🔎 Map: identify the real constraint

The discovery phase is the period when the consultant steps into the client’s operations to understand what’s really happening—without assuming the solution. It counteracts the bias of “I already know what you need.”

Consultants who skip discovery prescribe for the symptom, not the cause. The listening phase reduces the risk of bringing in the wrong solution and damaging the relationship in the first month.

Structured discovery; hypotheses to validate (not certainties); separate symptoms from causes; build trust before making recommendations.

The SPIN framework guides the sequence of questions: first map the current situation, then explore the perceived problem, dig into its implications, and finally articulate the real need for change.

Poorly sequenced questions lead to generic answers. SPIN has been validated over decades of consultative sales research—and works just as well for diagnosis.

Situation (context) → Problem (explicit pain point) → Implication (cost of the pain) → Need (what changes if it’s solved). Open-ended questions lead; closed-ended ones confirm.

Active listening means paying attention to what’s being said—and what isn’t. Confirmation bias is a consultant’s tendency to hear only what confirms their initial hypothesis, ignoring data that contradicts it.

The most common diagnostic trap: going in with a solution hypothesis and unconsciously gathering only evidence that confirms it. The result: a biased diagnosis and the wrong solution.

Listen without an agenda; note what contradicts the hypothesis; paraphrase to confirm understanding; use silence to go deeper.

Gemba (a Japanese word meaning "the actual place") is the practice of going where the work happens and observing directly. What’s written in the process is rarely what people actually do.

Documented processes reflect the ideal; actual work is full of hacks, workarounds, and informal steps that are invisible on paper. Anyone who doesn’t go to the gemba diagnoses the official process, not the real one.

Gemba walk; official process vs. actual process; shadow work (invisible parallel work); bottlenecks that only show up in real time.

Stakeholder mapping identifies who is affected by the problem, who has decision-making power, and who funds the solution. Often, the people who feel the pain (operations) aren't the ones who pay (leadership)—and the two have different agendas.

Making a recommendation without understanding internal politics is risky. The technically correct solution can get stalled by a lack of executive sponsorship or operational resistance.

Influence × interest map; executive sponsor; operational champions; who loses power with the change; align agendas before making a proposal.

After the interviews and observations, the consultant organizes the findings into a diagnostic document: identified patterns, confirmed and disproven hypotheses, and the two or three most critical constraints.

Raw interview data doesn’t turn into recommendations on its own. Synthesis is the intellectual work that turns listening into an actionable diagnosis—and demonstrates the consultant’s value to the client.

Discovery report; top 3 constraints; evidence for each finding; discarded hypotheses (and why); clear next steps.

Value stream mapping (VSM) is a Lean technique for visually mapping every step that transforms an input into value for the customer — including processing time, waiting time, and handoffs between departments.

Without seeing the full workflow, it’s impossible to identify where the real constraint is. Often the visible problem (slow performance in X) is caused by a bottleneck in Y that comes earlier.

Value stream map; lead time vs. processing time; wait time between steps; identify where work stops and waits.

Each process step should have numbers: average execution time, queue time, rework rate (the percentage of times an item is sent back for correction), and unit cost (hours × people involved).

Without data for each stage, prioritizing where to apply AI is subjective. With data, the consultant turns perception ("this is slow") into evidence ("this stage consumes 40% of total lead time").

Cycle time; lead time; throughput; rework rate; unit cost per step; identify the steps with the highest relative cost.

Eliyahu Goldratt’s Theory of Constraints (TOC) says every system has exactly one bottleneck — its slowest step — which determines the maximum flow rate. Improving any other step doesn’t speed up the result.

It’s the most important mapping principle. Without identifying the bottleneck, the consultant risks optimizing steps that don’t change the outcome—adding cost without noticeable gains.

TOC; the bottleneck as the slowest step; elevating vs. exploiting the bottleneck; making everything subordinate to the bottleneck; the bottleneck shifts after optimization.

Volume and frequency answer "how often does this happen?" — the basis for calculating the potential impact of any automation. A process with 5 occurrences per year rarely justifies the investment; one with 500 per day almost always does.

Automation ROI is directly proportional to the volume of repetitive work. Collecting frequency data turns a solution idea into a defensible business case.

Daily/monthly frequency; transaction volume; seasonality; total annual cost of the step = frequency × unit cost × people.

Classify each process step along a spectrum: from entirely rule-based (deterministic, automatable without AI) to entirely based on contextual judgment (requires a human). AI fits in the middle—where there are patterns but also variation.

The classification guides which pyramid level to use: pure rule → deterministic; pattern with variation → AI; complex context + decision → human or supervised agent.

Deterministic → AI → agent spectrum; rule-based vs. judgment-based work; fully automated vs. assisted; where to keep a human in the loop.

The process map drawn by the consultant needs to be validated with the people who do the work every day. Mapping errors—missing steps, incorrect sequence, estimated volumes—are common and invalidate the entire assessment.

A wrong map leads to a wrong recommendation. Validation with the operations team is the cheapest “reality check”—and it also builds buy-in from the people responsible for the change.

Validation workshop; walk-through of the map with operations; collaborative corrections; buy-in before the final diagnosis.

All generative AI use cases in a business context fit six patterns: classify (categorize text/data), extract (pull structured information from free text), generate (create content), summarize (condense), predict (infer a future state), and converse (dialogue interface).

Knowing the pattern catalog by heart speeds up diagnosis. When a consultant sees a step, they can quickly tell whether it’s “classify,” “extract,” or “generate”—and know which tools and risks apply.

