Stop Looking for AI Use Cases. Start Allocating Intelligence.

Most AI use case discovery goes wrong before the first sticky note reaches the wall.

The room usually opens with the same question: where can we use AI? It sounds practical and modern, and it quietly corrupts the exercise, because it makes the technology the centre of gravity. From then on, the team looks for places to apply a tool instead of places where the business is losing value.

The strongest framework starts from business problems and the allocation of intelligence, with AI use cases coming later. Most organizations have plenty of AI ideas. They fail because they generate too many weak ones, attach them to no material business outcome, underestimate the operating model required, and confuse visible experimentation with strategic progress.

So "where can we apply AI?" is the wrong opening question. A sharper one is:

Where do business outcomes suffer because the organization cannot understand, decide, create, coordinate, or act at the level or speed required?

Then a second question:

If we had near-unlimited intelligence available at low marginal cost, would this problem materially improve?

And a third, which most AI workshops avoid:

Is AI actually the best lever, or is the real answer process redesign, better data, simpler governance, rules automation, or role clarity?

The order is the philosophy: business problem first, intelligence gap second, intervention choice third and AI use case fourth.

Why use case brainstorming produces weak strategy

A technology-led workshop produces predictable output. Someone suggests a chatbot, then automated content, knowledge search, meeting summaries and a dashboard that writes recommendations. None of these ideas is necessarily bad. The trouble is that they are usually detached from the question that matters: which business performance constraint are we removing?

When the starting point is AI, the organization rewards novelty. When the starting point is business outcomes, it rewards materiality.

That is why so many AI portfolios become long lists of disconnected ideas. They look impressive in a steering committee deck and fail to explain what will improve, who owns the outcome, what process will change, which metric will move, what risk will be accepted and what legacy work will stop. The portfolio turns into theatre while the business stays the same.

I have seen this pattern across large transformations for most of my career. The language changes and the behaviour does not. Today it is AI use case discovery. Before that it was digital innovation pipelines, analytics use case factories, automation backlogs, agile product ideas and CRM opportunity maps. Each time, the volume of ideas was mistaken for strategic clarity.

A serious executive approach does the opposite. It reduces the field, makes weak ideas uncomfortable, and forces every opportunity to prove that a meaningful outcome is constrained by intelligence that is scarce, slow, inconsistent, expensive or hard to scale. BASICS is designed to do that.

The BASICS framework

BASICS is a practical executive framework for turning business problems into disciplined decisions about where to allocate intelligence. It stands for:

1. Business outcomes
2. As-is friction
3. Size the value pool
4. Imagine unlimited intelligence
5. Choose the right intervention
6. Sequence, pilot and scale

Its strength is the order. You do not earn the right to talk about AI until you have described the business outcome, diagnosed the friction, sized the value and shown that intelligence is a real constraint. That order sounds obvious, and in practice it is rare.

1. Business outcomes: start where the business hurts

Start with the business rather than the technology. Define the top outcomes the function or organization is trying to improve over the next 12 to 36 months, concretely enough to survive contact with a CFO, a business unit president or a sceptical operator.

Typical categories include revenue growth, margin improvement, cost efficiency, cycle time, quality and accuracy, risk reduction, customer experience, employee productivity, knowledge reuse and speed of innovation. For each area, define five things: the objective, the KPI, the gap against target, the cost of inaction and the strategic importance.

That sounds basic because it is, and it is also where many AI conversations fail.

A weak statement is "we want to use AI in marketing". A strong statement is: marketing-sourced pipeline is below target, lead leakage is too high, campaign cycle time is slowing the response to market signals, personalization cannot scale economically, and sales follow-up quality is inconsistent across regions.

Now there is something worth discussing: outcomes, measurable gaps and a business system under stress. Without this step, use case discovery is brainstorming. With it, the conversation becomes strategy.

2. As-is friction: find where the business breaks down

Once outcomes are clear, diagnose where the business breaks down today. Look across customer journeys, internal workflows, functional processes, decision forums, knowledge flows, content production, analytics and reporting, and handoffs between functions. For each pain point, document what happens today, where the friction occurs, who experiences it, what downstream impact it creates, whether it recurs, and whether it is structural or operational.

It helps to classify friction into six types.

Information friction appears when people cannot find, trust or synthesize the information they need. The data may exist and still be scattered, stale, ambiguous, inaccessible or politically contested.

Decision friction appears when decisions are slow, inconsistent, low in quality or dependent on a few experts, so the organization waits for scarce judgment.

