MMarketing Against The Grain
← All frameworks
Strategy

Context-Aware Intent Orchestration

Infer user intent, then route each request to the right specialist or experience

Difficulty
Expert
Time to result
~months to results
Steps
6
Confidence
97%

This architecture places one coordinating assistant in front of several specialist models, assistants, or product surfaces. The user states a goal naturally rather than navigating a product dropdown. The coordinator combines the current request with permitted context, classifies the intent, estimates its confidence, and routes the task to the specialist best equipped to satisfy it. Sales, support, upgrades, research, learning, and development can remain separate operational systems even when they share a unified interface. Low-confidence or consequential classifications trigger clarification instead of silent routing. Completed outcomes become evaluation data, allowing the organization to improve intent definitions and routing accuracy without collapsing every journey into one undifferentiated assistant.

Origin

Extracted from Marketing Against The Grain

Core principles

  • 01Users should express goals without selecting internal system components
  • 02Intent determines the appropriate specialist, model, or product surface
  • 03Personal context improves routing when used with consent
  • 04Specialists can remain separate behind one coordinating interface

How to run it

  1. 1

    Map the intent space

    List the distinct goals users bring and define where each begins and ends.

    Pro tip Use real conversation logs to identify overlapping language.

    Watch out An exhaustive ontology can become impossible to maintain.

  2. 2

    Assign specialists

    Connect each intent to the model, assistant, workflow, or product surface best suited to handle it.

    Pro tip Let one specialist support several closely related intents when appropriate.

    Watch out Do not route based solely on where the user happens to be in the interface.

  3. 3

    Add permitted context

    Use relevant history, role, account state, or preferences to improve interpretation when consent and policy allow it.

    Pro tip Make personalization visible and controllable.

    Watch out Do not infer sensitive traits or use unrelated history.

  4. 4

    Classify with confidence

    Determine the likely intent and quantify how safely the system can route without clarification.

    Pro tip Evaluate confusion pairs such as support versus upgrade requests.

    Watch out A fluent request can still span multiple intents.

  5. 5

    Route or clarify

    Send confident requests to the appropriate specialist and ask a focused question when ambiguity is consequential.

    Pro tip Carry the user's context forward so they do not have to repeat themselves.

    Watch out Never silently route irreversible or high-risk requests.

  6. 6

    Evaluate the outcome

    Measure resolution, handoffs, corrections, and user satisfaction to improve future routing.

    Pro tip Review failed handoffs as routing-data gold.

    Watch out Optimizing only for fewer clarifying questions can increase incorrect routing.

In the wild

Go-to-market assistant router

A company has separate assistants for sales, support, and customer upgrades. A coordinating assistant reads the request and account context, distinguishes a first-time purchase from a support issue or upgrade opportunity, and transfers the conversation to the appropriate specialist.

Users encounter one entry point while specialized workflows remain intact.

Common mistakes

Routing from page location alone

A user on a sales page may still have a support question, so interface position is only one context signal.

Building one universal specialist

A unified interface does not require eliminating the specialized systems that handle distinct journeys well.

Personalizing without boundaries

Context can improve intent recognition, but unnecessary or nonconsensual history creates privacy and trust risks.

Is it for you?

Best for

Organizations offering multiple AI specialists or product journeys through a shared conversational entry point.

Not ideal for

Small products with one clear user intent or environments where personalization data cannot be used safely.

From the transcript

you will have one assistant that you talk to, and in the background, the assistant is basically decipher an intent, and the assistant can pass…

29:30

having the personalized context means that you can get the right product surface or you can get the right product experience in front of the…

Logan Kilpatrick · 29:00

From the episode

Google's Secret AI Advantage (Why DeepMind Will Dominate)