Getting Camera Alerts Where You Actually Look: WhatsApp, ERP and the Shift Sheet

An alert nobody opens is not an alert. In an Indian plant, camera events need to land in three places: WhatsApp for the supervisor within seconds, an ERP downtime code for the costing that follows, and the shift handover sheet for the next in-charge. Anything routed to a dashboard alone is usually dead on arrival.

You run 200–1000 workers across two or three shifts in Chakan, Hosur, Sanand or Panipat. Your production head does not open a browser dashboard at 2 a.m. He opens WhatsApp. Any camera-AI system you buy has to meet that reality, or the capex buys you a screen that gets checked on the day the auditor visits.

Key points

• The channel decides whether the alert works. A machine-stopped event that reaches the line supervisor's phone in under a minute has a chance of converting to recovered production. The same event sitting in a web dashboard usually converts to nothing. Design the routing before you argue about detection accuracy.
• WhatsApp economics favour reply-driven flows. On the WhatsApp Business Platform, utility templates delivered inside an open customer service window are free, and all marketing template messages are charged — see Meta's pricing documentation. India moved to local INR billing on 1 January 2026, with eligible accounts required to migrate by 31 December 2026. Alerts should be utility-category, never marketing.
• ERP is where the money argument gets settled. A camera event only becomes a rupee figure once it carries a downtime reason code your ERP already knows. If a line contributing ₹40,000 of gross margin per hour loses 25 minutes to an unlogged changeover overrun, that is roughly ₹16,700 of margin that never appears in any MIS unless the event is coded. Substitute your own line rate — the illustration is arithmetic, not a benchmark.
• The shift sheet is the audit trail nobody replaces. WhatsApp gets deleted, ERP gets aggregated. The handover sheet — paper or digital — is what the next in-charge actually reads and what a customer auditor asks to see.
• Who receives footage is a legal question now. The Digital Personal Data Protection Rules, 2025 were notified on 13 November 2025 on a phased timeline: the Data Protection Board provisions took effect immediately, the consent-manager framework from 13 November 2026, and substantive data-fiduciary obligations — notice, security safeguards, rights handling, breach notification — from 13 May 2027. A worker's face forwarded into a 60-member WhatsApp group is personal data leaving your control.
• The old statute is gone. The Occupational Safety, Health and Working Conditions Code, 2020 came into force on 21 November 2025, consolidating thirteen central labour laws including the Factories Act, 1948. Registers, returns and inspection formats you built around the old Act need re-mapping.
• Three routes, three latencies. Seconds for safety and stoppage, hours for cost coding, one per shift for the narrative. Mixing them is why WhatsApp alert groups get muted in week three.

The three-lane routing model

Most vendors sell you one firehose. Split it into three lanes with different speeds and different audiences.

Lane 1 — act now (seconds, WhatsApp). Machine stopped past threshold, person inside a guarded zone, gate movement outside schedule, fire-path blocked. Goes to the shift supervisor and the security in-charge. One line of text, one still frame, one clip link behind authentication.

Lane 2 — code it (hours, ERP). Every stoppage over your minimum duration gets a reason code and lands against the work order. Nobody gets pinged. This is what the production and finance review consumes on Monday.

Lane 3 — hand it over (once per shift, shift sheet). A short summary: total stopped minutes by line, top three causes, any safety event, any open action.

If an event appears in all three lanes it should appear once in each, not three times in WhatsApp.

The routing matrix

WhatsApp: use it as a channel, not a filing cabinet

Three rules that separate a group people read from one they mute.

Cap the volume. In our own deployments, past roughly eight to ten alerts a shift the group gets muted. If your system cannot suppress duplicates and repeats of the same stoppage, it will train your supervisors to ignore it.

Make every alert actionable. "Anomaly detected on Camera 7" is noise. "Line 2 press stopped 6 min — no operator at station since 14:12" is an instruction.

Never put the evidence in the group. Send a link that opens an authenticated viewer with an access log. WhatsApp forwarding is one tap; once a clip of a worker is in a group, you have lost control of it, and the notice-and-purpose obligations under DPDP will not stop applying because it was convenient. This is why the worker CCTV privacy notice matters more than most owners expect.

On cost: keep alert templates in the utility category. Per Meta's pricing rules, utility templates delivered inside an open customer service window are not billed, so a plant whose supervisors actually reply — each reply keeping the window open — pays far less for its alert traffic than one that broadcasts one-way. Confirm current rate cards with your BSP before budgeting; Meta revises pricing periodically.

ERP: the downtime code is the whole point

Camera AI that does not write into your ERP produces opinions. Camera AI that writes a downtime reason code produces evidence.

Practical approach for a mid-size plant on SAP B1, Tally-plus-MES, or a local ERP:

1. Take your existing downtime reason master. Do not invent new codes — use the ones your supervisors already tick.
2. Map each detectable camera event to exactly one code. Machine idle with operator absent → manpower. Machine idle with operator present → material or setup. Unmapped events go to a single "uncoded" bucket, and that bucket's size is your honest accuracy metric.
3. Post hourly, not in real time. Real-time ERP writes create reconciliation arguments; hourly batches with a supervisor override window do not.
4. Keep the override. The supervisor must be able to recode an event with a reason. If he cannot, the data becomes contested and the whole system loses credibility inside a month.

