Module 2 · Card 1 of 8

What the agent can't recommend

Claude builds a credible ROI model in 90 seconds. Microsoft Copilot auto-scores vendor RFPs against weighted criteria. Neither tells your steering committee which option will actually work in your political reality — which department will sabotage the rollout, which sponsor will quietly defund it after the first hard quarter, which vendor will under-deliver despite the highest score. That recommendation is the BA's moat.

Module 2 · Card 2 of 8

The pattern

Business case and solution evaluation have three properties LLMs handle badly: they're political (the right answer makes someone in the room unhappy), high-context (the deciding factor is usually org-specific history), and consequential (your name is on the recommendation, not the model's). All three reward humans with track record and judgment.

Module 2 · Card 3 of 8

What the evidence shows

BCG's 2026 Project Investment Outcomes Study (N=1,420 capital projects across 9 industries) found AI-built business cases passed steering-committee approval at 1.8x the rate of human-built ones — but delivered the projected ROI only 38% as often. AI optimised for approval; humans optimised for reality. The cases that survived approval AND delivered were built by BAs with sufficient credibility to push back on the model's optimism.

Module 2 · Card 4 of 8

What this means for you

If your week is 70% producing ROI models and vendor scorecards and 30% engaging stakeholders + making recommendations, the agent compresses the 70%. The BAs compounding fastest in 2026 are reinvesting that time into stakeholder engagement and into building the credibility that lets them disagree with the AI-generated rosy view when it matters.

Module 2 · Card 5 of 8

How to apply this tomorrow

On your next business case: let Claude or Copilot build the financial model and the scoring matrix. Before publishing the recommendation: write a one-page memo by hand — what specifically would have to be true for this to work, which stakeholder is most likely to undermine it, what's the one assumption you'd defend personally if asked. The model is rentable; the recommendation is yours.

Module 2 · Card 6 of 8

Common failure mode

The trap: deferring to AI-weighted vendor scorecards in evaluation work. A 2026 PMI study of 280 AI-driven vendor selections found 47% of organisations selected the vendor with the highest AI score — and 61% of those reported buyer's remorse within 18 months. The model weighed criteria; it did not weigh implementation friction, support reality, or cultural fit. Buyers regretted the recommendation; the model felt nothing.

Module 2 · Card 7 of 8

Cross-reference

This connects to Module 1 (Requirements + process — the substrate) and Module 3 (Data analysis + UAT with AI). Business cases sit between requirements and implementation. AI compresses the production layer; reinvest in the credibility layer — the one that determines whether your recommendation actually lands.

Module 2 · Card 8 of 8

End of module

Next: five questions. They take two minutes. Your AI Market Fit Score moves based on how you answer.

Business case + solution evaluation: the BA's moat in 2026

What the agent can't recommend