AI ROI Calculator — Human vs AI + HITL

Should you optimize the human process, or bring in AI with a human-in-the-loop? Every input is a lever. Move a slider and the risk-adjusted P&L, the break-even projection, and the recommended human/AI split all recompute live. Same math as the Python engine (roi_tracker).

pricing: editable · auto-filled from litellm when the engine runs

Inputs

The CFO interview, as sliders. Type an exact number in any value box.

1 · Current human process (baseline)

What the work costs today, with people - and how much of it there is.

Fully-burdened operator rate x minutes per task. The number every alternative must beat.
How many transactions per month. Scales every per-unit number.

2 · Optimize-the-humans alternative

Before buying AI: how much cheaper could the human process get with lean / tooling, and what would that cost to do?

Realistic efficiency gain from redesigning the manual process.
Cost to redesign / retrain / re-tool the human process.

3 · Model pricing & routing (AI path)

Live token prices per 1M tokens. Router sends simple work to the cheapest fallback, complex work to the cheapest flagship. ● = auto-selected

ModelInput $/1MOutput $/1M

4 · Task mix & token size

Share of traffic hard enough to need a frontier model.

5 · Infrastructure & build cost (AI path)

One-time build spend. Drives the break-even projection.

6 · Risk (the Hallucination Tax)

AI is not free of consequence. These deduct from AI savings.

Damage of one bad hallucination or compliance failure in production.

7 · Process sensitivity (drives the split)

How error-costly and judgment-heavy is this work? High sensitivity keeps more of it human, even when AI is cheaper.

Share of work to keep human when AI is cost-favorable.

8 · Qualitative measures (Human ↔ AI)

Beyond cost. Slide toward whoever is stronger on each dimension: left = Human wins, right = AI wins.

Risk-adjusted P&L

Live. Recomputes on every change.

Human vs AI + HITL — cost per transaction

Three ways to run the same process. Lowest bar wins.

Break-even projection

Cumulative cash vs staying all-human. Where a line crosses $0, that investment has paid back.

Go AI + HITL (pays back CapEx) Optimize humans instead

Recommended approach

Where to draw the human / AI line, given cost and sensitivity.

Qualitative verdict

Cost is one axis. These are the rest.

Executive report

AI ROI Calculator · browser mirrors python -m roi_tracker · MIT