T22

Most GTM monitoring systems are designed to detect coefficient drift (a conversion rate fell by 15%) but not topology change (a new channel replaced an existing channel as the primary source of category awareness). Topology changes are rarer but more dangerous because they invalidate the model structure, not just the parameters. The claim is contested by practitioners who argue that 'topology change' is just a very large coefficient shift — the distinction is one of degree, not kind. The counter-argument: topology change requires a different response (model rebuild) vs. coefficient drift (parameter update).

Coefficient drift: the same GTM graph structure holds, but the parameters (conversion rates, channel CPMs, payback periods) have changed. Response: update parameter estimates, reoptimize within the existing model. Topology change: the graph structure itself has changed — a new node or edge appears, or an existing one disappears. Examples: (a) the rise of intent data (T12) added a new pre-funnel node B_r to the GTM graph that did not exist in 2010; (b) the collapse of third-party cookies restructured the paid digital graph; (c) AI-generated content saturation (T23) is potentially rewriting the 'organic search' node from 'high-PMF content wins' to 'brand authority wins' — a structural change, not a parameter shift; (d) category creation (T26) represents a topology change where the 'competitive evaluation' node is temporarily absent. The GTM World Model's versioning (v1.0 → v3.1) is itself evidence of topology change: each major version added structural nodes, not just updated parameters.

• Third-party cookie deprecation: companies that modeled this as a 'reduce retargeting efficiency' (coefficient drift) were wrong — it was a topology change that required rebuilding the identity resolution layer from scratch
• LinkedIn overtaking email as primary B2B outreach channel (2018-2022): companies that treated this as a parameter shift in their channel mix model lost years of compounding efficiency vs. those that rebuilt around LinkedIn as a primary node
• AI content generation: SEO research in 2024 shows that Google's Helpful Content updates restructured the 'content quality' node from 'length + keyword density' to 'author authority + original research' — a topology change that invalidated prior content strategies

• Distinguishing topology change from large coefficient drift in real time is extremely difficult — the diagnostic criteria are clear in retrospect but ambiguous at the time of change
• Declaring a topology change (and therefore justifying model rebuild) can be used as an excuse to avoid the harder work of fixing a fundamentally broken model
• Most GTM problems in practice are coefficient drift, not topology change — the claim's value is in creating vigilance for the rare but critical topology shift

Add a 'topology watchlist' to your quarterly GTM review: a list of structural assumptions your GTM model makes that would be invalidated by specific external events. Examples: 'this model assumes that organic search is a meaningful discovery channel — if AI-generated overviews reduce click-through by >50%, we need to rebuild the discovery layer.' For each item on the watchlist, set a quantitative tripwire that triggers a model audit rather than a parameter update. When a tripwire fires, run a structured review of whether the change is drift (update parameters) or topology (rebuild model) before taking action.