The AI Use-Case Funnel
Map what AI makes possible, rank use cases, and execute only the best few.
- Difficulty
- Moderate
- Time to result
- ~weeks to results
- Steps
- 6
- Confidence
- 99%
The AI Use-Case Funnel converts broad technological possibility into a small, executable portfolio. A discovery group first examines relevant language models, APIs, and integrations to understand what is genuinely possible. The organization then maps those capabilities onto its own workflows and customer problems, producing an intentionally broad inventory. Instead of attempting everything, the team evaluates use cases for potential value, feasibility, data availability, risk, and strategic fit. It selects only a few leading candidates, implements them, and measures operational and customer outcomes. HubSpot's example moved from 126 possibilities to three priorities, including a support-chat integration that answered a large share of questions while increasing customer satisfaction. The mechanism is divergence followed by disciplined convergence and evidence-based execution.
Origin
Extracted from Marketing Against The Grain through HubSpot's account of prioritizing 126 proposed AI applications.
Core principles
- 01Capability discovery should precede use-case selection.
- 02A long opportunity list is an input, not an implementation plan.
- 03Prioritization concentrates resources on the highest-value applications.
- 04Real customer and business outcomes determine whether an AI use case succeeds.
How to run it
- 1
Discover Capabilities
Review relevant models, APIs, products, and integrations to understand the tasks AI can currently perform. Document constraints as well as capabilities.
Pro tip Assign a small group to perform the initial technical and market scan.
Watch out Do not assume a demo will perform reliably with company data.
- 2
Inventory Use Cases
Ask teams where AI might remove friction, improve quality, accelerate work, or transform the customer experience. Capture the full list before narrowing it.
Pro tip Describe each use case as a problem and desired outcome, not merely a tool idea.
Watch out A large list can create false momentum if it is never prioritized.
- 3
Score the Candidates
Compare use cases using business value, user value, feasibility, data readiness, risk, and time to learning. Make trade-offs explicit.
Pro tip Include both customer-facing and internal applications.
Watch out Do not rank only by novelty or executive enthusiasm.
- 4
Select the Best Few
Choose a deliberately small portfolio of high-priority use cases. Allocate accountable owners and enough resources to test them properly.
Pro tip A top-three constraint forces useful decisions.
Watch out Attempting the entire inventory dilutes execution.
- 5
Implement the Workflow
Integrate the required models and systems into a usable end-to-end process. Include human review, fallback behavior, and data controls where necessary.
Pro tip Use existing systems when they shorten the path to a trustworthy pilot.
Watch out A disconnected prototype is not an operational implementation.
- 6
Measure Customer and Business Value
Track whether the implementation improves service, satisfaction, cost, speed, conversion, or another defined outcome. Expand only when evidence supports it.
Pro tip Pair an efficiency metric with a quality or customer metric.
Watch out Automation volume alone can hide a worse user experience.
In the wild
HubSpot surveyed models and APIs, generated 126 possible applications, prioritized three, and selected chat as one leading use case. It combined HubSpot Conversations, Google Dialog Flow, and OpenAI, fine-tuning the system to answer support questions.
→ The system answered 78% of support questions through AI while customer satisfaction rose almost 300%.
Common mistakes
Implementing the Entire Backlog
The inventory exists to expose options; trying to execute every idea defeats the prioritization mechanism.
Starting with Tools Instead of Problems
Tool-led projects can produce impressive demos that solve no consequential organizational need.
Measuring Automation Alone
An AI system should also preserve or improve customer quality, satisfaction, and business outcomes.
Is it for you?
Best for
It is best for teams beginning structured AI adoption across several functions or customer journeys.
Not ideal for
It is not ideal for a known urgent problem where one obvious, validated solution should be implemented immediately.
From the transcript
“The first thing that we did at HubSpot, I'll give you the behind the scenes is we collected all the use cases.”
“We came back with 126.”
“we took those use cases, we boiled it down, we found the top three.”
From the episode
How To Guarantee AI Won’t Replace You (#155)