AI in Compliance: Let It Read, Let People Decide

1. The Promise and the Pitfall

Indian employers face a stream of paper: notices from provident fund and ESIC offices, state circulars, gazette notifications, format changes. Reading all of it, matching it to the right sites and drafting responses is slow and tiring work. It is also work where language models and document recognition are genuinely useful.

The pitfall is the leap from "can read" to "can decide". A tool that reads a notice accurately is helpful. A tool that replies to a regulator unsupervised is a risk.

2. What AI Reads Well

• Scanned notices. Pulling the allegation, the section cited and the reply deadline from a photographed letter.
• Circulars and amendments. Matching a new instruction to the companies, states and registers it affects.
• First drafts. Producing a structured reply and an evidence checklist from a template and your records.
• Patterns. Spotting a contractor whose remittance proof is slipping, before the gap becomes a notice.

In practice the saving is in the first few hours. One platform, Iztty, reports an average of 4 hours against 3 to 5 days manually for drafting a notice reply.

3. What People Must Decide

• How an ambiguous provision applies to your arrangement, for example whether a contractor relationship is genuine.
• Whether to concede, contest or seek a hearing.
• What goes out under the company's name and the signer's DSC.
• When a rule change is actually operative for you. The four Labour Codes took effect on 21 November 2025, with detailed rules still being notified, so timing questions are real.

4. Designing the Handoff

The best set-ups put a person at every point where something leaves the building. AI prepares, a Checker reviews and may reject with remarks, a DSC holder signs. Each step is recorded. That pattern keeps the speed and removes the risk.

5. Questions for Any Vendor

• Show me a real notice being read, not a demo file.
• Where does a person review before anything is enforced or sent?
• What happens when the AI is unsure?
• Is there an audit trail of what was suggested and what was approved?

The Iztty guide to regulatory software solutions goes through the wider checklist.