From Fax to Kroll in 30 Seconds: How Multi-Agent AI Handles Canadian Prescriptions

A fax arrives at a Canadian pharmacy. It contains a prescription — possibly handwritten, possibly a photocopy of a photocopy, possibly with a correction written in the margin in a different pen. A pharmacist or technician needs to take that document and get an accurate entry into Kroll before the patient arrives at the counter.

In a manual workflow, this takes three to five minutes. In an automated workflow with AutoRx, it takes under thirty seconds. Here is exactly what happens.

The problem with fax prescriptions

Fax prescriptions are notoriously inconsistent. They come from hundreds of different prescriber offices, on dozens of different prescription pad formats, with handwriting that ranges from typed to illegible. They arrive skewed, with toner marks obscuring key fields, or with handwritten corrections that override printed text.

The obvious solution is OCR — optical character recognition, which converts the fax image into text. OCR has been used in pharmacy software for years. But OCR alone is not enough.

OCR reads what is on the page. It does not know what is in your formulary. It does not know what the patient was dispensed last month. It does not know that when your fax says “atorva 20,” there are three possible DINs it could mean — and that your pharmacy carries two of them, but the patient’s prior fill was with the third from a different pharmacy.

A system that relies on OCR alone will produce a text extraction. What happens with that text extraction depends on downstream logic that, in most systems, is limited.

What multi-agent AI does differently

AutoRx uses a multi-agent pipeline. Multiple AI systems work together on each prescription, each responsible for a specific part of the process. Here is the sequence:

Step 1: Document processing. The incoming fax image is de-skewed, normalized, and processed for extraction. Vision AI — not just OCR — reads the document. Vision AI can interpret documents the way a human does: reading context around fields, recognising that a crossed-out value was replaced, understanding that a handwritten note in the margin is a dose adjustment and not random notation.

Step 2: Field extraction and normalization. Drug name, strength, quantity, sig, prescriber, and patient identifiers are extracted and normalized. Ambiguous fields are flagged with a confidence score — so the system knows what it is certain about and what needs more context.

Step 3: Kroll context read. Before any write decision is made, AutoRx reads from Kroll: the patient’s profile, their drug history (recent fills, therapy context), your drug catalog for the products you carry, and your mix catalog if the prescription is a compound. This is the step that most automation systems skip — because it requires native Kroll integration, not screen automation.

Step 4: DIN resolution. Using the extracted fields combined with the Kroll context, the agent selects the correct DIN. It uses your formulary first — picking from products you carry, with the patient’s history as a tie-breaker. If the product is not in your catalog, it falls back to the Health Canada Drug Product Database (DPD). The result is a specific DIN selection, not a text string left for staff to match manually.

Step 5: Write to Kroll. The validated entry is submitted to Kroll through the native integration, not through the UI. The write is queued and executed asynchronously, with retry logic for transient errors.

Step 6: Completion and exception routing. If the write succeeds, a completion webhook fires to your systems. If the write fails permanently — a prescriber not found, a DIN not in catalog — the exception is routed to your dashboard immediately, with the original document attached and a clear failure reason. Your staff resolves it in seconds.

What happens on exceptions

No automation system is right 100% of the time. AutoRx targets a 95%+ first-attempt success rate. What happens to the remaining cases matters as much as the success rate.

When AutoRx cannot confidently complete a write, it does not guess. It escalates — immediately, cleanly, with everything your staff needs to resolve the exception. The original fax image, the extracted fields, the Kroll context that was read, and the specific reason for the failure are all visible in the dashboard. Your technician does not need to re-read the fax from scratch; they pick up where the system left off and complete the entry in seconds.

This is a fundamentally different model from automation systems that either guess on low-confidence inputs (and introduce errors) or fail silently (and leave prescriptions in a queue with no clear status).

The 30-second number

From the moment a fax arrives in the AutoRx queue to the moment the Kroll write confirmation fires, the average is under thirty seconds. During that time, no staff member touches the keyboard. The prescription is in Kroll and ready for the dispensary workflow — accurately, with the right DIN, with the patient’s history taken into account.

At a pharmacy processing fifty faxes a day, that is two-and-a-half hours of manual entry time reclaimed per day. Every day.

What this means across your pharmacy’s week

The compounding effect of 30-second processing becomes clear when you look at weekly numbers. A pharmacy receiving 250 faxes a week — roughly 50 per day — would spend 12–20 hours on manual entry at 3–5 minutes per prescription. At 30 seconds per prescription with automation, that same volume takes just over 2 hours of total machine time, with your staff handling exceptions only.

For the exceptions — say a 4% rate on 250 faxes — that is 10 prescriptions per week that need human review. At 1–2 minutes each for an informed exception resolution (where the system shows you the original document and the specific problem), that is 10–20 minutes of exception handling per week. Your team goes from 12–20 hours of repetitive entry to under 30 minutes of active decision-making.

The time does not disappear — it gets reallocated. That is the operational case for automation: not that your team does less, but that what they do is clinical, not clerical.

See it in action — book a demo.