Overview

Tomoko Hayashi's leadership team had already approved a $1.2 million AI investment when someone asked the question no one had asked yet: "Are we solving the right problem?" The project - an AI tool to automate the weekly executive report - had been championed by one enthusiastic VP and moved quickly through budget approval. No one had looked at the full landscape of where the organization struggled most. Two months later, a cross-functional scan revealed that the highest-cost inefficiency was in contract renewal tracking, a process that was losing the company an estimated $3 million per year in missed deadlines and renegotiation gaps. The executive report took four hours a week. Contract renewal churn took months.

Starting with the technology and finding problems to fit it is one of the most reliable ways to underinvest in AI while feeling busy with it.

The Case for Scanning Before Committing

Most organizations discover AI opportunities the way they discover most opportunities: through champions. Someone sees a tool at a conference, a vendor pitches a solution, a team lead reads a case study. These inputs are real and sometimes excellent. They are also random. They reflect who happened to be paying attention, not a systematic reading of where AI would add the most value.

A structured opportunity assessment replaces random with deliberate. It gives your leadership team a common map - across the whole organization - before anyone commits resources. That map usually reveals two things: that the highest-value opportunities were hiding in plain sight, and that the use cases getting the most attention weren't necessarily the best ones.

The assessment is not a long-term research project. Done properly, a baseline scan of a mid-sized organization can be completed in two days of structured workshop time - if the right people are in the room and the right framework is running the conversation.

Step 1: Map Your Value Streams

A value stream is the sequence of activities that takes your organization from input to outcome. In a manufacturing company: procurement to production to delivery. In a financial services firm: client intake to underwriting to servicing. In a healthcare system: patient intake to diagnosis to treatment to follow-up.

Start by mapping four to eight major value streams in your organization. Not departments - streams. The difference matters because AI opportunities often live at the handoffs between departments, not inside them. That's where work accumulates, errors compound, and delays are hardest to see.

For each value stream, identify the main stages and estimate the volume of work at each stage: transactions per month, hours of human effort per week, error rates where known. You don't need precise numbers at this stage. Orders of magnitude are enough - a stage processing 500 invoices per week looks different from one processing 5,000.

This step typically takes half a day with the right cross-functional participants. The output is a rough map, not a process audit. Accuracy matters less than coverage.

Step 2: Score Each Stage for AI Readiness

Not every stage of every value stream is a good candidate for AI. Four factors predict AI-readiness reliably.

Data availability

AI needs data to learn from and operate on. Does this stage generate structured, accessible data? Is it captured digitally? Is it clean enough to use, or would it require significant preparation? A stage that relies on unstructured paper records or tribal knowledge scores low. A stage with years of digital transaction records scores high.

Task repetitiveness

AI performs best on tasks that follow recognizable patterns - the same inputs producing the same type of output, repeatedly, at volume. A task where every case is genuinely unique and requires fresh judgment every time is a poor fit. A task where 80% of cases follow three known patterns and 20% are exceptions is an excellent fit: the AI handles the 80%, humans handle the 20%.

Volume

The business case for AI investment improves with volume. A process that handles 10 cases per year is unlikely to justify significant investment regardless of its AI-readiness on other dimensions. A process that handles 10,000 cases per year - even at low unit value per case - starts to look different.

Error cost

What does a mistake in this stage cost? This includes financial cost, regulatory risk, customer impact, and downstream rework. High error-cost stages have two things going for them as AI opportunities: the potential upside of error reduction is significant, and there is usually executive attention and budget available to fund improvement.

Score each stage on each factor - a simple 1 to 3 scale works - and sum the scores. The highest-scoring stages are your priority candidates. The scoring is a structuring device, not a verdict. Review the top ten with the workshop group before drawing conclusions.

Step 3: Distinguish Quick Wins from Strategic Bets

Not all high-scoring opportunities are equal in terms of time-to-value. Some are quick wins: a high-volume, repetitive, well-documented task that can be automated with an existing tool in 60 to 90 days. Others are strategic bets: high-value opportunities that require data infrastructure investment, organizational change, or custom development before they can deliver results.