The rupee logic is simple, and you should run it on your own numbers rather than a vendor's. Take the gross margin your line earns per running hour, multiply by the stopped minutes the cameras surface that were previously invisible, then discount hard for the share of those minutes you can realistically convert — response capacity, not visibility, is usually the binding constraint. The full method is in does camera AI pay for itself.

The shift handover sheet

The handover sheet is the one artefact that survives. Keep it to a single page: stopped minutes by line, top three causes with minutes, safety events with status open or closed, and actions carried to the next shift.

Two design points. First, the camera system should fill the numeric fields and leave the judgement fields blank for the outgoing in-charge — an auto-generated narrative gets signed without being read. Second, the sheet must reconcile with the ERP totals. If the sheet says 62 minutes and ERP says 41, your supervisors will trust neither.

Governance: who is allowed to receive footage

This is the section vendors skip.

Under the OSH Code framework in force since November 2025, your statutory obligations are registers, returns, safety committees and working conditions — not the surveillance of individuals. DPDP obligations run the other way: personal data must be handled for a stated purpose, with notice, with reasonable security safeguards, and erased when the purpose is over. A face on a shop-floor camera is personal data.

Build a written recipient list before go-live:

• Aggregate counts (stopped minutes, PPE compliance percentage) — any supervisor, any group. No personal data.
• Identifiable stills or clips — a named, short list. Production head, security head, owner. Logged access, no forwarding, no group distribution.
• Anything involving a woman worker on night shift — tightest tier. Under the OSH Code, women are entitled to be employed in all establishments for all types of work; where the work is hazardous or dangerous, the appropriate government may require the employer to provide adequate safeguards before employment (PRS legislative summary). Night working for women is permitted subject to the worker's consent and to conditions on safety, holidays and hours prescribed by your state — check your state's notification, as the conditions differ. In practice "safeguard" means escort confirmation and lighting compliance recorded in the register; it does not mean a clip of her walking to the gate circulating in a supervisors' group. Our women's night-shift retention checklist sets out the storage side.

Set retention deliberately, not by whatever the NVR defaults to. Longer retention is more liability, not more safety.

Where this does not help — and where it can hurt you

Be clear-eyed.

Routing does not fix a plant with no response capacity. If maintenance takes 45 minutes to reach a line regardless, faster alerts only produce better-documented waiting. Fix response first.

Alerts can manufacture conflict. Push identifiable footage into a supervisors' group and you risk a labour dispute long before you get a productivity gain. In plants with active unions in Maharashtra, Tamil Nadu and Haryana, this is a common way camera projects die.

Cameras cannot code intent. They see a stopped machine. Whether it stopped for material shortage or for a legitimate quality hold is a human judgement. Systems that claim otherwise are inflating their downtime savings.

More alerts is not more control. The failure mode is rarely a missed event; it is thirty events per shift, all muted. Start with two alert types, prove response, then add.

WhatsApp is not evidence-grade. Deleted messages, forwarded clips, no access log. For anything that might become a disciplinary matter or an insurance claim, the authenticated viewer and the retained recording are the record — WhatsApp was only the doorbell.

FAQ

Should alerts go to a WhatsApp group or to individuals? Groups for aggregate and machine-level alerts, individuals for anything identifiable. A group of forty means forty uncontrolled copies of any image you send. Keep the group to the shift team on duty, rotate membership with the roster, and route person-level events to two or three named recipients through a logged portal instead.

What does it cost to send factory alerts over WhatsApp in India? Under Meta's pricing rules, utility template messages delivered inside an open customer service window are free, while all marketing templates are charged. India moved to local INR billing on 1 January 2026, with migration of eligible accounts required by 31 December 2026. If supervisors reply to alerts, most traffic stays inside free windows — but confirm current rates with your provider, since Meta changes them.

Which ERP fields should camera events write to? Only three matter at the start: work order or line ID, downtime reason code from your existing master, and duration in minutes. Post hourly with a supervisor override window. Resist adding fields until the uncoded-event bucket has shrunk to a share of stopped minutes you are willing to defend in a review — set that threshold yourself before go-live.

Does the OSH Code 2020 require me to send camera alerts anywhere? No. The Code, in force since 21 November 2025, governs registers, returns, safety committees and working conditions — not alert routing. No central law makes shop-floor cameras themselves a statutory requirement; obligations that look like a mandate usually come from a customer contract, a tender condition, or a state-level night-shift condition. See is CCTV mandatory in Indian factories.

When do DPDP obligations actually bite? The DPDP Rules, 2025 were notified on 13 November 2025 in phases: Data Protection Board provisions immediately, consent managers from 13 November 2026, and the substantive data-fiduciary duties — notice, security safeguards, rights handling, breach notification — from 13 May 2027. Build the recipient list and retention policy now; retrofitting governance into a live routing setup is far harder.

Do I need cloud, or can this run on my existing NVR? Alert routing needs outbound internet from the plant, but footage can stay on-premise with only stills and links leaving. That hybrid keeps recordings inside your gate while alerts reach phones — the trade-offs are laid out in NVR vs cloud VMS.