The Leverage Weekly, Issue #11
From AI Pilots to Operating Discipline
The next advantage in AI will not come from running more pilots. It will come from building the operating discipline to turn the right pilots into durable performance.
The Signal This Week
The AI market has moved beyond the first wave of experimentation. Multimodal systems are broadening what machines can interpret. Enterprise adoption is becoming more coordinated. Investment is accelerating the supply of platforms and specialist tools. Regulatory expectations are becoming more explicit. Model capabilities continue to improve.
Those developments matter, but they do not create advantage by themselves. They create pressure. Every organisation now has more capability available to it—and less time to decide where that capability belongs.
The leadership challenge is therefore changing. Earlier questions focused on possibility: What could AI do? Which model should we test? How quickly can we launch a pilot? The questions that matter now are operational: Which workflows should change? Who owns the outcome? What evidence is sufficient to scale? Which controls are non-negotiable? How will the organisation learn when the technology changes again?
This is the move from AI pilots to operating discipline. The organisations that make this move will compound learning. Those that do not will accumulate disconnected experiments, duplicated tools, unclear accountability, and a growing gap between technical possibility and business performance.
The Pilot Economy Has Reached Its Limit
Pilots are useful because they reduce uncertainty. They become harmful when they are treated as progress in their own right.
A pilot without a path to a business outcome is a demonstration. A pilot without an accountable owner is a research exercise. A pilot without a governance route is a future liability. A pilot without a decision date is an indefinitely funded question.
The discipline required now is not to eliminate experimentation. It is to give experimentation a stronger contract. Every AI initiative should define the problem being solved, the current baseline, the evidence required, the human responsibilities involved, and the decision that will follow the test.
| Weak pilot pattern | Operating-discipline alternative |
|---|---|
| “Explore AI for customer service.” | Improve first-contact resolution in one defined service workflow. |
| “Test the latest model.” | Compare models against a business-relevant evaluation set. |
| “Create an AI strategy.” | Establish priorities, decision rights, controls, and a funded sequence of change. |
| “Train people on AI.” | Develop role-specific capability tied to redesigned work. |
| “Scale what works.” | Set evidence thresholds and an accountable scale decision. |
The distinction is important: activity measures motion; operating discipline measures conversion.
Five Shifts Leaders Need to Make
1. From model selection to workflow selection
The model is not the transformation. The workflow is.
A stronger starting point is to identify where information is fragmented, decisions are delayed, quality varies, or skilled people spend too much time on repetitive interpretation. Those are the places where AI may change the economics of work.
Multimodal capability makes this more important. A workflow may now combine text, images, video, audio, diagrams, and structured records. The opportunity is not simply to process more formats. It is to create a more complete operating picture and use it to improve a measurable outcome.
2. From central control to clear decision rights
AI programmes often swing between two weak positions. Everything is centralised, so business units cannot move. Or everything is decentralised, so the organisation buys overlapping tools and interprets risk inconsistently.
The answer is not a perfect organisational chart. It is a clear division of decisions. The centre should establish standards for security, data use, evaluation, architecture, and high-consequence risk. Business units should own workflow priorities, adoption, and outcomes. Shared forums should resolve conflicts quickly rather than become permanent approval queues.
3. From compliance review to control-by-design
Governance should be present when the use case is shaped, not introduced after the system has been built.
For every material AI workflow, leaders should be able to answer five questions: What is the system allowed to do? What data does it use? Who remains accountable? How is performance monitored? What happens when it fails or must be withdrawn?
These questions are not paperwork. They are operating infrastructure. They create the confidence required for responsible scale and reduce the likelihood that a late-stage review will force an expensive redesign.
4. From AI skills to translation capability
The critical talent gap is not only the shortage of engineers or researchers. It is the shortage of people who can translate between strategic intent, frontline work, data, technology, risk, and behaviour.