Creation friction appears when content, code, analysis, proposals, plans, reports or other assets take too long to produce. The organization knows what it needs and cannot create it at the required volume, quality or speed.

Coordination friction appears when work is trapped in handoffs, approvals, follow-ups, dependencies and status chasing. The missing intelligence here is orchestration as much as cognition.

Execution friction appears when the work is known but not done reliably or fast enough. The gap is operational consistency rather than strategy.

Learning friction appears when the organization repeats mistakes because experience does not compound. Lessons stay local, best practice is rediscovered, and institutional memory leaks through every reorganization.

Real AI opportunities start to appear here, as long as you resist naming solutions too early. Pain points and friction types are diagnostic inputs rather than use cases. The point is to understand the constraint before designing the intervention.

3. Size the value pool: materiality before novelty

Not every pain point deserves investment, and this is where many teams lose discipline. A use case can be elegant, impressive and technically feasible while still being strategically irrelevant. Demonstrable value is the standard.

Size each pain point along one or more dimensions: revenue upside, cost reduction, hours saved, cycle time, error reduction, risk reduction, experience improvement and strategic capability gained. Use both quantitative and qualitative sizing. At executive level the question is simple:

If this pain point disappeared, how much better would the business perform?

Look for pain that is frequent, costly, cross-functional, located in strategic growth areas, or created by scarce expert judgment.

The AI ROI Calculator helps here. A calculator cannot replace executive judgment, but it forces the assumptions into the open. A value case that depends on heroic adoption, perfect data or productivity savings that never leave the cost base is a story rather than a value case.

The output of this step is a value-ranked list of business problems worth solving, rather than a list of AI ideas.

4. Imagine unlimited intelligence: the core reframe

This is the intellectual centre of the framework. Instead of asking what AI can do, ask what would change if intelligence were abundant, cheap, fast and always available. What would we understand or predict better? What would we create faster or decide more consistently? What could we personalize at scale, coordinate automatically, monitor continuously, or learn once and reuse?

That question moves AI away from tool adoption and toward capability design.

It helps to break intelligence into seven jobs.

1. Understand: summarize, classify, extract, interpret and contextualize information that is otherwise too fragmented or slow to process.
2. Predict: forecast, score, estimate likelihood, find patterns and anticipate exceptions before they show up in lagging indicators.
3. Recommend: suggest next best actions, prioritize work, advise users and make judgment more consistent across teams.
4. Create: draft content, code, designs, analysis, plans, responses and proposals that used to depend on scarce human capacity.
5. Decide: support or automate bounded decisions within thresholds, with clear rules for confidence, escalation and override.
6. Coordinate: trigger workflows, follow up, orchestrate steps across systems and reduce the manual effort of keeping work moving.
7. Learn: capture institutional knowledge, improve from outcomes, reuse best practice and stop paying repeatedly for the same lesson.

Then take each important pain point and ask four questions. Is the constraint really a shortage of intelligence? If so, which kind of intelligence is missing? Would more of it materially change the outcome? Does the problem need human judgment, or can the intelligence be partly industrialized?

Lead follow-up is a good example. The superficial use case is automated outreach. The stronger diagnosis is inconsistent follow-up quality and lead leakage, where the missing intelligence may be prioritization, qualification, next-best-action guidance, contextual message drafting and escalation. The hypothesis is that abundant intelligence could score leads, recommend follow-up, draft outreach and flag risk automatically. That is intelligence allocated against a specific commercial constraint, which is much narrower and more useful than "AI for sales".

5. Choose the right intervention: AI is not always the answer

This is the most underrated step. Not every intelligence problem needs AI. Sometimes the answer is better process design or governance, system integration, master data, rules-based automation, standardized templates, training, enablement or clearer ownership. For each opportunity, force a real choice between seven interventions:

1. No technology: clarify roles, fix policy, redesign the workflow or remove unnecessary steps.
2. Basic digital: dashboards, forms, workflow tooling and integrations.
3. Rules automation: deterministic logic, robotic process automation and explicit business rules.
4. Analytics: BI, forecasting, segmentation, measurement and structured insight.
5. AI assistance: a person remains the actor, and AI improves speed, quality or consistency.
6. AI automation: AI performs bounded tasks with human review, escalation and control.
7. Agentic orchestration: AI reasons, decides, coordinates and acts across tools with oversight.

