✨Schedule from promptLive
Build a schedule from a sentence
Describe the week in plain English — coverage needs, who's off, who prefers what — and PeopleBridge drafts the schedule. You tweak, you publish.
How it works
Claude Sonnet 4.5 reads your roster, availability windows, time-off requests, and historical preferences. Returns a draft schedule as a structured payload that hits the same validation pipeline your manual schedules do — no special path, no hallucinated shifts.
In practice
"Cover Friday dinner rush + weekend, give Maria Tuesday off, prefer evening folks after 5pm" → 7 days × 14 shifts drafted, 0 conflicts, 0 overtime risk.
⚡Conflict detectionLive
Every conflict caught before you publish
Overlaps, time-off violations, double-booking, coverage gaps, overtime risk. Red, yellow, green — at a glance, before a single notification goes out.
How it works
Runs server-side on every shift insert and update. The same validation pipeline that catches manager mistakes catches AI-suggested mistakes. Conflicts are explained in plain language, not just flagged.
In practice
Pre-publish: "Overlap: Maya R. already on 11a–7p, cannot also take 2p–10p. Overtime risk: Jordan T. at 38h, adding Sun 3–11p pushes to 46h (1.5× wages)."
🎯Smart open-shift rankingLive
Open shifts go to the right person first
Drop an open shift — PeopleBridge ranks candidates by availability, hours worked so far this week, acceptance history, travel distance, and skill match. The most-likely "yes" gets first dibs.
How it works
Explainable scoring: each candidate shows why they're ranked where they are. No black box. You see the inputs, you can override the ranking, you can flip back to FCFS at any time.
In practice
Open shift Fri 2–6 pm → Top candidate: Marcus (rank 1, 92% match) — Available Fri afternoons, 18h this week (capacity), 4/4 last accepts, 2 miles from location.
💰Labor watchLive
Labor cost watch with anomaly alerts
Set a target labor percentage of revenue. PeopleBridge watches it daily against Toast sales data and alerts you when you're trending hot before the pay period closes — not after payroll.
How it works
Daily cron job compares projected hours × wages against actual sales pulled from the Toast integration. Threshold is per-location. Notifications go to the owner via Make.com SMTP + in-app push.
In practice
Watters Creek labor at 34.2% of sales this week vs target 28%. Suggested: drop Sunday 7–11p shift (saves ~$180, low historical traffic in that hour band).
📍Smart clock-inLive
Smart multi-location clock-in routing
Float staff who cover multiple stores get auto-routed to the closest geofence at clock-in. No more "I punched in at the wrong location" payroll fixups.
How it works
On clock-in tap, the device GPS is matched against every store geofence the employee has access to. Closest match wins. Manager override always available. Each punch is tagged with its actual location for cost-center allocation.
In practice
Sarah covers Frisco South + Watters Creek. Punches in at 9:58 AM. GPS shows she's 240ft from Watters geofence (radius 300ft) and 2.3 miles from Frisco. Auto-routed to Watters. No tap required.
🔔Smart re-notifyLive
Smart re-notification — only when it matters
Employees don't see every push notification. PeopleBridge re-notifies only when the message is urgent (shift starts soon, manager approval needed, time-off response) and the recipient hasn't opened the app.
How it works
Presence-aware suppression: if the recipient's tab is focused on the relevant chat or page, skip the push entirely. If they've been inactive 30+ minutes, re-notify once at the urgency threshold. Otherwise, in-app badge only.
In practice
Shift starts in 30 min and Marcus hasn't clocked in or opened the app. → Push + SMS fallback fires once. If he was already in the app on the schedule page 5 min ago, → no push fires.
📄Doc extractionLive
Document extraction on onboarding
New hire uploads a driver license photo and a voided check during onboarding. PeopleBridge extracts the data automatically: name, address, DOB, license number; routing and account numbers from the check.
How it works
Claude vision API reads the document, returns structured fields, populates the W-4/I-9/direct-deposit form. Employee reviews + corrects before submit. Original document stored encrypted in Supabase Storage with masked-display in the UI.
In practice
Three-week paper onboarding chase compressed to a 12-minute mobile flow. Two of every five paper packets had a missing field; AI-extracted packets have under 5% error rate.
⏱️OT predictorLive
Overtime predictor before publish
Pre-publish warning when an employee is on track to cross 40 hours that week — fix it before the schedule goes out, not after payroll runs.
