Most firms are stuck in Phase 1: giving people AI chat tools and hoping for the best. Every employee gets a little faster, but the firm itself doesn't change. Phase 2 is different. Processes get redesigned around what AI makes possible. Triggers fire automatically. Institutional intelligence accumulates in a database. The firm stops being a collection of people using AI and becomes an organization that runs on it.
60 workflows. 7 business functions. One slider that shows what happens to a firm's economics across two phases of AI adoption.
Start by entering your firm's numbers below. Then scroll down and drag the slider.
The downside is $20K/year. The upside is a different kind of firm.
Join the AI for Business Owners CourseThese margins assume you adopt while competitors don't. If every firm adopts simultaneously, excess profits get competed away on price. The 30-40% EBITDA margin is a window, not a permanent state. Over time, gains flow to clients as lower prices or to employees as higher comp. First movers get 2-3 years of advantage.
Throughput increases only help if you can sell more work. The revenue upside assumes your pipeline is deep enough to fill freed capacity. A firm already struggling to fill its funnel gets the cost savings but not the revenue lift. Pipeline is the binding constraint, not productivity.
Enterprise software typically sees 30-40% real adoption. We're assuming much higher. The "people actually use it" problem has killed every productivity tool wave since Lotus Notes. AI's conversational interface helps, but culture eats technology for breakfast.
We assume AI-assisted deliverables are at least as good as human-only work. If quality drops, you lose clients despite being faster. One hallucinated number in a financial model erases the whole efficiency gain. The human review loop is non-negotiable.
We assume billing rates stay flat even though work takes less time. If clients realize a deliverable took 4 hours instead of 40, they push back on fees. Firms will need to shift to value-based pricing. The ones who don't will see margin compression, not expansion.
Some industries and clients will resist AI-augmented deliverables or require disclosure. Government, healthcare, and financial services all have data handling constraints. Your most regulated clients may be the last to benefit.
The headcount model assumes natural attrition lets you right-size without layoffs. But attrition is lumpy - the wrong people leave at the wrong time. You might lose a star consultant while the role you want to eliminate stays filled. Workforce planning gets harder, not easier.
We model AI subscription costs ($20-30K/year) but not the real cost of organizational change: training time, productivity dips during transition, morale management, potential departures from people who feel threatened. The hidden costs are real.
The slider implies smooth improvement. Real adoption stalls, accelerates, hits walls, and restarts. Month 3 might be worse than month 1 as novelty wears off and edge cases surface. Plan for a J-curve, not a straight line.
The model treats current tools as the steady state. Anthropic's CEO predicts AI will handle most knowledge work in 1-3 years and envisions "a country of geniuses in a data center" by 2027. If he's right, the "optimistic" scenario might actually be conservative on a 3-year horizon.