DECISION TOOLKIT · PORTFOLIO DESIGN
Turn a list of ideas into an investment sequence
The portfolio becomes strategic when each candidate has an owner, a baseline, a risk posture and a next decision. Potential alone creates theatre; executability alone creates small optimisations.
Enable
shared identity, data, evaluation and governance
capacity before applicationsProve
restricted workflows with measurable baselines
evidence before scaleScale
repeatable cases with owners and run-rate
reinvest only after proofPortfolio rule: no initiative enters Scale without an evidence file.
An enterprise AI roadmap is not a list of tools or use cases. It is a sequence of investment decisions: where to intervene, what must be prepared, which risks are acceptable and what evidence will justify the next tranche of capital.
This distinction matters in Switzerland, where organizations often combine high-value knowledge work, sensitive data, multilingual operations and exacting expectations around trust. A roadmap copied from a technology vendor rarely captures that operating reality.
Start with workflows, not ideas
Most opportunity lists are collected through workshops: each function proposes assistants, copilots or automation. The result can be energetic and still be unhelpful. Fifty disconnected ideas do not form a portfolio.
Map the workflows where value is created or lost. Look for decisions with slow access to knowledge, repeated handoffs, high rework, constrained expertise or material quality variation. For each workflow, name the owner and establish a baseline.
If nobody owns the workflow today, nobody will own the AI outcome tomorrow.
Score the opportunity and the conditions
Evaluate each candidate across two axes. The first is potential: economic materiality, strategic relevance, user impact and time-to-value. The second is executability: data access, integration complexity, control requirements, adoption capacity and evaluation feasibility.
The strongest first initiatives are not always those with the highest theoretical upside. They combine meaningful value with a short path to trustworthy evidence.
| Decision dimension | Direct question |
|---|---|
| Value | Which measurable workflow outcome should change? |
| Data | Can the required context be accessed lawfully and reliably? |
| Control | What harm could occur, and how will it be detected? |
| Architecture | What must integrate with identities and existing systems? |
| Adoption | Who changes how they work, and why would they? |
| Evidence | What result would justify scaling, adapting or stopping? |
Build a portfolio, not a queue
A credible roadmap balances three horizons:
- Enable — identity, data access, evaluation, governance and platform foundations shared across initiatives.
- Prove — a small number of workflow interventions designed to produce operational evidence quickly.
- Scale — only the initiatives that clear explicit quality, control, adoption and value gates.
Dependencies should be visible. If several opportunities require the same permission model or knowledge layer, that capability may deserve investment before the individual applications.
Make governance part of sequencing
Governance is not a review at the end of a build. Classify risk while shaping the portfolio. A low-risk internal drafting assistant and a system influencing customer decisions should not face the same evidence threshold.
For each initiative, assign an accountable business owner, technical owner and risk owner. Define the decision gates before delivery begins: permission to experiment, permission to use sensitive data, permission to expose the system to users and permission to scale.
Attach value to the roadmap
Before claiming a benefit, record the current workflow. Useful baselines include cycle time, completion rate, rework, exception rate, unit cost, quality variation and risk exposure. Usage is not value; it is only an input into a changed operating process.
The roadmap should be governed monthly. Review evidence, run-rate, emerging risk and portfolio trade-offs together. Some initiatives will scale. Others should change direction or stop. A roadmap is valuable because it makes those decisions faster and better — not because it predicts the next eighteen months perfectly.
The minimum credible output
A useful enterprise AI roadmap contains:
- a shared executive thesis;
- a workflow-based opportunity portfolio;
- a readiness and dependency map;
- risk classification and decision gates;
- a rolling 12–18 month sequence;
- named owners, baselines and funding logic.
That is enough to move from AI ambition to an executable first portfolio without creating a strategy program larger than the work itself.
Build the opportunity inventory from operating evidence
The first version of the portfolio should be assembled from facts, not a brainstorming wall. Interview workflow owners, inspect queue data, sample completed cases and identify the points where expert capacity is scarce or rework is expensive. In a Swiss enterprise, also segment by language, business entity, data residency, professional secrecy and customer commitment. A use that appears simple in one market may require a different control posture in another.
