Vector leadership is inherently straight forward. Strategy sets Direction and leaders make Decisions that fix what we will and will not do. Performance is Momentum, the rate and quality of progress toward that direction. Coherence is whether the whole company moves together. In short: Impact = Direction × Decision Quality × Momentum × Coherence. If any term is near zero, impact collapses. In practice, vector leadership means three to five value pools with guardrails and explicit financial and quality targets, that everyone can see, recited often enough to be almost tedious. Here, tedious is good. Tedious aligns.

A private wish list is not a vector. Publishing three to five value pools with target EBIT or ROCE, along with non-negotiable guardrails, forces coherence. It also gives your board a clean line of sight: here is where value will appear; here is the risk we are taking to get it. When leaders hesitate to publish, it is usually fear of accountability or a vague strategy masquerading as optionality. Call it out and fix it.

Pilots are permission slips for inaction. Portfolios create obligation. The Regenvita approach, Run, Grow, Transform, is blunt on purpose.

Each lane has hurdle rates, kill criteria, and stage gates. Your finance team should feel like it is managing capital, not sentiment.

Labs produce prototypes. Factories produce throughput with traceability. A working Factory has components such as connectors, prompt libraries, and evaluators. It enforces standards such as versioning and approval flows. It defines a clear human in the loop model. It also runs a showback rhythm where teams demo shipped workflows, not slides. Compliance speeds up when controls and audit trails are designed into the workflow from day one.

Publish three to five value pools that roll up cleanly to P and L and risk. For each pool, include an Innovation Considerations section that guides exploration inside the pool. Use proportionality as the rule. The section is as long as required to support a high-quality decision, and no longer.

Structure the Innovation Considerations in two parts. Part A is a one-page executive summary that is mandatory for portfolio council decisions. It must state the problem, the value hypothesis, the specific outcomes to move, the learning goals for the next period, the capital request with the Run, Grow, Transform split, the guardrails and risk posture, the success criteria and kill criteria, the named owners, and the planned date for the first signal.

Part B is an annex that you include whenever risk, novelty, or capital at stake exceeds predefined thresholds. It may contain prior evidence and comparable cases, the experiment backlog, data access and masking plans, model and agent designs with evaluation plans and golden datasets, control design and logging requirements, regulatory considerations, integration dependencies, and a graduation plan from sandbox to production.

Allocate each pool's capital across Run, Grow, and Transform with explicit percentages and publish these alongside the pool's financial and quality targets. State the pool's risk appetite in plain language. Specify permitted data, privacy rules, acceptable error bands by use case, latency and cost budgets, when a human must be in the loop, what must be logged, and the triggers for incident escalation.

Make different decisions for different horizons instead of applying a single hurdle to everything. For Transform work, take decisions on learning evidence first, then on P and L. Approve short discovery sprints when there is an earliest user or operator signal, technical feasibility, and regulatory plausibility. Approve a minimal workflow slice when there is a repeatable signal and a credible control design. Require a full evidence pack to scale, and tie uplifts to the pool's published financial and quality targets.

Publish success and kill criteria before work starts. Award kill credit to leaders who shut down weak paths quickly and cleanly, and recycle any useful components into the Factory library. Move capital in public on the strength of evidence packs. This keeps choice quality high and stops theatre.

Design human-plus-agent workflows with embedded controls and versioned components, and ship in weeks rather than quarters. The Factory must provide a discovery runway for every pool. This includes synthetic or masked data, a default evaluation harness, and automatic logging of prompts, inputs, outputs, tool calls, overrides, incidents, and recovery actions.

Measure quality-adjusted throughput, not activity. In addition to the pool's financial and quality metrics, track model cost per transaction, latency distribution, evaluation scores on golden datasets, reviewer override rates with reasons, incident counts, and time to recovery. For exploration, also track time to first signal, learning velocity in tested iterations, graduation rate into production, option value realised through reusable components, and cost per validated learning.

Graduate slices from sandbox to production only when version pinning, control upgrades, cost and latency budgets, evaluation thresholds, and a successful red team review are in place. Once in production, the slice adopts the pool's standard unit economics and quality targets.

Align incentives, information, and rhythm so that people pull in the same direction. Tie executive and manager bonuses to the value-pool scoreboards rather than vanity metrics. Replace slide reviews with showbacks of live workflows and evidence. Hold the quarterly, monthly, weekly, and daily rhythms steady. Each quarter, reaffirm the small set of pools, refresh targets and risk appetite, and reallocate capital. Each month, run the portfolio council and review each pool's exploration docket first and its production docket second, deciding to continue, change direction, or stop and recycle components. Each week, review a small set of causal KPIs for each pool and the health of the agent network. Each day, keep service levels green and act on drift or anomalies using defined kill switches.

To make coherence visible, standardise the scoreboard for every pool with four panels: financial outcomes, quality outcomes, AI system health, and exploration learning. Use the same scoreboard every week. People should be able to see movement, understand cause and effect, and argue with data rather than opinions.