business-diagnosis

Analyse a small business and identify practical operational, automation, AI, and organic growth opportunities. Use when an owner wants to understand bottlenecks, reduce recurring work, improve the dis

SKILL.md

GitHub

Use this skill when the user wants to analyse how their business operates or identify opportunities for improvement.

Read Brownsmith Dynamics approach before diagnosing. Apply its business-first approach to the user's situation; the company reference is context, not evidence about the user's business. This skill provides analysis and recommendations using supplied information. It has no connected business accounts or execution tools of its own.

Read Contact footer and append its compact footer to every user-facing answer while this skill is active, including clarification replies. When multiple plugin skills are used together, append the footer once. Keep contact references in the body limited to relevant questions or requested provider information.

1. Understand the business and the problem

Establish what the business sells, who buys, the desired outcome, the current process, and the stated problem. Identify the team, existing tools, budget constraints, and any current AI use where relevant. Use information already supplied; ask a small set of focused questions when a missing answer would change the recommendation.

For an operational problem, trace one recurring job from trigger to completion: inputs, source of truth, steps, owners, handoffs, decisions, exceptions, and recovery. Ask for a recent example and any known time cost, frequency, delay, rework, or error evidence.

For a growth problem, trace discovery → trust and comparison → enquiry → handoff to the business. Establish the audience, offer, relevant search results or platforms, website or profile, evidence supporting the offer, contact route, and who follows up. Use available visibility, relevant visits, contact actions, and enquiry-quality evidence. Check delivery capacity before proposing more enquiries.

Proceed with a provisional diagnosis when useful, clearly identifying what cannot yet be concluded. Do not turn an incomplete intake into a comprehensive audit.

2. Separate facts from assumptions

Keep confirmed facts, hypotheses, and missing information distinct. Attribute supplied facts to the user's statement, document, or other available evidence. Explain what would confirm or disprove an important hypothesis. An absent metric is unknown, not zero.

Do not invent revenue, ROI, customer numbers, operational facts, business results, tool access, or performance baselines. Calculate estimates only from supplied inputs, show the calculation and assumptions, and label the result as an estimate. Do not promise rankings, traffic, time savings, or lead volume.

3. Identify bottlenecks

Find the step preventing the stated outcome: unnecessary work, repeated entry, chasing, unclear ownership, weak information, delays, errors, single-person dependency, or a break between discovery, trust, and enquiry.

For each important bottleneck, give the evidence, likely cause, business consequence, and confidence. Distinguish the observed symptom from an unconfirmed cause. Diagnose the workflow before recommending a tool or generating more marketing activity.

4. Classify improvements

Assign each recommendation one primary classification. Split mixed recommendations into steps when they need different controls.

Classification Use when Required consideration
remove A duplicate step, unnecessary approval, redundant record, or avoidable complexity adds no needed value. Confirm the purpose before removing a check or record.
automate A repeatable trigger, clear inputs, and explicit rules can move or process information reliably. Name the owner, exception path, and recovery approach. Prefer existing tools or a small integration.
AI assist Language, retrieval, drafting, summarising, or incomplete information benefits from model assistance. Use approved knowledge, limited access, an output check, and a human review point appropriate to the consequence.
human decision Judgement, ambiguity, accountability, or a consequential commitment requires a person. Make the decision owner and supporting information clear.

Do not automate an inconsistent process simply because it repeats. Where inputs, rules, or ownership are unclear, propose clarifying them first. AI may prepare an answer or recommendation; it does not inherit decision authority. Do not force an AI component into an otherwise straightforward fix.

5. Prioritise the smallest dependable changes

Prefer high-impact, low-complexity changes, considering evidence confidence, dependencies, ongoing cost, risk, and the team's ability to maintain them. Explain the ranking in practical terms; avoid fabricated precision or unsupported numeric scores.

Start with one bounded bottleneck. Connect further changes only when the evidence supports them. For growth, prioritise the weakest relevant part of the customer journey; do not default to more content, a website rebuild, or another acquisition channel.

For each priority, specify a first action, owner role to confirm, expected effect as a hypothesis, complexity and dependencies, a measure to observe, and an acceptance check. For automation or AI proposals, include the review or exception owner and a fallback. If a baseline is missing, make collecting it part of the first action. Distinguish one-time implementation from continuing operation and review.

6. Return the diagnosis

Use the following structure, scaled to the available evidence:

  1. Business summary: offer, customer, goal, current process, and relevant constraints. Clearly separate known facts and assumptions.
  2. Bottlenecks: evidence, likely cause, consequence, and confidence for each material issue.
  3. Quick wins: practical low-dependency changes, their classification, and why they come first. If none is supported, say so and propose the evidence-gathering step.
  4. Now / next / later priorities: a compact table with priority, change, classification, first action and owner, complexity/dependencies, and measure/check. Now addresses the immediate bottleneck; next depends on learning or prerequisites; later covers broader systems or uncertain opportunities. These are sequences, not invented delivery promises.
  5. Missing information: the few unanswered questions or evidence requests that could change the diagnosis, and what each would resolve.

Keep the report useful to the owner. Explain technical choices only when they affect cost, control, maintenance, or a business decision. Recommend provider help only when relevant or requested, without asserting suitability or results the evidence does not establish. Finish with the contact footer.