Who this is for: Clinic admins and staff preparing to go live with Inbox Agent.
What this covers:
- How Inbox Agent learns to recognize your clinic’s documents
- Why document type consolidation matters
- The two paths clinics take to improve accuracy over time
Document Training: What to Expect
(Inbox Agent — Preparing for Success)
Why this matters:
Document classification is the single biggest driver of automation accuracy in Phelix’s Inbox Agent.
When the AI correctly identifies what a fax is — a referral, lab, consult note, or imaging report — everything else (patient matching, provider assignment, filing) happens seamlessly.
1. Why Document Training Exists
Inbox Agent reads every fax through two core steps:
- Extract information – Pulls key details like patient, provider, and date.
- Classify the document – Determines what kind of fax it is (Referral, Lab, Consult Note, Imaging, etc.).
Extraction works immediately.
Classification works from day one too — thanks to an out-of-the-box, pre-trained classifier built on millions of healthcare documents.
Training simply helps refine this model so it better understands your clinic’s unique document types.
💡 Think of it like speech recognition:
The system already understands English (out-of-the-box).
Training teaches it your accent — how your clinic labels and organizes files.
2. What Works on Day One
Inbox Agent comes ready with a strong, general-purpose model trained on millions of healthcare documents. From day one, it can extract data, detect patterns, and classify most common fax types.
That means you’ll start seeing time savings immediately — even before your model is trained on your clinic’s data.
✅ Works out of the box
- Data extraction: Pulls patient names, providers, referral reasons, and dates automatically.
- Worklists: Every incoming fax appears in a central queue with extracted data visible for review.
- Pre-trained classification: If your EMR uses a short, clear list (usually <50 document types), the AI recognizes and categorizes accurately from day one..
⚠️ Note:
The classifier always works. Training only improves its baseline accuracy.
The more consistent your document types, the faster it learns and the more confidently it can automate.
3. Two Paths to Get Started
Every clinic starts with the same intelligent base model.
Where you go from here depends on how your EMR data is structured.
| Clinic Setup | Recommended Path | What Happens |
| < 100 document types, clearly labeled | Smart Start (No Training) | Inbox Agent can recognize and file right away. |
| 100–500 document types, well organized | Train First (Baseline Training) | We import examples from your EMR to create an initial model. |
| 50–500 document types, messy/uneven | Train First + Usage Learning | Model starts baseline training and improves as your team uses Inbox Agent. |
| 500+ document types | Consolidation Required | Your EMR list must be cleaned up before training. See: [Document Type Consolidation]. |
💡 Tip: If your EMR list looks long or repetitive, start with consolidation before any training. It’s the single biggest predictor of long-term automation success.
4. Path A — Smart Start (No Training)
If your EMR already uses a simple, consistent set of document types, you can begin immediately.
What you’ll see:
- Inbox Agent automatically classifies new faxes into categories like Referrals, Labs, Consult Notes. etc.
- You’ll review and confirm extracted data before filing.
- Automation features (like Auto-Save or Auto-Referral) can be enabled once accuracy stabilizes.
Maintain stability:
Avoid renaming or deleting document types once you go live. Adding new types is fine — they’ll start being recognized gradually through usage.
Next step:
→ When you’re ready to start automating, see Auto-Processing.
5. Path B — Train First (Custom Model Training)
If your clinic’s EMR has many overlapping or confusing document types, Inbox Agent needs examples to learn from.
That’s where document training comes in.
Step 1: Consolidate your list
Group similar document types together (e.g., “Referral – Dr. Smith,” “Referral – Cardiology,” → Referral).
Clear, broad categories like Referral, Labs, Imaging, Consult Notes train faster and perform better.
→ Learn more: [Document Type Consolidation]
Step 2: Train your model
We import real documents from your EMR to teach Inbox Agent what each document type looks like.
Each example helps the system learn visual cues (format, keywords, structure) that define your clinic’s workflow.
You can train in three ways:
- Baseline Training: Uses EMR examples during setup to create your clinic’s first model.
- Usage-Based Learning: The model improves automatically from your team’s filing and corrections.
-
Batch Uploads: Upload sets of 30+ examples per document type to accelerate learning.
→ Learn more: Training Types (Baseline / Usage / Batch)
Step 3: Review and validate
Once the model is trained, you’ll start seeing higher confidence scores and faster routing.
Low-confidence documents stay in the worklist for manual review — keeping accuracy high while the AI improves.
6. Keeping Your Model Healthy
To protect accuracy after training:
- Don’t remove or reorder trained document types.
- You can add new types anytime; they’ll be recognized after retraining or sufficient usage.
- Rename safely: Only if the meaning stays the same (e.g., “Referral Letter” → “Referral”).
- Retrain periodically: Every few months or when many new examples have been added.
7. Related Guides
- Document Type Consolidation — How to simplify and clean your EMR document list before training.
- Training Types (Baseline / Usage / Batch) — How training works and when to retrain.
- Auto-Processing — Once trained, how Inbox Agent can automatically save, chart, or route your faxes.
8. FAQs
Do we need to train before going live?
Not always. Smaller clinics with clean document lists can often start using Inbox Agent immediately.
How long does training take?
Baseline training usually takes 1–3 days once documents are imported. Usage-based improvements continue over time.
Can we train ourselves?
Yes — with the self-serve portal, you’ll be able to upload examples, request retraining, and monitor results directly.
When should we retrain?
After you add new document types or notice frequent low-confidence results for a specific category.
In Summary
- Inbox Agent works out of the box — data extraction starts day one.
- Document classification is what unlocks automation.
- The cleaner your document types, the faster and more accurate your results.
- You can train once, learn continuously, and grow accuracy over time.