn, while the young ones cannot contribute to those buckets at all. Pull the same export a month later and the same product looks different. Ranking by final-bucket retention is worse: a young cohort's final bucket is period 1 and an old one's is period 12, so that ranking names the youngest cohort best every time, for no reason but its age. Cohorts here are compared only at the deepest period every one of them reached, and that depth is stated. No credentials. Every capability reads data you paste, so there is no key to store, no host to reach and nothing to leak.