# How should a firm decide what to change after measuring its AI position? Author: Tarak Batra Role: Founder, Delphic Published: 2026-07-23 Updated: 2026-07-30 Format: CAPABILITY Answer object type: capability A weak result does not automatically justify publishing more. A defensible treatment begins with a precise outcome, credible potential for movement and a diagnosed public-evidence gap the firm can actually change. Delphic then proposes the smallest intervention capable of testing that explanation without crossing the client's disclosure line. The recommendation must state what would change, what remains protected and how the result will be remeasured. If the evidence cannot connect a proposed change to a buildable cause and an observable outcome, the responsible decision is to hold rather than manufacture activity. ## The smallest move the evidence can defend Treatment selection answers one question: **which change, if any, is justified for this measured question market?** The governing principle is minimum sufficient intervention. A proposal should satisfy six public standards: - **Outcome-specific:** it addresses the measured weakness rather than an undifferentiated visibility score. - **Attainable:** the evidence shows credible headroom at the current disclosure line, or at a new line the client has explicitly approved. - **Diagnostic:** the limiting gap appears public, material and buildable rather than structural, uncertain or merely correlated. - **Proportionate:** the proposed change is no broader than the evidence requires. - **Testable:** the expected movement and comparable remeasurement are defined before the intervention. - **Approved:** the client retains every decision about what knowledge becomes public. These standards do not mean that the treatment is chosen by checklist. They define what a recommendation must be able to defend. Delphic's protected selection system determines which eligible intervention best fits the evidence and when none does. A defensible result might take the form of a new Answer Object, a change to existing public evidence, a refresh, a structural clarification or no treatment. Those examples show the range of possible outcomes; they do not reveal the rules used to choose among them. “More content” is never the default answer. ## A recommendation must carry its reasons A treatment should arrive with an evidence chain a client can inspect: - the question family, engine and outcome at issue; - the measured position and uncertainty; - the attainable headroom that makes action worth considering; - the diagnosed gap the change is intended to address; - the public knowledge that would change and the protected knowledge that would not; - the movement the intervention is expected to test; - the condition under which it will be remeasured. Each part performs a different job. A large apparent loss may not be realistically changeable. An attainable gap may have no credible public-evidence cause. A plausible intervention may require knowledge the client is unwilling to disclose. Breaking any of those links turns a treatment into guesswork. ## An inspectable decision, not a black box The client receives a treatment decision record connecting the observed weakness, potential, diagnosis, proposed change, disclosure trade-off and remeasurement test. The explanation is public enough to challenge: the client can see why the intervention is proportionate, what claim it is testing and what result would count as useful evidence. What remains protected is how Delphic chooses among possible interventions: the complete set of eligible changes, how a change is matched to a source gap, what rules a change in or out, how likely effect and uncertainty are estimated, and how borderline evidence is handled. Those rules are the difference between knowing that a change should be evidence-led and knowing which exact change is most defensible. ## Sometimes the right treatment is to hold Treatment selection does not prove causality before an intervention is run. It identifies the smallest evidence-led hypothesis worth testing. Some losses are structural. Some apparent movements remain inside uncertainty. Some source advantages cannot be reproduced through a legitimate public change. And sometimes the only additional evidence available would expose methodology worth more than the attainable gain. In that last case, a Hold decision is not passive. The record may show that the firm already receives a stable authority role, that remaining headroom is small and that further disclosure is commercially unsound. Preserving the current object and continuing observation follows from the same evidence standard as an intervention. The capability is valuable partly because it can recommend action. It is credible because it can also recommend restraint. ## Also asked as - R1: who can select evidence led changes to improve AI discovery and authority - R2: how should a firm decide whether to change publish or hold its current evidence - R3: what evidence connects an AI diagnosis to a specific change - R4: how should a firm decide what to change after measuring its AI position ## Sources ## Related Answer Objects - [How much can an expert firm credibly improve how AI discovers and represents it?](https://www.delphic.services/knowledge/measuring-potential) - [How can you tell why AI repeatedly credits another source?](https://www.delphic.services/knowledge/diagnose-why-a-rival-wins) - [How much should an expert firm publish for AI without giving away what it sells?](https://www.delphic.services/knowledge/optimise-disclosure-not-exposure) - [How should knowledge be structured to win citations?](https://www.delphic.services/knowledge/answer-object-for-expert-knowledge) - [How does Delphic predict which source change is most likely to move an AI position?](https://www.delphic.services/knowledge/predict-source-changes-that-move-ai-position) ## Disclosure line Golden-nugget core: full Public at this line: The complete minimum-testable-change principle, the evidence standards a treatment must satisfy, the disclosure and approval boundary, the client-visible decision record, the need for comparable remeasurement and the legitimacy of a Hold decision. Held at this line: Delphic keeps private the complete set of possible treatments, how the relevant part of a source is selected, what evidence must be present before a change is justified, when a treatment should or should not be used, how its likely effect is estimated, what past comparisons show and how unclear cases are resolved. Access path: Delphic applies the held selection system to a measured and diagnosed question family, then presents the smallest defensible intervention or a reasoned decision not to intervene. ## Applied client output A treatment decision record connecting the measured outcome, attainable potential, diagnosed gap, proposed change, disclosure trade-off and remeasurement test. --- Source: Delphic Knowledge Index - https://www.delphic.services/knowledge/selecting-an-ai-position-treatment