Both belong in your portfolio. Quick wins build credibility, fund subsequent investment, and demonstrate to skeptical stakeholders that AI delivers real value. Strategic bets create sustainable competitive advantage but require patience and sustained investment.

A common mistake is pursuing only quick wins. Organizations get good at automating small tasks and never build the foundational capabilities - data pipelines, model governance, change management - that unlock larger-scale transformation. Another common mistake is pursuing only strategic bets, burning budget and patience on long-horizon initiatives before the organization has demonstrated it can execute on AI at all.

A healthy portfolio has both. A rule of thumb that works in practice: target 60 to 70% of initial AI investment toward opportunities that can show measurable results within six months, and 30 to 40% toward longer-horizon bets that build future capability.

Step 4: Prioritize with ICE

ICE stands for Impact, Confidence, and Ease. Originally developed for product prioritization, it adapts cleanly to AI opportunity assessment.

• Impact: If this opportunity delivers on its potential, how much value does it create? Consider financial return, risk reduction, customer experience improvement, and employee time freed for higher-value work. Score 1–10.
  - Confidence: How confident are you that this opportunity is real and that the AI approach will work? High confidence if there are proven implementations at comparable organizations. Low confidence if the use case is novel and the data situation is uncertain. Score 1–10.
  - Ease: How difficult is implementation? Consider data readiness, change management complexity, integration requirements, and budget. High ease if existing tools and data can get you 80% of the way there. Low ease if significant infrastructure work is required first. Score 1–10.

Multiply Impact × Confidence × Ease to get an ICE score. Rank your candidates by ICE score. The top of the list is where to start your business case development.

ICE scores are not predictions - they are structured opinions. Scoring forces the conversation that would otherwise happen informally, where the loudest voice or the most enthusiastic champion wins. When everyone scores independently and then compares, disagreements surface, assumptions get examined, and the group reaches better decisions.

Running the 2-Day Workshop

A baseline assessment for a 500-to-2,000-person organization typically takes two structured workshop days.

Day 1: Map value streams (morning), identify stages and assign readiness scores (afternoon). Participants: six to ten people representing major functions, including at least one person from operations and one with data or technical literacy. Pre-work: each participant documents the two or three AI opportunities they're already aware of in their area.

Day 2: Review the top twenty candidates from Day 1, apply ICE scoring, distinguish quick wins from strategic bets (morning). Draft a prioritized opportunity map with a provisional portfolio allocation (afternoon). Output: a single-page ranked list with rationale, plus a brief for leadership review.

The two-day format creates enough space to do the thinking without becoming a months-long strategy exercise. Most organizations have enough information already to reach 80% of the insight - they just haven't had the structured conversation to surface it.

>
"The organizations that find the best AI opportunities are not the ones with the best technology instincts. They're the ones that asked where the real pain is before they bought anything."

Key Takeaways

• Starting with technology and finding problems to fit it is the most common - and most costly - failure mode. A structured opportunity assessment reverses the sequence: find the highest-value problems first.
  - Value streams, not departments, are the right unit of analysis. The highest-value AI opportunities often live at inter-departmental handoffs where inefficiency is hardest to see.
  - Four factors predict AI-readiness: data availability, task repetitiveness, volume, and error cost. Scoring each stage against these factors creates a defensible, common basis for prioritization.
  - A healthy AI portfolio includes both quick wins and strategic bets. Quick wins build credibility and fund future investment; strategic bets build sustainable competitive advantage.
  - ICE scoring - Impact × Confidence × Ease - structures the prioritization conversation. It replaces informal influence with shared, documented reasoning.
  - A baseline assessment can be completed in two workshop days if the right cross-functional participants are engaged and a clear framework is running the conversation.
  - The output is a prioritized map, not a plan. The assessment tells you where to invest further analytical effort - it doesn't replace the detailed business cases that follow.