A four-step funnel for finding where AI has the highest business impact

Executive overview

Most businesses either apply AI to everything and get disappointed, or are paralysed by choice. The answer is to start with the business, not the technology.

A four-step funnel — goal, bottleneck, function, activity — narrows focus to the single highest-impact starting point before any AI tool is chosen.

Identify the priority bottleneck first; the right AI application becomes obvious from there.

The two failure modes

• Over-eager teams automate broadly, then hit the limits of what AI can realistically do
• Overwhelmed teams can't choose among models and tools, so they never start
• Both groups make the same root error: starting with AI instead of the business problem

Step 1: Identify the one priority goal

• Ask: what is the single priority goal for the business in the next 12–18 months?
• Clients typically offer 5–10 goals; the work is ruthlessly choosing one
• Examples: reduce churn from 20% to 12%, increase qualified leads from X to Y
• Everything downstream is scoped to this one goal

Step 2: Identify the priority bottleneck

• Ask: what bottlenecks are preventing you from achieving that goal?
• Multiple bottlenecks will exist; rank them by which, if removed, moves you furthest toward the goal
• A secondary ranking criterion: ease of automation with today's AI tools
• Focus on the bottleneck that moves you 80% of the way, not the ones worth 1–2%

Step 3: Map the functions inside that bottleneck

• Break the bottleneck into its constituent functions — the distinct workflows that make it up
• Example: a two-week proposal-writing process breaks into: aggregating spec sheets, repurposing past proposals, calculating overhead and pricing
• This step makes the bottleneck concrete and actionable

Step 4: Map and rank the activities inside each function

• For each function, list every discrete activity ("GoPro on your chest" level of detail)
• Rank activities by impact first: which activity, if automated, most advances the goal?
• If impact is unclear, rank by ease of automation — use AI itself to help rank
• Query multiple high-reasoning models (e.g. Grok, Claude, GPT) with full context and compare outputs
• The ranked list becomes your implementation roadmap