Ask AI What It Can Do
Make AI inventory available data before choosing the highest-value use cases.
- Difficulty
- Starter
- Time to result
- ~days to results
- Steps
- 4
- Confidence
- 94%
Ask AI What It Can Do is a capability-discovery loop for unfamiliar integrations. Instead of guessing at prompts, the user first asks the AI to inspect the data it can access and explain what that data makes possible. The model can identify the business model, growth priorities, commonly used tools, target audience, and industry focus, then propose high-value applications. A second prompt narrows those possibilities for a specific role, such as a sales manager, and asks for the top use cases. The output becomes a map from available inputs to feasible workflows. The user then chooses a few valuable candidates, validates them against real needs, and builds targeted prompts. This method turns vague experimentation into a data-informed adoption process.
Origin
Extracted from Marketing Against The Grain after the hosts used deep research to inspect a demo HubSpot portal and generate ten growth-oriented prompt ideas.
Core principles
- 01Capability discovery should precede workflow design.
- 02Available data determines which use cases are feasible.
- 03Role and business priorities help rank candidate applications.
- 04The first research output can become the input to more focused prompts.
How to run it
- 1
Inventory Accessible Data
Ask the AI to inspect the connected system and describe the data classes it can use. Include records, activities, notes, outcomes, and contextual fields where available.
Pro tip Let the first pass be broad enough to reveal unexpected capabilities.
Watch out The inventory may take substantially longer than a normal chat response.
- 2
Infer the Business Context
Have the model identify the business model, growth priorities, target audience, industry focus, and commonly used tools from the accessible evidence.
Watch out Treat inferred context as a hypothesis and correct material errors.
- 3
Generate Role-Specific Use Cases
Name a role and ask which applications would be most valuable given the available data. Request a ranked shortlist rather than an unbounded brainstorm.
Pro tip Ask for the top three use cases for the person who will execute them.
- 4
Select and Expand
Choose the most valuable feasible use cases and ask the AI to create detailed prompts or workflows for each. Test them against representative records.
Pro tip Prefer applications tied to current growth priorities.
Watch out Do not automate every suggested use case before validating output quality.
In the wild
After the AI inventories HubSpot data, a sales manager asks it to identify the three highest-value analyses available. It proposes renewal outreach, upsell planning, lead scoring, deal-risk analysis, and other workflows grounded in the portal's records.
→ The manager receives a prioritized starting point instead of experimenting blindly with generic prompts.
Common mistakes
Starting With Random Prompts
Unstructured experimentation can miss the highest-value applications because the user does not know what evidence the integration exposes.
Accepting Inferred Context Uncritically
The AI may infer the business model or priorities incorrectly, so users should validate those assumptions before acting.
Is it for you?
Best for
Teams adopting a newly connected AI system and unsure which workflows to automate or improve first.
Not ideal for
Situations where the desired task and required data are already precisely defined.
From the transcript
“I actually love what you've done. Actually, do a deep research to figure out all of the data it has access to.”
“So I think that's a cool way for everyone in terms of how you would get started is like ask AI what it can do.…”
From the episode
AI Just Replaced a $100K Sales Analyst for $20/Month