AI Leadership Frameworks: Applying FLOW to Your Technology Roadmap
Four moves that decide which AI work belongs on the roadmap, which belongs in a pilot, and which should end this quarter.
Most AI leadership frameworks describe maturity. FLOW decides. It is one of Jacqueline V. Twillie's four named frameworks, and it moves in four steps, from the facts, through the objections, to a written condition for walking away. Applied to a technology roadmap, it turns the recurring argument about which AI initiatives to fund into a sequence you can run in a single meeting, with a defensible answer at the end and a written record of what would change it.
The reason a decision framework beats a maturity model here is timing. Maturity models tell you where you are. Roadmap arguments are about what to do next while the evidence is incomplete, the vendor is persuasive, and someone senior has already said the word transformation out loud.
FLOW is a four-part decision framework: Facts over feelings, Listen to understand, Open to new ideas, Walkaway when necessary. It is built for contested decisions where the pressure to move is high and the evidence is partial.
Why do AI roadmap decisions get made on feelings?
Because the feelings are reasonable ones. Fear of falling behind a competitor, discomfort at telling a board there is no AI line item, genuine excitement after a demo that worked. None of that is irrational, and all of it is unusable as evidence.
Facts over feelings, applied to a roadmap, means every proposed initiative arrives with three things attached before it gets discussed: what it would replace, who currently does that work, and what result would prove it worked. An initiative that cannot answer the first two is still just a capability someone wants, waiting to become a workflow anyone runs.
The instrument matters here. BNEDai's AI Readiness Index scores eight dimensions of how a team already runs AI, and one of them is Truth in Every Draft, whose operating rule is never invent a credential, a number, or a relationship. That rule was written for AI output. It applies with equal force to the deck arguing for AI spend. A roadmap built on a vendor's projected efficiency figure has the same defect as an agent that invents a statistic, and it fails for the same reason. The capability audit in how to audit your team before buying AI agent training is the cheapest way to put real numbers under the first move.
"A projected efficiency figure is not a fact. It is a feeling with a decimal point."
What do you do with the strongest objection in the room?
The second move is the one most executives believe they already make. The test is whether you can state the strongest version of the objection in the objector's own words before you answer it.
On AI roadmaps, three objections recur, and each hides a different fact.
The operations lead who says the tool will not work on their process is often describing an exception rate. Ask what percentage of cases are unusual and what happens to those today. That number belongs on the roadmap.
Compliance rarely raises AI as the real issue. What they are actually asking about is access and reversibility: what the tool can reach and whether anything it does can be undone. Both are answerable, and answering them converts an opponent into a reviewer.
Then there is the objection nobody says out loud, from the person whose work is being automated. It is a question about credit and headcount, and it decides adoption anyway. That one is covered in handling cultural friction during a rapid AI rollout.
Each of those objections carries a fact the roadmap needs and does not have. Listening is how you get it out of the room and onto the page.
What does staying open to new ideas cost?
Open to new ideas is the move that keeps a roadmap from calcifying around the first architecture someone proposed. It has a specific application in AI work, because the shape of what is possible changes faster than an annual planning cycle.
Two practices carry it. Re-examine the build-versus-buy call on each initiative at every planning cycle rather than inheriting last year's answer. And let the people doing the work propose the initiatives. That is where most of the good ones come from anyway. A department manager who has watched the same reconciliation take four hours every Friday for two years knows more about what to automate than any consultant will learn in a discovery phase. That principle runs through reducing IT dependency for operational teams.
Openness has a boundary, and naming it is what keeps this move from becoming permanent indecision. Open to new ideas governs how the work gets done. It does not reopen the operating rules. The yes gate, scoped access, and the grounding rule are not subject to a better idea from a vendor with a faster demo.
When should you walk away from an AI initiative?
Walkaway when necessary is the hardest of the four and the one that makes the other three honest. A decision framework without a walkaway is a justification framework.
Set the walkaway condition before the pilot starts, in writing, while you are still capable of writing it. Three forms work on AI initiatives.
A result condition. The specific outcome that must appear by a named date, the thing the initiative existed to change. Adoption numbers and enthusiasm don't count.
A cost condition. The point at which the effort to keep it running exceeds the manual process it replaced. This is where the review burden lives, and review burden is the most underestimated cost in enterprise AI.
A rule condition. Any initiative that can only work by suspending an operating rule ends. If the business case requires an agent to send without a human yes, or requires an unscoped credential, the answer is no, and the answer does not improve with a better projection.
The walkaway is also what makes a pilot cheap. A pilot with a written end condition can be approved quickly. Its downside is bounded. Without one, a pilot turns into a permanent commitment nobody voted for.
"Set the condition that ends a pilot before the pilot starts."
How do the four moves run on one decision?
In order, in a single session, on one initiative at a time.
- 01
Facts.
What it replaces, who does that work now, what result proves it worked. Anything unanswered is written down as an open question, not assumed.
- 02
Objections.
State the strongest objection in the objector's words. Extract the fact inside it and add that to step one.
- 03
Open.
Ask whether the proposed approach is the only one, and whether the operating rules are being respected. Rules stay out of scope for revision.
- 04
Walkaway.
Write the result, cost, and rule conditions that end it, with dates and a named owner for each.
What comes out is short. One page per initiative, and a roadmap where every line has a stated reason for being there and a stated condition for leaving. That is a document a board can read and put in front of a skeptical operations lead who will argue with it productively.
FLOW is one of four named frameworks in this body of work. The others apply to adjacent problems: R4 for protecting core relationships, A.H.A. for the individual pressure of a tech shift, and L.A.T.T.E. for the negotiation itself. For concise definitions of each, see What Is the FLOW Framework? and What Is the L.A.T.T.E. Framework?.
Frequently asked questions
What is the FLOW framework?
FLOW is a four-part decision framework: facts over feelings, listen to understand, open to new ideas, and walk away when necessary. It is one of Jacqueline V. Twillie's four named frameworks and it is built for contested decisions made under time pressure with partial evidence. That describes most AI roadmap calls.
How is FLOW different from an AI maturity model?
A maturity model tells you where an organization currently sits. FLOW decides what to do next. Roadmap arguments happen while evidence is incomplete and a vendor is persuasive, so the useful tool is a decision sequence that produces a written outcome per initiative.
What counts as a fact on an AI roadmap?
Three things attached to every proposed initiative before discussion: what it would replace, who currently does that work, and what result would prove it worked. A vendor's projected efficiency figure is a guess dressed up as data. An initiative that cannot answer the first two questions is still just a capability someone wants, waiting to become a workflow anyone runs.
When should a company walk away from an AI initiative?
When a written condition set before the pilot is met. Three forms work: a result condition naming the outcome required by a date, a cost condition marking where upkeep exceeds the manual process it replaced, and a rule condition ending anything that can only work by suspending an operating rule such as the human yes gate or scoped access.
Does being open to new ideas mean the operating rules can change?
No. Openness governs how the work gets done, including build-versus-buy decisions and which initiatives get proposed in the first place. It does not reopen the operating rules: the yes gate, scoped access, and the grounding rule are not subject to a better idea from a vendor with a faster demo.
Build something that actually runs your workflow.
A focused, free 60-minute live session with Jacqueline. You build alongside her, on your own real task, and leave with an agent that is already running.