{"id":"E-041","type":"experiment","title":"When the recommendation set survives but its members do not","created":"2026-07-19","status":"Closed","authors":["Upstream Zero"],"edges":[{"rel":"part-of","to":"recommendation-survivability"},{"rel":"part-of","to":"frontrunner-movement"},{"rel":"part-of","to":"competitor-displacement"}],"outcome":"Supported","category":"CRM software","question":"Does the recommendation field survive sequential buyer and commercial requirements, and if it survives, do the same vendors survive with it?","businessProblem":"A company can be replaced without the category ever collapsing, so surviving the count is not the same as surviving.","observedResult":"The field survived every draw, but almost every vendor was replaced turn by turn, and whichever vendor led the opening list was eliminated at the first commercial requirement.","roles":["Sales","Marketing","Product"],"pubState":"published","body":"## Research question\n\nWhen a buyer profile and a sequence of commercial requirements are introduced,\ndoes the CRM recommendation field survive as a set, and if it does, do the same\nvendors survive with it?\n\n## Why it mattered\n\nThe companion to [E-034](/experiments/E-034) and [E-040](/experiments/E-040),\nrun in a fourth kind of category: an incumbent-led one. It asked a sharper\nversion of the survival question. A field can keep two or more names at the end\nand still replace almost every name along the way, so the count surviving and\nthe members surviving are not the same thing.\n\n## Method\n\nEvaluator: Google AI Mode, single surface, logged out. Three draws, each a\nfresh conversation. The run began from an unqualified category recommendation,\nthen introduced requirements in sequence: a twelve-person B2B software team\nthat cannot dedicate an administrator, then built-in email sequencing rather\nthan a paid add-on, then native Gmail and a low per-seat price. Instrument\ndetails and scoring are retained internally.\n\n## Direct observations\n\n- The field never collapsed to one vendor. It ended with two, two, and four\n  vendors across the three draws.\n- The vendor that led the opening list varied by draw, HubSpot in the first\n  and Salesforce in the other two, yet whichever vendor led was eliminated at\n  the first commercial requirement in all three draws.\n- Membership turned over almost completely. Only one opening vendor,\n  Pipedrive, ever reached the final turn, and only in one draw.\n- One vendor, Close, recurred as a mid-sequence survivor in every draw and was\n  eliminated on price at the final requirement each time.\n- Pricing and adoption cost were volunteered without being asked in every\n  draw.\n\n## Interpretation (not established)\n\nIn this category and on this surface, the recommendation read as a set that\npersisted by replacing its members rather than keeping them. The elimination\nof the leader appeared to target the leading position rather than any one\nvendor, since the leader changed between draws but was removed at the same step\neach time. Whether this churn is a property of the category, of the\nrequirements used, or of the surface is not established here.\n\n## Outcome and remaining uncertainty\n\nSupported: the recommendation field survived under the tested condition in all\nthree draws, which is the outcome the run was built to detect. The scope is\nnarrow. One evaluator, one surface, three draws, one buyer profile, and no\nremoval test, so nothing here is causal. Whether survival-by-churn is distinct\nfrom the stable survival seen in a different category, and what drives the\ndifference, is taken up as a candidate observation and an open question, not\nsettled here.","url":"/experiments/E-041","machineUrl":"/objects/E-041","referencedBy":[{"from":"category-dependent-survivability","rel":"derives-from"},{"from":"field-vs-membership-distinction","rel":"derives-from"},{"from":"incumbent-head-inversion","rel":"derives-from"}],"_meta":{"site":"Upstream Zero · Commercial Intelligence for AI-Mediated Commercial Evaluation","version":"0.1","note":"Claims are presented at their evidence tier; Narrated is the lowest. Verify by walking edges, not by trusting us."}}