This protects the organization from putting AI on top of a broken process. If the approval chain is irrational, automating the memo will not fix the operating model. If a data definition is contested, a smarter interface will only make bad data more persuasive. If nobody owns the outcome, adding AI gives the problem a more modern costume.

AI is usually a strong fit when the work is high in volume and variability, knowledge-intensive, language-heavy, dependent on pattern recognition or decision support, too expensive to scale with people alone, speed-constrained, or dependent on personalization and context. It is usually a weak fit when the process is broken, the data is too poor, the task is simple and deterministic, the business rules are stable and explicit, the bottleneck is organizational rather than cognitive, or the risk of a wrong action is too high for current controls. That sentence should be on the wall of every AI steering committee.

The goal is to allocate intelligence and intervention capacity where it changes business performance, rather than to maximize AI adoption.

6. Sequence, pilot and scale

The last step turns qualified opportunities into a sequenced portfolio. Each justified use case gets a card, a score and a place in the portfolio, and each pilot gets a baseline, a success threshold and a decision gate before it starts.

From opportunity to use case card

Once AI is justified, define the use case properly. A serious card has twelve elements: business problem, outcome KPI, user or stakeholder, process point, intelligence job, AI pattern, data and systems needed, human-in-the-loop design, risk profile, value hypothesis, feasibility and pilot design.

That structure prevents vague use cases such as "AI for marketing" or "AI for operations". Which pain point are we solving? Which metric improves, and who benefits? Where does it sit in the workflow? Is the job to understand, predict, recommend, create, decide, coordinate or learn? Should the solution be an advisor, copilot, analyst, creator, triage engine, decision engine, agent, knowledge engine or simulation engine? What data, sources, integrations and permissions are needed? What stays human, what is supported by the machine and what is automated? What could go wrong, and how will we prove value quickly? A use case without these answers is an idea with a name.

That is why the Use Case Prioritization tool is built around value, feasibility, readiness and risk. The discipline lies in forcing each idea to compete, rather than in producing it.

Prioritize with a serious scoring model

A strategic portfolio needs more than high, medium and low. I score each qualified use case from 1 to 5 on six dimensions.

The AI advantage score stops teams from calling everything AI when analytics, workflow tooling or rules would be cheaper, safer and more robust. The readiness score matters just as much. A use case that is technically feasible but operationally impossible belongs in the future-state vision rather than on the roadmap.

Build a portfolio, not a list

Do not end with 50 disconnected use cases. Build a balanced portfolio with four buckets. Quick wins are low-complexity opportunities with visible value and near-term adoption; they build confidence but should not become the whole strategy. Strategic bets are high-value use cases that change how a function works, and they deserve senior sponsorship and stronger operating-model design. Foundational enablers are the data, knowledge architecture, APIs, governance, identity, permissions and workflow integrations that make advanced use cases possible. Watchlist items are interesting opportunities that are not ready yet and stay visible without consuming execution capacity.

Many organizations try to launch advanced use cases without the enabling layer. They want autonomous workflows while their knowledge base is unmanaged, permissions are unclear and process ownership is fragmented, which is a sequencing failure rather than ambition.

A mature organization does not say it has 120 AI ideas. It says something like this: we identified 12 priority business problems where scarce intelligence limits performance; five are best solved with AI, three with process redesign, two with rules automation, and two need data remediation first; we are piloting three high-value AI use cases and building two foundational enablers. That is what strategic maturity sounds like.

Pilot with a decision gate

Before a pilot starts, define its scope, a control group where possible, the KPI baseline, the success threshold, the owners, the timeline and the rollout decision. The organization should know in advance what scales if the pilot works and what stops if it does not.

Design the right AI archetype

A common failure is treating every use case as a chatbot. The better question is what form the intelligence should take.

An advisor gives insights and recommendations. A copilot helps a person work faster. An analyst synthesizes data and interprets it. A creator drafts content, plans, designs and assets. A triage engine prioritizes cases, leads, tickets or requests. A decision engine recommends or makes bounded decisions. An agent executes multi-step workflows across systems. A knowledge engine captures and reuses institutional memory. A simulation engine models scenarios and trade-offs.

The same business pain can lead to very different designs. Lead leakage could become a prioritization model, a sales copilot, a routing engine, a next-best-action advisor or an agentic follow-up workflow. The right archetype depends on the process point, the risk profile, data readiness and the human role. Design must follow diagnosis, or the business problem ends up bent to fit the tool.

Separate three levels of opportunity

AI portfolios often stay shallow because teams focus on task-level productivity. There are three levels.

