Top-Down Enablement, Bottom-Up Discovery
Centralize AI foundations while empowering frontline experimentation
- Difficulty
- Advanced
- Time to result
- ~months to results
- Steps
- 6
- Confidence
- 98%
This model separates the parts of AI transformation that require central control from those that benefit from distributed initiative. Leadership appoints one directly responsible individual, approves the data practices, selects shared tools and infrastructure, and chooses a few high-impact organizational priorities. Individual teams then experiment with how AI can improve their specialized work and surface opportunities leadership may not see. Regular reviews connect the two directions: teams report use cases and blockers, while leaders remove legal, security, procurement, or infrastructure constraints. Proven bottom-up discoveries can become top-down standards. The result is a managed blend rather than a binary choice. Centralization supplies safety, clarity, and reusable foundations; decentralization supplies domain expertise, speed, and unexpected breakthroughs.
Origin
Kipp Bodnar and Kieran Flanagan contrasted HubSpot’s top-down approach with Zapier’s bottom-up experimentation before converging on a blended model. Extracted from Marketing Against The Grain.
Core principles
- 01Shared data, tools, and infrastructure require central governance.
- 02Frontline specialists are best placed to discover role-specific opportunities.
- 03One directly responsible individual must coordinate the transformation.
- 04Leaders should remove blockers while teams generate experiments.
- 05Existing-work transformation and net-new discovery may require different directions of leadership.
How to run it
- 1
Name one owner
Assign one directly responsible individual to coordinate AI technology, use cases, priorities, and organizational learning.
Pro tip Give the owner an explicit charter and decision rights.
Watch out Co-ownership can turn a fashionable initiative into an unaccountable cluster.
- 2
Build shared foundations
Set centrally approved rules and systems for data access, security, tools, and infrastructure.
Pro tip Make the approved path easier than using unsanctioned tools.
Watch out Do not invite experimentation with sensitive data before defining safe handling.
- 3
Gather distributed use cases
Ask leaders and frontline practitioners where AI could improve current work or enable something new.
Pro tip Seek input from people who understand both their role and the technology.
Watch out A centrally invented list will miss practical opportunities in specialized workflows.
- 4
Prioritize centrally
Select a small number of initiatives with outsized organizational impact while leaving room for local experiments.
Pro tip Distinguish shared strategic projects from lightweight team-level tests.
Watch out Trying to execute every submitted use case destroys focus.
- 5
Run recurring transformation reviews
Meet with teams regularly to examine AI usage, discoveries, results, and blockers.
Pro tip Ask what approval or infrastructure would unlock the next experiment.
Watch out Reviews should remove friction, not become permission-heavy status meetings.
- 6
Scale proven discoveries
Convert successful local practices into shared tools, playbooks, or centrally supported initiatives.
Pro tip Let evidence determine what becomes standardized.
Watch out Do not standardize experiments before they demonstrate repeatable value.
In the wild
Leadership provides a private company-approved LLM interface and names one AI owner. Marketing teams use that foundation to test role-specific workflows, report results in monthly reviews, and request help with legal or technology blockers. The central owner then promotes the strongest experiments into shared programs.
→ Employees can move quickly without exposing company data or fragmenting the strategic portfolio.
Common mistakes
Giving AI to a committee
Multiple nominal owners diffuse accountability and make prioritization difficult.
Centralizing every tactic
Leadership cannot see all the opportunities available to specialists working at the front line.
Allowing tools without foundations
Unmanaged experimentation can create privacy, security, and procurement problems.
Is it for you?
Best for
It is best for medium or large teams that need security and strategic focus without suppressing frontline innovation.
Not ideal for
It is not ideal for a tiny team where formal governance would cost more than it saves.
From the transcript
“Not plural, it has to be one person.”
“there's like three categories of things that you need. There's the data, the tool and the infrastructure, and the tactics.”
“there's some blend of tops-down and bottoms-up that's gonna work for every team.”
From the episode
Use This A.I. Marketing Strategy To Grow Your Business In 2023 (#114)