Top-Down and Bottom-Up AI Adoption
Combine strategic priorities with employee-led experimentation to embed AI.
- Difficulty
- Advanced
- Time to result
- ~months to results
- Steps
- 6
- Confidence
- 97%
Top-Down and Bottom-Up AI Adoption treats transformation as both a strategic and cultural challenge. The top-down side identifies major business opportunities, establishes boundaries, and directs resources toward a limited set of important initiatives. The bottom-up side allows employees who understand daily workflows to experiment, automate tasks, and propose applications leadership may not anticipate. Hackathons provide a structured mechanism for this discovery by giving teams time and permission to build. Promising workflows are then evaluated, standardized, and shared so experimentation becomes organizational capability. The episode contrasts HubSpot's centralized use-case prioritization with Zapier's company-wide hackathons and notes that nearly half of Zapier employees subsequently automated parts of their jobs. The result is coordinated adoption without suppressing local ingenuity.
Origin
Extracted from Marketing Against The Grain through contrasting HubSpot's prioritized initiatives with Zapier's employee hackathons.
Core principles
- 01Leadership must provide focus, resources, and strategic priorities.
- 02Employees closest to the work must discover practical applications.
- 03Hackathons and experiments turn AI from theory into organizational behavior.
- 04Successful adoption becomes part of culture rather than remaining a specialist project.
How to run it
- 1
Set Strategic Direction
Leadership identifies why AI matters, which outcomes deserve attention, and which boundaries teams must respect. This creates focus without prescribing every solution.
Pro tip Express direction in terms of business and customer outcomes.
Watch out A vague mandate to use AI can encourage performative experimentation.
- 2
Fund Central Priorities
Select and resource a small number of organization-level implementations. Assign ownership and measurement standards.
Pro tip Choose projects whose success would visibly demonstrate value.
Watch out Central projects alone may miss workflow-level opportunities.
- 3
Enable Local Experimentation
Provide employees with approved tools, training, time, and safe data practices. Invite them to improve the work they understand directly.
Pro tip Make experimentation accessible to nontechnical teams.
Watch out Do not expose sensitive data through uncontrolled tools.
- 4
Run Hackathons
Create bounded events where cross-functional teams build AI-assisted workflows or customer experiences. Require a working demonstration and a stated problem.
Pro tip Use real recurring work as the starting point.
Watch out A hackathon without follow-through can become innovation theater.
- 5
Evaluate and Scale
Assess submissions for value, safety, repeatability, and maintainability. Standardize successful patterns and make them available to other teams.
Pro tip Recognize the employees who produce reusable workflows.
Watch out Do not scale a prototype before checking reliability and governance.
- 6
Embed the Culture
Track adoption, share results, and make responsible AI experimentation part of normal improvement work. Continue feeding bottom-up discoveries into top-down planning.
Pro tip Create an accessible internal library of proven use cases.
Watch out Mandated usage targets can reward low-value automation.
In the wild
Zapier held hackathons where marketing and other teams could determine how AI should integrate into their own work. Employees built practical automations using Zapier, OpenAI, and other tools rather than waiting for one centralized implementation team.
→ After the hackathon, more than 45% of the company was automating parts of its work through Zapier and AI tools.
Common mistakes
Relying Only on Leadership
Executives can set priorities but may not see the recurring friction inside every team's daily work.
Relying Only on Grassroots Experiments
Uncoordinated experimentation can fragment tooling, duplicate effort, and create governance risks.
Ending with the Hackathon
Experiments create lasting value only when useful patterns are evaluated, operationalized, and shared.
Is it for you?
Best for
It is best for organizations that need both governance and broad employee participation in AI adoption.
Not ideal for
It is not ideal for highly restricted environments where unstructured experimentation with data or models would create unacceptable risk.
From the transcript
“I think AI needs to be somehow top down and bottoms up.”
“And I think bottoms up means you have to build it into the culture of your team.”
“At Zapier, what we did is we had hackathons again, where the teams could figure out how to like integrate AI into the work they…”
From the episode
How To Guarantee AI Won’t Replace You (#155)