Inspired by a class discussion

Why Do Some GenAI Pilots Never Scale?

An exploratory scrollytelling inspired by our discussion of Markov Chains, organizational adoption, and Monte Carlo methods.

This is not a prediction tool. It is a visual thought experiment about what can happen after a successful GenAI pilot.

↓ scroll
The mystery

The pilot worked. Six months later, nothing changed.

The demonstration was successful. The model produced useful outputs. Leaders were interested. But the workflow was never redesigned, no team owned implementation, employees remained unsure, and the pilot slowly disappeared.

The question is not only whether GenAI works. The more interesting question is what happens between a successful experiment and everyday organizational use.
Markov in simple terms

Think in states, not in promises.

A Markov Chain gives us a simple way to think about movement. A system is in one state now, and several next states may be possible. It does not tell us why a transition happens by itself, but it helps us organize the possible pathways.

Current state

The organization has a successful pilot.

Pilot

The pilot is not the finish line. It is only the starting point for the next organizational decision.

Possible next states
AssimilationDepartment UseStalledAbandoned

The same technical result can lead to very different organizational futures.

History

Andrey A. Markov

Markov studied dependent sequences in the early twentieth century. His analysis of vowel and consonant patterns in Pushkin's Eugene Onegin became a well-known illustration of transitions between states. Here, that same intuition is used only as an exploratory lens for organizational adoption.

The missing middle

Organizations can acquire technology without assimilating it.

A useful idea from organizational adoption literature is that adoption happens in stages. Buying, building, or piloting a technology does not mean it has been assimilated into work or integrated into the organization.

InitiationThe organization identifies a problem or opportunity.
AcquisitionThe organization selects or builds the technology.
PilotThe idea is tested in a limited context.
AssimilationTeams learn, adapt processes, and begin regular use.
IntegrationThe technology becomes part of workflows, governance, roles, and measures.
A possible answer Many GenAI pilots stop after acquisition and experimentation. They never achieve assimilation and integration.
Explore the next state

What happens after the pilot?

Choose how the organization responds. Each choice reveals a different pathway. This is a thought experiment, not a forecast.

Pathway A

Technical success, organizational stall

Pilot→Stalled→Abandoned

Why it happens

The organization treats the pilot as a model-development project. It does not clarify who owns implementation, how work should change, or how employees will be supported.

  • Adoption thresholds stay high.
  • Knowledge remains segregated inside the pilot team.
  • Diffusion and cross-team connectivity remain weak.
  • Governance and accountability remain unclear.
Lesson: A technically successful pilot can still fail to become an organizational capability.
Pathway B

Local adoption, limited spread

Pilot→Department Use→Isolated Capability

Why it happens

One department sees value and continues using the tool, but knowledge, workflows, and support do not move across organizational boundaries.

  • Diffusion happens locally.
  • Expertise becomes segregated.
  • Percolation across teams never occurs.
  • No organizational tipping point is reached.
Lesson: Local adoption is progress, but it is not yet enterprise integration.
Pathway C

The pilot becomes part of work

Pilot→Assimilation→Integration

Why it happens

The organization treats adoption as a work-design challenge. It establishes ownership, trains users, clarifies governance, connects teams, and embeds the tool into operating routines.

  • Thresholds fall as confidence grows.
  • Knowledge diffuses through champions and shared practices.
  • Segregated capability begins to connect across teams.
  • Percolation creates conditions for a tipping point.
Lesson: Scaling happens when the organization changes—not simply when the model performs well.
Why the pathways differ

Five supporting models help explain the transition.

Markov helps us describe the possible states. These supporting ideas help explain why an organization moves toward assimilation and integration—or remains stalled and isolated.

Threshold Theory

When are people ready to adopt?

Employees and teams may wait until they see enough usefulness, trust, support, social proof, or managerial encouragement. If thresholds remain high, the pilot stays inside the original team.

Effect on the pathway: High thresholds make Pilot → Assimilation less likely.
Segregation Theory

Why does capability become trapped?

AI knowledge may cluster among specialists, innovators, or one business unit. The pilot appears successful, but the wider organization remains separated from the capability.

Effect on the pathway: Segregation encourages Pilot → Isolated Capability.
Diffusion

How does knowledge spread?

Prompts, workflows, examples, confidence, and lessons must move through formal and informal networks. Weak diffusion means the pilot remains a demonstration rather than a shared practice.

Effect on the pathway: Strong diffusion supports Pilot → Assimilation.
Percolation

Do the adoption clusters connect?

Local pockets of GenAI use do not become enterprise capability unless enough links exist across teams, functions, workflows, and governance structures.

Effect on the pathway: Weak connectivity blocks Assimilation → Integration.
Tipping Point

When does use become self-reinforcing?

A tipping point may occur when enough people, teams, use cases, and support structures are connected. After that point, adoption is no longer driven only by the pilot team; it begins to feel like an organizational norm.

Effect on the pathway: Connected adoption can move the organization from scattered use toward integration.
How the pieces fit together Markov describes the states and possible pathways. Thresholds, segregation, diffusion, percolation, and tipping points explain why one pathway becomes more plausible than another.
Where Monte Carlo fits

One pathway is only one possible story.

Monte Carlo introduces the idea of exploring uncertainty through repeated hypothetical sampling. It does not provide evidence by itself, and it should not create the impression that imagined trials are real organizations.

What it contributes

Thinking beyond one pathway

A single pathway shows what could happen. Monte Carlo reminds us that uncertain systems may follow many possible paths, depending on the assumptions built into the model.

Why there is no simulation here

The assumptions matter more than the number of trials.

Without observed or defensible transition probabilities, repeated sampling would only reproduce our assumptions. This page therefore focuses on understanding the possible organizational pathways first.

History

Nicholas Metropolis and Stanislaw Ulam

Their 1949 paper, The Monte Carlo Method, described the use of random sampling for difficult mathematical and physical problems. Here, it serves only as inspiration for thinking about multiple possible futures—not as a forecasting claim.

Inspired by these works

Three ideas behind the story

These works are shown as inspiration, not as proof that this exact GenAI model already exists.

Technology transitions

Markov Models of Policy Support for Technology Transitions

Max T. Brozynski and Benjamin D. Leibowicz

This work inspired the idea of representing technology change as movement through uncertain states. Our page applies that intuition lightly to the journey after a GenAI pilot.

Organizational adoption

A Model for the Adoption Process of Information System Security Innovations in Organisations: A Theoretical Perspective

Mumtaz Abdul Hameed and Nalin Asanka Gamagedara Arachchilage

This work inspired the distinction between acquiring an innovation and actually assimilating and integrating it into organizational use.

Uncertainty and repeated sampling

The Monte Carlo Method

Nicholas Metropolis and Stanislaw Ulam

This work inspired the idea that uncertain processes can be explored through repeated hypothetical trials. In this page, that remains an explanatory concept rather than a numerical forecast.

Answer

Why do some GenAI pilots never scale?

Because proving that GenAI works is only one step. Scaling requires assimilation and integration: people must understand and trust the tool, workflows must change, ownership and governance must become clear, and knowledge must move beyond the pilot team. The pilot does not become capability until the organization changes with it.