Industries/ Healthcare/ Referral and intake reading
Referral letters read on arrival, with doubts sent to a person
Referral letters, insurer authorisations and scanned forms are read as they arrive, their administrative details extracted and matched to the right patient. Anything uncertain waits for a person, and the clinical content is left for clinicians to read in full.

- Used by
- Reception, the referrals coordinator and the practice manager
- Takes the place of
- Retyping referral letters and insurer forms at the end of the day
- Connects to
- Your referrals inbox, scanner, practice system and record
- Built with
- Our document intelligence service
Where it hurts today
A normal week, before it exists
Referrals arrive in a shared inbox as PDF attachments, faxes turned into emails, insurer portal downloads and letters that came by post and were scanned at the desk. Each one is opened, read and its details typed into the practice system by hand: name, date of birth, address, phone number, referrer, registered GP and insurer. On a busy day the typing waits until the last patient has gone.
Mistakes are small and expensive. A date of birth typed wrong creates a second record for a patient who already exists, and their history is split until someone notices. A letter filed against the wrong patient is an incident. A referral that arrives on Friday afternoon may not be opened until Monday, and nobody can say for certain who has looked at it.
Insurer authorisations bring their own problems. Each one says how many sessions are covered and until when, and that information lives in the letter, not in the diary. When sessions run out or the authorisation expires unnoticed, the patient carries on booking, and the gap is discovered when the claim is rejected and the clinic has to explain an invoice to someone halfway through treatment.
What it does
Referral and intake reading, feature by feature
Reading the page is the easy part. Most of the work is in matching the right patient and knowing when to stop and ask a person.
- 01
Referrals collected from every inbox
Documents are picked up from the referrals mailbox, the fax-to-email address, the scanner folder and portal uploads. Files holding several documents are split, sideways pages are turned the right way, and each document is sorted as a referral letter, an insurer authorisation, an intake form or a clinic letter.
- 02
Data extraction, field by field
From each document the system reads the details reception would type: patient name, date of birth, address, phone numbers, referrer, registered GP, insurer, membership and authorisation numbers, sessions covered and expiry date. Each field carries its own confidence score, so one doubtful date does not hold up the rest.
- 03
Patient matching to the right record
Extracted details are checked against the practice system by name, date of birth, postcode and any record number on the letter. An exact match links automatically. Near matches, possible duplicates and new patients go to a person, and the system never merges or creates a record on its own.
- 04
A checking screen built for reception
Doubtful documents open with the original on one side and the extracted fields on the other, each highlighted where it was found on the page. The person checking confirms or corrects a field in seconds instead of retyping the letter, and every correction is recorded.
- 05
Clinical content left to clinicians
The system does not summarise, interpret or rank what a letter says about a patient's condition. It files the whole letter into the record for the clinician to read. If the referrer has marked it urgent, that marking is shown as written to the person checking, and your own procedure decides what happens next.
- 06
Insurer authorisations tracked against bookings
Sessions covered and expiry dates from insurer authorisations are held against the patient's bookings. Reception sees how many sessions remain when booking the next one, and gets a reminder before the last session or the expiry date, so the patient or insurer can be asked about an extension in good time.
A day with it
How the day runs instead
A normal day for the referrals coordinator at an outpatient clinic, with documents read as they arrive.
- 7.45amReferrals coordinator
Overnight documents are already read. Most have matched an existing patient or been sorted as new referrals waiting for her confirmation. She opens the needs checking list first, where four documents are waiting, and starts with the consultant letter its author marked urgent.
- 8.05amReferrals coordinator
A handwritten intake form has a date of birth that could be read as two different years, and two patients share the name. She compares the phone number with both records, picks the right one, confirms the date and moves to the next document.
- 10.30amReceptionist
Booking an insured patient's next session, he sees two sessions left on the authorisation and an expiry date next week. He asks the patient to contact their insurer about an extension while they are still at the desk, rather than finding out when a claim is refused.
- 12.00pmPhysiotherapist
Before a new patient's assessment, she opens the record. The referral letter is already filed in full under correspondence, against the right patient, with the original attached. She reads the letter herself, as she always has, before calling the patient in.
- Friday 4.00pmPractice manager
She reviews the week's corrections. One GP practice's new letter layout keeps placing the registered GP in the wrong field, so she flags it for tuning. She also checks that every document received this week has been filed or is waiting with a named person.
What it connects to
Works with what you already run
Documents come in through the routes you already use and leave as entries in the systems you already trust.
Referrals mailbox and scanner
Your shared referrals mailbox, your fax-to-email address and the folder your scanner or multifunction printer saves to. Documents are collected from all of them automatically, and nothing changes for the people who refer to you.
Practice management system
Patient search for matching, new patient records created once a person has confirmed them, and insurer and authorisation details written against the right patient through its interface.
Electronic record or document store
Each letter is filed against the right record with its original file attached, so the clinician reads the document itself, not an extract of it, and can see when it arrived.
AI document models
Chosen with you for each document type, on terms that stop documents being retained or used for training, and processed in the region your obligations require or inside your own cloud account.
The first version
Start small, then grow it
A first version should be useful on day one and small enough to get right. This is where we would usually draw the line, and what tends to follow once people are using it.
In the first version
- Documents collected automatically from your main referrals mailbox and scanner folder
- Referral letters and insurer authorisations sorted and read
- Patient matching, with every near match and new patient sent to a person
- A checking screen with the original document beside the extracted fields
- Every document checked by a person until you agree which fields can file directly
Once it is in use
- Handwritten intake forms and documents that arrive by post
- Authorisation tracking with reminders before sessions or cover run out
- New patient records created from referrals once a person has confirmed them
- More document types, such as discharge summaries and clinic letters from other providers
Buyer's guide
Before you commission one
Four decisions worth making before you automate how referrals and intake documents are read.