For each candidate, record a minimum viable opportunity card:
| Field | What a useful answer looks like |
|---|---|
| Workflow owner | a person who controls the process and can change it |
| User group | named roles, volumes and current workarounds |
| Trigger and output | the event that starts the work and the artifact or action produced |
| Baseline | current time, quality, cost, exceptions and escalation |
| AI mechanism | retrieve, classify, draft, recommend, route or act |
| Data and systems | sources, permissions, integrations and exclusions |
| Risk posture | consequence, human role, reversibility and required evidence |
| First decision | the next gate and what would justify stopping |
The card is deliberately operational. “Deploy a copilot for sales” is an aspiration. “Reduce proposal rework for the Swiss industrial sales team by retrieving approved technical evidence and flagging unsupported claims” is a portfolio candidate that can be baselined, evaluated and owned.
Score potential and executability separately
Use two scores rather than one blended number. Potential captures the upside; executability captures the conditions required to reach it. A high-potential, low-executability case should not disappear. It should become an enabler or a staged research item with an explicit dependency. A low-potential, high-executability case may be useful as a learning vehicle, but it should not consume the attention reserved for material outcomes.
For each score, use a five-point rubric and record the evidence behind the rating. Potential can combine economic materiality, strategic differentiation, user pain and scale. Executability can combine data readiness, integration complexity, control burden, evaluation feasibility and adoption capacity. Avoid false precision: the written rationale is more valuable than a decimal score.
The portfolio conversation should ask four questions:
- Which two or three cases could produce credible evidence within one quarter?
- Which shared capabilities unlock more than one case?
- Which high-potential cases require a regulatory, data or operating decision first?
- Which ideas are being kept only because nobody wants to say no?
Sequence by dependency, not by enthusiasm
Roadmaps fail when every use case is scheduled as if it were independent. Create a dependency map. Identity and access can unlock several workflows. A controlled document index can support search, drafting and review. An evaluation harness can reduce the cost of every subsequent pilot. Conversely, an immature approval process can block all customer-facing work regardless of model quality.
Use a simple sequencing rule: fund the smallest enabler that materially changes the executability of several candidates. Do not build a generic platform “for future AI” without a live workflow that will consume it. The right foundation is a product of the first portfolio, not a parallel infrastructure program.
| Sequence decision | When to choose it | Evidence required |
|---|---|---|
| Start the case | value is material and conditions are ready | owner, baseline, scope and stop criterion |
| Prepare the enabler | one capability unlocks multiple cases | dependency map and named consuming workflows |
| Split the case | risk or integration makes the full scope too large | a smaller slice with independent value |
| Defer the case | critical condition is unresolved | explicit owner and a trigger to revisit |
| Retire the idea | value is weak or the mechanism is not credible | decision record, not silent disappearance |
Use a 12-month horizon with rolling decisions
An executive roadmap should be directional over 12–18 months and precise over the next 90 days. The first quarter should contain named work, named owners and a decision date. The second and third quarters should express outcomes and dependencies, not pretend that a model and a workflow can be predicted in detail.
Quarter 1 — establish and prove. Complete the inventory, select the first cases, baseline the workflows, set up the evaluation environment and run restricted pilots. The output is evidence, not a platform launch.
Quarter 2 — scale what clears the gate. Industrialise the strongest mechanism, instrument cost and quality, train the operating team and decide which second-wave cases can reuse the foundation.
Quarter 3 — widen the portfolio. Add cases only when the operating model can absorb them. Review whether the first cases changed capacity, cycle time, quality or risk, not only whether users opened the tool.
Quarter 4 — rebalance. Retire weak cases, renegotiate vendor boundaries where needed, fund the most valuable enablers and set the next annual thesis from measured outcomes.
Govern the roadmap as a capital allocation system
The roadmap owner is not responsible for making every initiative succeed. The role is to make trade-offs visible and move capital, senior attention and engineering capacity toward the best evidence. A monthly review can be short if it has a fixed pack: forecast versus realized value, current cost, control status, adoption by eligible workflow, major dependencies, decisions due and recommendation to scale, adapt or stop.
This creates a healthier relationship with uncertainty. A pilot that stops because evidence is weak has produced useful portfolio information. A pilot that continues because the team is attached to its original idea has produced neither learning nor value.
The best roadmap is therefore not the one with the most initiatives. It is the one in which every initiative has a next decision, every dependency has an owner and every scaling claim can be traced back to a baseline.