Rung 1 — Aware: leadership has heard the terms; nothing is deployed. Risk: buying tools to feel progress. Right move: quantify one process leak.

Rung 2 — Experimenting: individuals use ChatGPT-style tools personally; nothing is governed. Risk: shadow AI with company data. Right move: a usage policy and one sanctioned pilot.

Rung 3 — Piloting: one process has a bounded, measured AI pilot with human oversight. This is the first rung that creates business value. Right move: measure against a baseline, publish internally.

Rung 4 — Operating: AI-assisted processes run in production with owners, audits, and training. Right move: scale sideways to adjacent processes, not upward into complexity.

Rung 5 — Compounding: each new process reuses the governance, data plumbing, and trust built by the last. This is where the margin gap over competitors becomes structural.

Most companies we meet are on rung 1 or 2 and believe they're behind. They aren't — rung 3 is reachable in under a month with the right scoping.

Filter 1 — Volume: the process repeats at least weekly. One-off tasks never repay implementation effort.

Filter 2 — Describable: an experienced person can explain how they'd do it to a new hire. If the rules can't be spoken, AI can't learn them reliably yet.

Filter 3 — Painful: someone senior already complains about it. Pilots need an internal sponsor who feels the leak.

Filter 4 — Measurable: you can state today's cost in hours or rupees. Without a baseline, success will be a matter of opinion — and opinions kill rollouts.

Score your candidate processes 0–2 on each filter. Anything under 6/8 is a second pilot, not a first one. Resume screening, invoice processing, and support triage routinely score 7–8, which is why they're the classic entry points.

1. 'What's the baseline, and who measures it?' — If they don't insist on measuring your current state first, their ROI claims are decorative.

2. 'Show me a decision your system got wrong, and what happened next.' — Mature vendors have error-handling stories; weak ones have only demos.

3. 'Who is the named human accountable for each decision?' — 'The model decides' is a governance failure, not a feature.

4. 'What happens to our data — drawn as a diagram?' — DPDP and your legal team both need this; hesitation is your answer.

5. 'What would make you tell us NOT to buy?' — Vendors with no disqualifying criteria are selling licenses, not outcomes.

6. 'Can our team run this without you in six months?' — Dependency by design is a business model; make sure it isn't theirs.

7. 'Which claim of yours can we audit?' — Published methodology separates evidence from marketing.

Stage 1 — Inbox hiring: resumes in email, decisions by memory. Symptom: every recruiter has a different bar. Cost: quality depends on luck and stamina.

Stage 2 — Tracked hiring: an ATS stores applicants, but screening is still manual reading. Symptom: 'we have the data' but decisions haven't changed. This is where most mid-size companies sit.

Stage 3 — Assisted hiring: every applicant is scored against the role with explainable criteria; humans decide from ranked, reasoned shortlists. Screening time collapses; consistency becomes structural.

Stage 4 — Governed hiring: KPIs (time-to-shortlist, submittal-to-interview, screening hours) are baselined and reported; every decision is auditable. This is what enterprise procurement means by 'responsible AI in hiring'.

Moving one stage typically takes weeks, not quarters — and stage 3 is where the measurable ROI lives. That's the move MinMaxHR and CandidRanker were built for.