Zero Data Entry
AI & Automation Design — Last Updated: 2026-05-15
The CRM should update itself.
The #1 reason CRMs fail is that nobody fills them in. Field crews don't have time to type between jobs. The software becomes a ghost — paid for, opened occasionally, trusted by nobody.
The solution isn't better training. It's removing the requirement to enter data at all. Every signal the field generates — location, photos, time, audio, receipts — is already enough to keep the CRM current. The tech does the job. The system watches, interprets, and records. The human only touches it when the machine isn't sure.
Automatic Signals & What They Determine
Every action a field tech takes emits a signal. Here's what each signal determines automatically — and what the fallback is when confidence is low.
| Signal Source | What AI Determines Automatically | Fallback (low confidence) |
|---|---|---|
| Phone GPS (live) — Always-on | Which project the tech is currently at. When they arrived. When they left. Total time on site. | GPS matches multiple nearby projects → notify tech to confirm which one. |
| Photo EXIF GPS — Per photo | Which project a photo belongs to. Whether it's inspection, progress, or completion based on project stage. | GPS doesn't match any known project → show 2–3 nearest projects → tech taps one. |
| Photo AI Vision — Per photo | Photo type (roof, interior, exterior, receipt, material). Project stage (before/during/after). Damage or condition visible. Materials present. | Low-confidence classification → tagged as "unclassified" for review. Tech can relabel in one tap. |
| Receipt Photo (OCR) — On demand | Vendor name, amount, line items, date. Cross-referenced with GPS at photo time → matched to nearest active project as an expense. | GPS not near any project (e.g., supply store off-route) → show list of active projects → tech selects one. |
| Geofence Entry/Exit — Always-on | Clock-in when tech enters a job site radius (50–100m). Clock-out when they leave. Time entry created automatically on the project. | Brief entry/exit (under 5 min) → not logged. Ambiguous location → tech gets a nudge: "Were you at [Project Name]?" |
| Video Audio — Per video | Tasks mentioned. Notes and observations. Materials or issues called out. Customer mentions (if customer is present). | Transcription exists but AI didn't extract structured data → raw transcript stored, flagged for review. |
| Calendar / Schedule — Background | Which project a tech is expected to be at right now. Boosts GPS match confidence when tech is on-schedule. | Tech at unscheduled location → prompt to log it as a new visit or associate with a project manually. |
| Tech Identity — Login | Which crew member created every photo, receipt, time entry, and note. Labor cost rate for time tracking calculations. | Shared device → prompt for quick "who's this?" selection at app open. |
| Call Transcripts — Digital Receptionist | Contact details. Project mentioned. Appointment booked. Follow-up required. Personal details to store in relationship memory. | Unknown caller → new contact created. Contractor reviews and merges if duplicate. |
Receipt Photo Capture
Photo a Receipt → Expense Logged Auto + 1-tap confirm. Tech takes a photo of any job-related receipt. System does the rest.
| Step | Action |
|---|---|
| Capture | Tech taps "Receipt" in the app and takes a photo. That's the only required action. |
| OCR | AI extracts: vendor name, date, total amount, line items from the receipt image. |
| GPS Match | System checks GPS at the moment of capture. If within range of an active project → expense auto-assigned. |
| Confirm | Tech sees: "$142.50 at Home Depot → assigned to 123 Oak Street." Tap confirm or select a different project. |
| Logged | Expense appears on the project timeline and rolls up to the project cost summary. No typing. |
When GPS Doesn't Match
Receipt was taken at a supply store between jobs — no nearby project. System shows the tech's 3 most recent active projects. One tap selects. The receipt is still fully extracted; only the project assignment required human input.
GPS-Based Time Tracking
Arrive at Site → Timer Starts Automatically Fully automatic. No clock-in button. No forgetting. Time is captured from the moment they show up.
| Step | Action |
|---|---|
| Geofence In | Tech's phone enters a 75-meter radius around a project address → timer starts silently. |
| Notify | Push notification: "Timer started — 123 Oak Street. Tap to review." One tap if they want to adjust, or just ignore it. |
| Geofence Out | Tech leaves the radius → timer stops. Time entry saved to the project with tech identity, start, end, duration. |
| Labor Cost | Time × tech's hourly rate → calculated and added to project cost automatically. |
Edge Cases Handled
- Drive-by (under 5 minutes) → not logged. Threshold is configurable.
- Lunch break on-site → tech can split or pause from the app. Or just leave it — owner can review.
- Multiple techs at same site → each tracked independently from their own device.
- No GPS signal → last known location held, flagged for review when signal returns.
- Tech forgets to carry phone → manual override always available from the project page.
Automatic Photo Routing
Take a Photo → It Lands in the Right Project Auto + 1-tap fallback. No tagging, no folders. Photos go to where they belong based on where they were taken.
Confidence Ladder
| Confidence | Threshold | Behavior |
|---|---|---|
| High | GPS matches exactly one active project within 50m | Auto-assigned, no prompt. |
| Medium | GPS matches 2–3 nearby projects or confidence is 70–90% | Shows top match with one-tap confirm. "Is this for 123 Oak Street?" Yes / No. |
| Low | No GPS match, or confidence below 70% | Shows list of tech's active projects. Tech picks one. Still one tap. |
AI Vision Classification
After routing, AI analyzes the photo and tags it automatically:
- Stage: Before (inspection) / During (progress) / After (completion)
- Subject: Roof, siding, foundation, interior, material, equipment, receipt
- Condition: Damaged, repaired, new, clean
- Notes extracted: If a measurement or label is visible in the photo, AI reads it
Live Team Location Map
See Every Tech on a Map in Real Time Always-on. Owner or office sees where everyone is without calling anyone.
- Each tech shown as a pin on a map with their name and current project
- Pin color shows status: on-site (green), driving (blue), idle (grey)
- Tap a tech to see their today's timeline: arrived at X, left Y, currently at Z
- Alerts if a tech has been idle at an unknown location for more than 30 minutes
The One-Tap Rule
Every feature follows the same design rule: AI makes the best determination it can. If it's confident, it just acts. If it's not sure, it shows the most likely answer and asks for a single tap to confirm or correct.
There is no scenario where the tech is asked to type anything. The choices are:
- Nothing — AI was confident. It handled it.
- One tap — AI showed its best guess. Tech confirms.
- One selection from a short list — AI wasn't sure. Tech picks from 2–4 options.
No forms. No dropdowns with 30 options. No "please enter project code." The rule is never broken.
What Never Requires Manual Entry
- Time on site — GPS geofence handles it
- Which project a photo belongs to — GPS + AI handles it
- Receipt amount, vendor, items — OCR handles it
- Photo stage (before/during/after) — AI vision handles it
- Tasks from field video — audio transcription handles it
- New contact from a call — digital receptionist handles it
- Project created from new job site photos — GPS + reverse geocode handles it
- Labor cost per project — time tracking × rate handles it
- Which tech was on-site when — phone identity + geofence handles it
Open Questions
- Geofence radius — 50m, 75m, or configurable per project?
- Minimum dwell time before logging a visit (currently proposed: 5 minutes)
- Receipt OCR — build in-house (Claude Vision) or use a dedicated service (Veryfi, Taggun)?
- How does time tracking handle overnight or multi-day jobs?
- Does the owner see live GPS at all times, or opt-in per tech?
- Privacy — do techs know they're being tracked? (Likely yes — document consent flow)
- What's the fallback for techs without smartphones?