Support that already knows the worker's shift, attendance and pay before they finish typing
Frontline workers leave for the agency that answers. Most support today is a shared phone, a WhatsApp group and someone's memory. Worker Care answers instantly, in the worker's language, from the records you already hold, and hands real problems to a human with the whole story attached.
What this work looks like today, and what changes.
Today, per worker question
- Worker messages the group or calls the office: "where is my shift tomorrow?" or "my pay is short".
- Coordinator is on another call. The message waits, sometimes until the next day.
- Coordinator opens three systems to find the worker, the shift and the pay line.
- Half the questions are the same five questions. Each is answered from scratch.
- A pay dispute is a screenshot and a promise to "check with payroll".
- The worker takes a shift from a competitor while waiting.
With Worker Care
- Worker messages on WhatsApp in their language. The agent replies in seconds with their next shift, site, start time and how to get there.
- Common requests are resolved outright: resend shift details, update availability, upload an expiring document, confirm a payday.
- A pay query is answered from the actual pay lines and attendance records. If there is a discrepancy, a case is opened with the evidence attached.
- Anything sensitive, unusual or emotional is escalated to a named person, with the transcript and the records already on screen.
- Every conversation feeds a weekly view: top issues by site and client, so you fix causes, not tickets.
- Workers who go quiet are noticed and checked in on before they leave.
Step by step: what goes in, what the automation does, where a person decides.
Green steps run on their own. Amber steps are where a human approves or decides, by design.
WhatsApp, SMS or voice, in any language. "Where is tomorrow?" "Pay is short." "Can I change my availability?"
Next shifts, site details, attendance records, pay lines, documents on file, open cases. From your systems, at the moment of the question.
AutomatedThe agent replies with specifics: the shift, the site, the pay line, the document that expires. It never invents an answer it cannot see in the records.
AutomatedResend details, log availability, accept a document upload, confirm a payday, register a late notice. Each action is logged against the worker.
AutomatedPay discrepancies, disputes, distress, anything outside policy: a named human gets the case with transcript, records and a suggested next step.
Human handlesFaster answers, fewer repeat contacts, and a weekly view of what workers actually struggle with, by site and client.
One screen, three panes: the input, the reasoning, the result.
- Worker
- Anjali Nair · active since Mar 2026
- Next shift
- 2 Sep · 08:00–16:00 · Kochi Site 2
- Pay Aug
- ₹18,420 · 22 shifts · paid 31 Aug
- Attendance
- 19 Aug: check-in 07:58, check-out 16:03
- Timesheet
- 19 Aug recorded 6 h · mismatch
- Documents
- Bank verified · ID expires Dec 2026
The agent saw the check-in and check-out times and the 6-hour timesheet line. That is why it could explain the difference instead of promising to check.
- Case #2291 opened · pay discrepancy · evidence attachedhuman
- Assigned to Meera (payroll) · due tomorrow 18:00SLA
- Shift details resend scheduled 06:00auto
- Worker sentiment: neutral → satisfiednoted
A composite conversation. The agent answers from records it can see and opens a case with evidence when the records disagree. A person owns the correction.
Where the advantage comes from, step by step.
| Step | Today, by hand | With ShiftMargin | Why it matters |
|---|---|---|---|
| First response | Hours, or the next day. | Seconds, at any hour, in the worker's language. | Workers choose the agency that answers. |
| Knowing the worker | Coordinator opens three systems. | Context pulled automatically: shifts, attendance, pay lines, documents, open cases. | Every answer is specific. Nothing is "let me check". |
| Common questions | Answered from scratch, every time. | Resolved by the agent: shift details, availability, document uploads, paydays, late notices. Logged. | Roughly two-thirds of contacts never need a person. |
| Pay disputes | A screenshot and a promise. | Answered from pay lines and attendance records. A discrepancy opens a case with evidence attached and an owner. | Pay questions are the ones that decide whether a worker stays. |
| Escalation | A forwarded message with no context. | Transcript, records and a suggested next step handed to a named person with an SLA. | The human starts at the solution, not at "who are you?". |
| Learning | Nobody sees the pattern. | Weekly view of top issues by site and client. Silent workers flagged for a check-in. | You fix the cause at the site, not the ticket. |
The components, in plain language, and why each one matters to you.
Everything here is chosen so your ops team can maintain it after we leave. The only pieces we keep maintaining are the ones that move money.
Meta Cloud API via 360dialog or Gupshup. Service replies inside the 24-hour window become billable from October 2026; the run-cost figures already assume it.
A Haiku-class model answers from retrieved records only. A five-turn conversation costs under a cent. If the records do not contain the answer, it says so and escalates.
Scheduling, attendance, payroll lines and documents via API or nightly sync. Nothing is copied that does not need to be.
Resend details, update availability, accept uploads, open a case, schedule a reminder. Each action is a logged function, not a free-form capability.
Escalations land where your team already works, with the transcript and records attached and an SLA clock.
Top questions by site, client and language. Sentiment trend. Workers gone quiet. The report your ops lead reads on Monday.
What stays human, and what happens when something goes wrong.
Any sign of distress, harassment, injury or a safety issue is routed to a named human immediately, with the transcript. The agent does not counsel.
If the records do not hold the answer, the agent says so and escalates. It never guesses a payday or a start time.
The agent sees only the worker it is talking to, verified by number and a challenge where needed. Conversations are stored under your data processing agreement, in region.
The office number and the group still work. If the agent is down, workers reach the same people they reached before.
What it costs by hand, what it costs to run, and where the margin goes.
per contact in coordinator time at typical loaded rates, most of it spent finding information. Plus the shifts lost while workers wait.
per conversation in message and model cost. About $110–400 a month for 2,000 workers, depending on country mix.
The number that matters most is worker retention by site after go-live. It moves slowly and it is the one that pays for everything else.
Six weeks to production, behind a manual fallback.
Sample two weeks of worker contacts: what they ask, how long they wait, how many are repeat questions, how pay queries are resolved today.
Retrieval from your systems, answer templates per language, the action list, escalation routes and SLAs, the weekly insight view.
The agent answers alongside coordinators for one site or client. Every escalation and every answer is reviewed.
Contacts resolved without a person, first-response time, pay cases closed on time, coordinator hours redirected, retention by site.
Before they let this near the business.
Which languages does it support?
The workforce's languages, including the ones your coordinators do not speak. Hindi, Malayalam, Arabic, Swahili and Zulu are common asks. We pilot-test any language that matters to you before committing, because nobody publishes per-language accuracy.
Will workers know they are talking to an AI?
Yes. The agent introduces itself as an assistant, and a worker can ask for a person at any time and get one, with the conversation attached.
Can it change pay or mark attendance?
No. It can explain a pay line and open a case with evidence. Changes to pay, attendance or status are made by a person, through your existing process, and logged.
Do we need a helpdesk tool?
Not necessarily. Escalations can land in the tool you have or in a simple internal review screen we set up. What matters is a named owner and an SLA on every case.
Start with your number, not our demo.
Bring the volumes and the time your team spends on this workflow today. In thirty minutes we will tell you whether it is your most expensive line, and if it is not, which one is.
Book a free 30-minute teardown →