Task-level opportunities help a person do one task faster or better, such as drafting emails, summarizing meetings, producing campaign copy or generating first-pass analysis. They are useful and rarely transformational on their own.

Workflow-level opportunities make a process faster, more consistent or more automated, such as lead qualification and routing, proposal generation, customer response handling, case triage, planning cycles and reporting. Value starts to compound here.

Operating-model opportunities change how the organization works, such as always-on personalization, autonomous campaign optimization, AI-powered account intelligence, AI-supported commercial decisions and new services built on intelligence at scale. This is where the strategic ceiling rises.

Most organizations overinvest in the first level because it is easy to demonstrate and politically safe, and underinvest in the other two because they require process redesign, governance, role clarity and executive commitment. That is exactly why the other two matter.

Governance early, not late

Governance should arrive before the room has fallen in love with a use case. Screen every AI opportunity early for data privacy, security, IP exposure, regulatory concerns, brand risk, bias and fairness, explainability, human override, auditability and the severity of failure modes.

Then assign a risk tier. Low-risk augmentation helps people work faster without acting on its own. Medium-risk workflow support influences process outcomes while keeping meaningful review. High-risk decision or action automation affects decisions, customers, money, compliance or operations directly. The tier should shape the human role, evidence requirements, monitoring, escalation path and speed of rollout.

The Change Resistance Review belongs here because risk is not only technical. Many AI use cases fail because they collide with incentives, identity, trust or local power structures, so adoption risk is execution risk.

The workshop sequence

If I were running this inside an organization, I would start with six working sessions rather than a prompt library or a vendor demo.

1. Business pain discovery: participants identify the top business goals, constraints, pain points, bottlenecks and missed opportunities.
2. Intelligence reframing: for each pain point, the group asks what intelligence is missing, what abundant intelligence would change, and whether this is really an intelligence problem.
3. Intervention selection: compare process redesign, rules automation, analytics, AI assistance, AI automation and agentic workflow.
4. Use case definition: convert the selected opportunities into structured use case cards.
5. Prioritization and sequencing: rank by value, strategic fit, AI advantage, feasibility, readiness and risk.
6. Pilot planning: define scope, a control group where possible, the KPI baseline, the success threshold, owners, timeline and rollout gates.

The AI Readiness Assessment supports this sequence by making capability gaps explicit before the organization commits to use cases it cannot yet absorb.

The executive decision logic

The whole framework reduces to six questions.

1. Is this a meaningful business problem? If not, stop.
2. Is the root cause materially related to scarce or inconsistent intelligence? If not, fix the process, data, governance or ownership.
3. Would abundant intelligence change the economics of the problem? If not, deprioritize it.
4. Is AI the best way to provide that intelligence? If not, choose analytics, rules or process redesign.
5. Can we deploy it with acceptable risk, data readiness and adoption potential? If not, move it to the enabling backlog.
6. Can we prove value fast? If so, pilot it. If not, redesign the scope until you can.

That is the cleanest executive filter I know.

The strongest objection

A pragmatic executive will say this is too slow. Six workshops before anyone touches an AI tool, while competitors ship quick wins that build momentum and AI literacy across the workforce, and every framework eventually turns into bureaucracy.

Part of that is right. Quick wins and broad AI literacy have real value, and task-level tools can roll out in parallel under light governance without passing through every BASICS step. The framework is for the decisions where money, operating-model change and risk are at stake. Those are also the decisions where skipping the diagnosis costs the most, because a well-funded AI initiative aimed at the wrong constraint uses up the budget and the patience the right one would have needed.

What good looks like

The strongest AI strategy explains the quality of its choices rather than celebrating the number of ideas. It does not report fourteen workshops and 120 opportunities. It reports the business problems where scarce intelligence is limiting revenue, margin, speed, quality, risk or growth, separates AI opportunities from other fixes, builds a portfolio, funds enablers, and defines pilots with measurable thresholds, so that everyone knows what scales if it works and what stops if it does not.

That standard is less theatrical and harder to decorate, and much more likely to produce operating results. Start with where the business pays too much for intelligence, waits too long for it, accepts inconsistent intelligence, or fails to reuse intelligence it already paid to create. And if abundant intelligence would not materially change the outcome, be disciplined enough to solve the problem another way.

Monday move

Take your current AI use case list and pick the ten items with the most budget or attention. Next to each, write the business outcome, the KPI it should move and the friction type it addresses. Any item that cannot name all three goes back to step 1 before it receives more funding.