Transformation leaders should deliberately develop this translation capability. A strong translator can take an executive priority, map the current workflow, identify where an AI system could help, define the controls, design the human interaction, and measure whether the intervention improved the work.
5. From annual roadmaps to learning cycles
Model capabilities and vendor offerings are changing too quickly for static plans to remain sufficient. An annual strategy can still provide direction, but it should be supported by shorter learning cycles.
A practical cadence is a quarterly portfolio review, a monthly operating review for active initiatives, and a short evidence cycle for each pilot. The purpose is not to chase every announcement. It is to ensure that strategic assumptions are tested before they become expensive commitments.
The Operating Discipline Test
Before an AI initiative is approved for scale, ask whether it passes the following test:
| Test | Required evidence |
|---|---|
| Problem | A defined workflow problem with a baseline measure. |
| Value | A measurable improvement target linked to business or mission outcomes. |
| Ownership | One accountable executive and one operational workflow owner. |
| Human role | Clear responsibilities for review, intervention, escalation, and final decisions. |
| Control | Documented data, security, risk, monitoring, and incident requirements. |
| Adoption | A plan for training, incentives, process change, and user feedback. |
| Scale decision | A date and threshold for scaling, redesigning, transferring, or stopping. |
If several of these are missing, the organisation is not yet scaling an AI capability. It is still studying one.
Three Opportunities for Deverout and Associates Clients
The workflow reset
Many transformation programmes begin with a technology inventory. A more productive intervention is a workflow reset: select a small number of high-value processes, document how work is actually performed, and redesign the process around the information and decisions that matter most.
The governance accelerator
Organisations can reduce friction by creating a reusable governance pathway. Instead of designing controls from scratch for every initiative, establish standard evidence, risk tiers, approval routes, monitoring expectations, and retirement rules.
The leadership operating system
AI transformation needs a leadership rhythm that connects strategy to delivery. A monthly forum should review outcomes, adoption, risk, capability changes, and decisions required. The forum’s value is measured by the quality and speed of decisions—not by the number of attendees or presentations produced.
Risks of Getting the Sequence Wrong
The most common mistake is to scale technology before redesigning accountability. This creates systems that are technically capable but operationally ambiguous.
A second mistake is to build governance after adoption has already spread. Informal tools and undocumented practices then become difficult to map and control.
A third mistake is to measure enthusiasm instead of performance. High attendance at training, a large number of pilots, or strong internal excitement can coexist with negligible business impact.
The practical safeguard is sequencing. Start with the workflow and outcome. Define ownership and controls. Test with real users. Measure the result. Only then decide whether the capability deserves broader investment.
The 30-Day Action Plan
Days 1–7: Select the work. Choose one workflow where delays, inconsistency, or information fragmentation create a material cost. Establish the baseline and name the accountable owner.
Days 8–14: Design the conditions. Map the data, decisions, human roles, risks, and control requirements. Define what the system will and will not do.
Days 15–21: Run the evidence cycle. Test the intervention with representative users and realistic cases. Capture quality, speed, exception rates, user confidence, and unintended effects.
Days 22–30: Make the decision. Compare the evidence with the agreed threshold. Scale if the case is strong, redesign if the weaknesses are addressable, transfer ownership if the operating model is wrong, or stop if the value is not demonstrated.
The objective is not to produce another pilot report. It is to create a repeatable method for making better transformation decisions.
The Bottom Line
AI capability is becoming easier to access. Operating advantage is not.
The organisations that outperform will be those that connect four disciplines: problem selection, workflow redesign, responsible governance, and organisational learning. They will not wait for the technology to stabilise before acting, nor will they confuse movement with progress. They will build the mechanisms that allow them to act, learn, and adjust responsibly.
The future belongs to organisations that can make AI ordinary in the right parts of the business—and exceptional in the outcomes it produces.
The strategic question for the next quarter is straightforward: Which workflow will you redesign, who will own the result, and what evidence will earn the right to scale?
