The “funnel” marketing – this canonical scheme that guides customers from discovery to purchase through consideration – has structured the strategic thinking of marketing teams for decades. Linear, sequential, measurable, it was suitable for a world centered on mass diffusion, temporalized campaigns and marked customer routes.
But this model collapses silently as the conversational artificial intelligence takes over from the interface and that the customer no longer interacts with a platform, but with a system. Funnel is not suitable for dynamic, exploratory and personalized interfaces. It is no longer a tunnel that must be mapped, but a graph.
A linear model in a fluid world
Historically, the funnel made it possible to rationalize marketing in the advertising era: each step (Awareness, Interest, Consideration, Conversion, Retention) corresponded to an implicit intention, measured by behavioral signals (visit, click, time spent, purchase, re -desk).
But at a time when the main point of entry into an interface becomes a Conversation with an AI modelthis sequencing collapses:
- The user no longer “descends” a funnelhe navigate.
- He poses a complex question, jumps the steps, returns back, branches off.
- It can go from discovery to purchase in an instruction, or spend an hour testing, iterer, exploring, without triggering measurable action.
“Cat is the entry point, and everything else is a graph of intention behind him.” – Kevin Weil, Openai CPO
Towards a conversational cartography
The replacement of the funnel with a conversational graph is based on an observation: navigation is no longer controlled by the interface but by the uservia its natural formulations, its hesitations, its reformulations, its tests.
Consequently, the measure is no longer based on a conversion rate by step, but on theAnalysis of conversational knots and trajectories ::
| Traditional funnel | IA-STH GRAPH |
|---|---|
| Fixed and sequential steps | Dynamic and tree intentions |
| Conversion rate | Depth and buckle of interactions |
| Explicit call-to-action | Implicit or combined instruction |
| Pre-determinated course | Navigation generated in real time |
Example: a prospect in an IA-STIF tool (e.g. Banking assistant or B2B solution) does not click on “I want a quote” → He asks: “And if I choose the intermediate offer with 2 users, is the support included?” This sequence does not exist in a classic funnel, but it is a stage of strong intention.
What it changes for marketing teams
1. Tracking can no longer be satisfied with events
The useful signal becomes conversational : keywords, reformulations, tone, latency between instructions, etc. This imposes new layers of analysis (NLU, Intent Detection, Vector Embeddings).
2. The routes are no longer authorized with static rules
Nurturing or reactivation logics must be contextualized by the agents themselvesdepending on the signals detected, not a scenario defined in advance.
3. Interfaces must assume non-linearity
Good design does not seek to channel the user, but to help him Navigate freely in a base of responses, actions and offers.
4. The concept of “attribution” becomes blurred
To whom or what to attribute a conversion when the user spoke to an agent, read a page, tested an IA tool, explored 3 options in natural language, without ever clicking? The stake becomes relationalnot transactional.
Conclusion: Unlearn the funnel to learn the graph
Marketers who still think in funnel will have, as the formerly past developers from the imperative to asynchronous, unlearn a sequential logic To master the logics of tree, intention, and conversation.
This implies re -articulating the collaboration between UX Designers, Marketers, Data Scientists and Product Managers. And above all to accept that AI does not only structure the answer, but the very space of navigation.