- 01
A document platform or a custom pipeline
Intelligent document processing platforms read typed forms well, and some practice systems include basic scanning and filing. If your documents are mostly typed and you only need them filed, those may be enough. A custom pipeline earns its place when matching to the right patient, insurer authorisation tracking and a checking screen built for your reception matter more than reading the page.
- 02
What to gather first
Without sending us anything real, estimate what arrives each week by type and route: referral letters, insurer authorisations, intake forms and clinic letters, by email, fax-to-email, portal and post. List the fields reception types for each, and note which referrers send the most. Then ask whether any of them could send structured data instead of a letter.
- 03
Who should own it
The referrals coordinator or reception lead should own the checking queue and the corrections, because they know the documents best. The practice manager owns the matching thresholds and which fields can file without checking. Your data protection lead owns where documents are processed, and a clinician should confirm that clinical content always reaches clinicians whole.
- 04
How you will know it is working
Watch the size of the checking queue at the end of each day, the fields reviewers correct most often, the number of duplicate records created, and how long a referral waits between arriving and being filed. If letters are no longer typed in after the last patient has gone, and nobody asks where a referral went, it is doing its job.
Terms worth knowing
The words you will meet
- OCR
- Optical character recognition turns an image of text, such as a scanned letter or a fax, into characters a computer can search. On its own it does not know which line is the date of birth. Modern document models add layout and meaning on top, but quality still depends on the scan in front of it.
- Intelligent document processing
- Software that sorts documents by type, reads specific fields from them and checks those fields against your own records before they are used. It is the difference between a searchable scan and a completed patient record, and in a clinic the check against the practice system is what makes it trustworthy.
- Human in the loop
- A design where a person reviews the cases the software is unsure about, rather than letting it act on every document. Each field carries a confidence score, and anything below the threshold you set goes to a checking screen. It keeps people in charge wherever a mistake would matter.
Handled with care
The details that matter here
A misread form costs time. A letter filed against the wrong patient is an incident. These are the details we design around with your data protection lead.
Nothing filed to the wrong patient unnoticed
Matching thresholds are agreed with you and set cautiously. Near matches always go to a person, and every filing records which document went where, on whose confirmation. If a letter is filed against the wrong record, it can be traced, moved and logged as an incident under your own procedure.
No reading between the lines
We keep the system to administrative fields. It does not summarise a letter's clinical content, judge how urgent a referral is or decide who should see the patient. Those are clinical decisions, and the letter reaches the clinician whole, exactly as the referrer wrote it.
Tested without exposing real records
We build with synthetic and redacted sample documents. Measuring how well it reads your real letters is done on systems you approve, by named people, under terms agreed in writing with your data protection lead, and the results are reported to you field by field.
Where documents are processed
We agree with your data protection lead which model provider reads the documents, in which region, and on what terms. Documents are not retained by the provider or used for training, and copies held for checking are deleted on the schedule you set.
Guides
Worth reading first
Questions
What buyers ask first
Can AI decide which referrals are urgent?
Ours does not. Deciding how urgent a referral is, and who should see the patient, is a clinical judgement, and we do not build software that makes it. If the referrer has written or ticked urgent, the system shows that marking exactly as written to the person checking, so it is not lost in a pile. What happens next follows your own procedure, decided by your clinicians.
What happens if OCR reads a date of birth wrong?
It sometimes will, which is why every field carries a confidence score and every document is checked against your practice system before anything is filed. A doubtful date goes to a person with the original page highlighted. A date that produces no exact match is treated as a near match, never as a new patient, so a misreading is caught at the desk rather than creating a duplicate record.
Can OCR read handwritten intake forms and faxes?
Yes, but less of it will file without a person checking, and we would rather say that plainly. Handwritten intake forms are the hardest documents a clinic receives. Often the better fix is to stop receiving them: forms completed online before the visit arrive as typed answers and need no reading at all. We will tell you where that is the simpler route.
Do referrers have to change how they send referrals?
No. Coping with what already arrives is most of the point, so GPs, consultants and insurers carry on sending letters the way they do now. If a large referrer can send structured information through an electronic referral service instead of a letter, that is simpler and more reliable than reading a page, and we will point it out rather than build something you do not need.
Is AI document processing safe to use with patient data?
That depends on decisions you make with your data protection lead, not on the software alone. Under UK and EU GDPR, health data is special category data with extra conditions on its use. In the US, a supplier processing protected health information for you works under a business associate agreement. We agree the model provider, region, retention and access in writing before any real document is processed.
Tell us what arrives in your referrals inbox
Describe the letters and forms you receive and how they are handled today, but please do not send real patient documents. We reply within 48 hours and agree safe handling before we see anything.
More for healthcare
What else we build for healthcare
Patient portals
Booking and rescheduling against your real diary, forms completed before the visit, invoices paid online, and results or letters released only when a clinician chooses, with identity verified properly and every access recorded.
Booking and reminder apps
An app or mobile-friendly booking flow with reminders, one-tap rescheduling, waiting lists that offer a cancelled slot to the next patient, and deposits taken up front, so fewer appointments go to waste.
Clinic operations tools
Rotas, training and registration expiries, equipment checks, room bookings and incident logs in one place, with automatic reminders and per-site permissions, so inspection evidence builds up as the work happens rather than in a rush beforehand.
Front desk assistants
An assistant on your website and messaging channels that handles booking, opening hours, prices and preparation questions using your real calendar, with firm rules: no advice, no diagnosis, and a quick handover to a person.
Connected patient reporting
Controlled feeds from booking, records and billing into one reporting layer, with one patient identity across systems and personal details minimised for analysis, so managers see activity, capacity and revenue without spreadsheets full of patient names.