How it works
Sums projected weekly hours across all locations for each employee. Compares against federal + state OT thresholds (currently Texas; CA, NY on Q4 roadmap). Flags borderline cases (38-40h) so you can decide vs. preventing.
In practice
Pre-publish flag: "Jordan at 38h. Adding Sun 3–11p (8h) pushes to 46h → 6h at 1.5× = $87 in OT wages. Move shift to Tue instead? (Tue Jordan currently at 0h)."
📈Demand forecastLive
Demand forecasting from POS history
Pre-publish staffing recommendations based on historical Toast traffic for that day-of-week + hour band. Tells you what you SHOULD staff before you finalize what you ARE staffing.
How it works
Aggregates ~12 months of Toast order velocity by day-of-week × hour × location. Compares forecast to scheduled coverage. Surfaces gaps and over-staffing as soft warnings; you decide.
In practice
"Sat 11a–2p typically sees 47 orders/hr (95th percentile 62). Current staffing: 1 server. Last 4 Saturdays this slot averaged 2.3 servers. Add a second?"
📊Drift detectionLive
Drift detection between Toast & schedule
Spots when actual clock-ins drift away from published schedules — chronic late starts, early-outs, no-shows, ghost punches. Highlights patterns instead of just exceptions.
How it works
Daily comparison of time_entries vs. shifts table. Patterns are surfaced at the employee × day-of-week level: "Marcus is 12 min late on average for Friday opens (8/10 weeks)." Not just "Marcus was late today."
In practice
Weekly drift digest: 3 employees with >10 min average lateness on a specific shift, 2 employees consistently clocking out before scheduled end. Actionable on the spot.
📝Weekly digestLive
AI weekly operations digest
Monday morning summary email: labor % vs. target, overtime trend, top 3 drift patterns, time-off requests pending, anomalies in clock-in or sales. One scroll, no dashboard required.
How it works
Claude synthesizes the previous week's data into a plain-English briefing. Numbers are pulled from the same RPCs that power the Insights page; the AI just narrates them. Always opt-in per owner.
In practice
"Week of May 26: Labor at 29.4% (target 28%, $340 above). Two OT flags hit (Jordan, Maria — Sun). Three open shifts unfilled at McKinney. 4 time-off requests pending your review. Tip pool reconciles to penny."
🌎Auto-translateLive
Auto-translate for multilingual teams
Messages, announcements, and schedule notes auto-translate per recipient's language preference. Manager writes in English; back-of-house reads it in Spanish.
How it works
On-the-fly translation via Claude when the recipient's profile language differs from the message author's. Both original and translated text are preserved; recipients can toggle between them. Included for everyone at no extra cost.
In practice
Owner sends announcement "Tomorrow is Memorial Day — we open at noon" in English. Maria's profile is set to Spanish; her notification reads "Mañana es Memorial Day — abrimos al mediodía." Tap to see original.
AI features FAQ
Is AI really included with no per-prompt billing?
Yes. All AI features are included for everyone at no extra cost. There is no AI tier, no per-prompt billing, and no metered usage. PeopleBridge is one simple plan: free for your first month (up to 10 employees and 1 location), then $1 per active employee per month plus $10 per location per month — everything included.
What AI model does PeopleBridge use?
Claude Sonnet 4.5 from Anthropic for most workflows (schedule-from-sentence, weekly digest, auto-translate). Claude vision API for document extraction. Server-side validation pipelines for conflict detection, OT predictor, and demand forecasting — those use deterministic logic, not LLMs, so they're fast and explainable.
Are AI suggestions automatically applied to my schedule?
No. AI suggests; you are in control of every decision. Every AI output — drafted schedule, conflict resolution, open-shift ranking — surfaces in the UI for you to review, edit, and approve. Nothing is published, notified, or paid out automatically based on AI alone.
Do I need to be technical to use the AI features?
No. Every plan includes a 4-hour customization session with a real human (not a chatbot) who tunes the AI to your business: labor baselines, demand forecasting inputs, document extraction templates, prompt tuning, custom consent forms. After that, AI features just work inside your existing workflows.
Can I trial the AI features?
Yes — your first month is free with no credit card (up to 10 employees and 1 location). All 12 AI features are available from day one, included for everyone at no extra cost. The free month includes the 4-hour customization session (we usually schedule it in week 1).
Where does my data go when I use AI features?
AI calls hit Anthropic's Claude API directly from our backend. Anthropic does not train on API-tier data. Your data does not leave the request-response cycle. Document extraction images are encrypted at rest in Supabase Storage with row-level security and masked display in the UI.