AI can crunch the benchmarking numbers, but the strategic pitch still needs a human.

The analytical structure is consistent — extract metrics, rank, flag outliers — but each engagement involves a different industry context, peer set, and client situation that shifts what matters. It's repeatable in form but not fully plug-and-play in substance.

The quantitative outputs (rankings, growth rates, margin comparisons) have crisp success criteria. The strategic recommendations — where to focus cost control or investment — are inherently judgment calls with no objectively correct answer, making completeness hard to verify automatically.

The user says they've compiled the 10-K filings and internal P&L, so raw data is available, but the agent needs those files in a structured, parseable format and a code-execution environment to process them. If files are PDFs with inconsistent formatting, extraction adds meaningful friction and error risk.

This analysis feeds a client pitch — a miscalculated margin ranking or misattributed revenue trend could embarrass the consultant, undermine credibility, or lead to a flawed strategic recommendation the client acts on. Errors are not easily reversible once the pitch is delivered.

Deciding which competitors are 'gaining share' in a meaningful strategic sense, and translating that into prioritized recommendations for a specific client's situation, requires contextual judgment about industry dynamics, client constraints, and what a mid-market manufacturer can realistically execute. AI can surface the data patterns but not reliably weight their strategic significance.