Classification; entity extraction; text generation; summarization; prediction/scoring; chatbot/assistant. Each pattern has different risks and levels of technological maturity.

Each functional area has recurring opportunity patterns: sales (lead qualification, proposal generation), support (ticket triage, knowledge base), finance (invoice data extraction, reconciliation), HR (resume screening, onboarding), operations (production planning, data anomalies).

Recognizing patterns by department speeds up hypothesis generation during discovery. The consultant arrives at the meeting already knowing what to look for in each department.

Quick wins by department; more mature vs. more experimental cases; where adoption rates are highest; where hallucination risk is critical.

For each step in the mapped process, the consultant finds a match: which AI pattern solves this? The question isn't "what AI exists?" but "which pattern fits this specific step?" — a process that starts from the inside out.

Consultants who think from the outside in (starting with the tool) force artificial fits. The right method is to start with the process and ask which pattern solves it — radically improving the quality of the recommendation.

Process-first approach; matching question (which pattern solves this?); avoid looking for a problem to fit the tool; fit criteria: available data, acceptable variation, consequences of errors.

For each opportunity identified, the consultant places it on the pyramid: is it deterministic (rule + if/else)? A workflow with AI (LLM in one step)? An autonomous agent? The position determines implementation cost, risk, and timeline.

Many opportunities that seem like "agents" are actually "AI workflows" — or even "deterministic" solutions. Going to a higher level than necessary increases cost and risk without improving results.

Pyramid: deterministic → AI → agents; criteria for moving up a level; complexity overfitting; the lowest sufficient level that solves the problem.

Some AI use cases are "seductive"—they seem perfect but rarely deliver real value: a customer service chatbot without a structured knowledge base, financial report generation without clean data, a sales agent without a CRM. The problem comes before the solution.

Knowing how to identify red flags before proposing a solution saves the consultant from projects that predictably fail — destroying reputation and relationships. It's "saying no" applied to the catalog.

Prerequisites for each pattern (clean data, stable process, clear KPIs); chatbot without a knowledge base; generative AI without human validation in a high-risk context; automation of a broken process.

The deliverable for this phase is a catalog of all the AI opportunities identified for the client: process step, use case pattern, pyramid level, preconditions, and dependencies. No prioritization yet—that comes in the next module.

The catalog is the raw material for prioritization. Without it, deciding what to do first is intuitive and defensible only by the consultant’s prestige — not by analysis.

Opportunity table; fields: step, pattern, level, preconditions, data needed; deliberately broad scope before prioritization.

The matrix’s vertical axis measures the opportunity’s potential value. Four dimensions: impact on the end customer (quality, speed, satisfaction), process frequency (volume × cost per occurrence), costs avoided (how much spending is eliminated), and enabled revenue (what becomes possible to sell).

Perceived value without data is a weak argument. With the four dimensions, the consultant builds the numerator of the ROI — and the client can see concretely what's at stake.

Impact × frequency; cost of not changing; incremental revenue; qualitative value (satisfaction, reputational risk) vs. quantitative value.

The horizontal axis measures how feasible it is to implement the opportunity. Four dimensions: data readiness (does the data exist? Is it clean?), technical complexity (off-the-shelf vs. custom), integration complexity (do APIs exist? Legacy systems?) and organizational change (training, resistance).

Without assessing feasibility, high-value projects become sinkholes for time and budget. An honest assessment of the actual effort is what distinguishes consultative diagnosis from project sales.

Data (availability, cleanliness, access); available technical stack; required integrations; change management; realistic vs. optimistic timeline.

The combination of value and feasibility creates four quadrants: high value + high feasibility = quick wins (do now); high value + low feasibility = bets (invest, but cautiously); low value + high feasibility = fillers (only if capacity allows); low value + low feasibility = discard (don’t do).

The matrix forces an explicit conversation about what to prioritize. Without it, every project seems urgent and the client can’t decide—and the consultant loses the lead in the assessment.

Quadrant 1 (quick wins): top priority; quadrant 2 (bets): future roadmap; quadrant 3: low-priority backlog; quadrant 4: explicit rationale for discarding.

Quick wins are high-value, highly feasible projects that can be delivered in weeks. Starting with them creates a quick, visible win—which builds client trust and generates the political momentum for larger projects.

Long projects without visible results lose sponsorship. The quick win protects the larger project: when leadership wants to cancel, you have a concrete win to show and a reason to continue.

Short time to value; prototype or pilot in 2-4 weeks; measurable, visible results; a quick win as proof of commercial value.

Optimism bias leads consultants and clients to overestimate the value of AI projects and underestimate the real effort required for integration, data, and organizational change. The result is projects that run late, cost more, and deliver less than promised.

Calibrating estimates conservatively protects the consultant’s credibility. It’s better to promise less and deliver more than the other way around—especially on the first project with a new client.

Planning optimism; reference class forecasting; multiply effort estimates by 1.5×; conservative commitments; underpromise-overdeliver.

The completed matrix becomes the agenda for the prioritization meeting with the client. The consultant presents the quadrants, supports each placement with data, proposes the quick wins, and invites the client to make adjustments—making the decision collaborative.

Presenting the prioritization, not just the list of opportunities, positions the consultant as a decision-maker — not just an implementer. It’s the difference between charging for deliverables and charging for judgment.

Prioritization narrative; quick win 1 and 2 proposal; roadmap of bets; communicate what was ruled out and why; turn analysis into a decision.