AI can do the heavy lifting on CAC analysis, but the channel bets still need a human CFO's call.

The analytical structure — payback period, LTV by cohort, churn by segment — is formulaic and repeatable. But each engagement involves different data shapes, pricing tiers, and business context, so the agent must adapt its approach each time rather than run a fixed script.

Quantitative outputs like CAC payback period have crisp definitions, but 'most profitable cohort' and 'which channels to cut' involve judgment calls about thresholds, risk tolerance, and strategic priorities that aren't fully specified in the task.

The user has the data but hasn't yet provided it to the agent — it must be uploaded or connected. If the data is clean and well-labeled, the agent can proceed; if it's messy, inconsistent, or spread across systems, significant preprocessing is required before analysis can begin.

A miscalculated payback period or a flawed cohort comparison could lead to cutting a profitable channel or doubling down on a money-losing one — real capital allocation decisions with lasting consequences for a 12-person startup where every dollar matters.

The strategic recommendation layer — whether to cut a channel, how much to invest, what tradeoffs to accept — depends on competitive context, team bandwidth, investor expectations, and risk appetite that no agent can infer from 14 months of CAC data alone.