Building the
AI Flywheel
Sawyer Middeleer
Revi Systems
Chief of Staff Connect · New York · October 23, 2026
Who are you right now?
Analysis paralysis
Bought AI tools, nothing changed
Everything is chaos
They still won’t let us use AI
AI? Ew.
Using AI intentionally, seeing results
The case against huge AI transformations
More moving parts
More data, tools, and handoffs create more dependencies and complexity.
More stakeholder challenges
More teams, competing priorities, and approvals to navigate.
Hard to measure impact
More people and routines need to change. It gets harder to tell which change helped.
Longer before feedback
A larger first release delays feedback about whether the work is useful.
Build a flywheel of minimum viable transformations
A minimum viable transformation is the smallest useful change to a real business process that moves the needle and earns you the right to deploy AI deeper in your organization.
Impactful and goal-aligned
Move a measurable KPI or survey result in support of an OKR or strategy.
Tightly scoped
Keep the change small enough to try, adopt, and learn from quickly.
Verifiable success
Define what good looks like, how you’ll check it, and where human judgment stays.
- Map and diagnose
- Design and build
- Run and learn
- Improve and scale
Reuse the context, tools, standards, and feedback in the next project.
One huge bet—or a series of compounding projects?
One huge transformation
Compounding MVTs
The ingredients of a great MVT
What you need
- Business goal
- Quantify the business outcome and gather evidence about the baseline.
- Scoped process
- Choose one impactful, goal-aligned process. Keep the change tightly scoped.
- Data and tools
- Map the data and tools available, and the output you need.
- Assigned owner
- Name an accountable owner. Involve users and builders; the CoS coordinates.
- Success criteria
- Define a verifiable result, quality bar, and where human review is needed.
How to build it
Map the process
Map today’s inputs, outputs, and decisions.
Diagnose the gaps
Find the context and process gaps.
Design and build the change
Connect context; write instructions; define ownership and quality checks.
Run, learn, and improve
Work with users, train the team, test in real use, and iterate.
Make validated changes permanent
Retain what works, assign ownership, and scale with evidence.
An example from my work
Goal: scale sales output without more admin
MVT 1: AI-assisted pipeline review
A pipeline review brings the team together to check live deals, unblock stalled opportunities, and agree on next actions.
Map and diagnose
- Trace deal updates across email, calls, Slack, and tasks.
- Find gaps in context and review preparation.
Design and build
- Centralize deal context in the CRM.
- Check and refresh data; draft the weekly agenda.
Run and learn
- I facilitated; sellers checked the facts.
- Commercial leaders prioritized needs and assigned actions.
Improve and scale
- Use feedback to fix missing context and unclear deal definitions.
- Improve capture, update rules, and quality checks.
From one MVT to a flywheel
Scale sales output without more admin
Do
Build and use a bounded transformation. Make validated changes permanent.
Learn
Capture feedback, fix process and data gaps, and decide what to improve or build next.
WORKSHOP: Design three MVTs for your team
Work alone or in a small group on one real business goal. Use the worksheet to design the first transformation and sketch what could follow.
Choose the goal
What should improve? What evidence describes today’s baseline?
Map the first process
How does the work happen now? Identify its inputs, outputs, decisions, and context gaps.
Define MVT 1
Scope the smallest useful change. Name the tools, owner, human review, and verifiable success criteria.
Sketch MVTs 2 and 3
What could you build next toward the same goal? What must you learn first, and what could you reuse?
Stay in touch
Sawyer Middeleer
linkedin.com/in/sawyer-middeleerWorkshop materials
Slides and worksheet
revi.systems · AI Flywheel materialsYour first three experiments
Use the two-page worksheet to choose a workflow, map three tests, and make the first one executable.