Medical information response, explained.

A pharmacist's dialysis question, from his email to a signed reply

One email about Dravenzo and dialysis, in five steps

1. The question arrivesMartin Kessler, PharmD, emails at 9:52 a.m. with two questions about one patient on Dravenzo 10 mg who starts hemodialysis on Monday.
2. Someone reads and splits itPart 1 is on the label. Part 2, a 5 mg dose after dialysis, is off-label. No single standard response covers both, so it needs a custom letter.
3. The letter is built, sentence by sentenceFrom the label, a standard response and two small studies, each sentence tied to its passage. This is the step this guide is about.
4. Named people signThe specialist signs. In this desk's rules, a medical reviewer adds a second signature because part of it is off-label.
5. It goes to him alone, and the record staysSent privately with its references and the label. What was asked, what was sent and who signed stay on file.

The people, products, publications and figures in this example are fictional.

Why a helpful sentence is the risk, and three more facts about these letters

On-label, off-label, standard response and unsolicited, in plain terms

Fourteen terms are defined in the one-page summary.

Live answers, standard response letters and custom letters, and how one question can need two sets of rules

In shortA desk has three ways to answer: on the call, with a stored letter, or with a letter written for this person. Label status belongs to each part of a question, and one sentence of a question can be on-label while the next is off it. Reading the whole question as one status is the first mistake a desk can make.

Read that chart as a picture of the sorting. Your own mix may differ, and we found no public figure for what is typical. On this sample, live answers are the largest share at 514 of 1,184, and each custom letter takes far more skilled time than a live answer.

Why label status belongs to each part of the question

A question usually comes with a story, and the story holds more than one question. Dr. Kessler's email has two. Dr. Helena Marsh, a dermatologist in the same sample, asks three in one web form: pregnancy data, breastfeeding and dosing every 2 weeks. The first two are answered from the label's sections on those subjects. The third is outside the approved dose.

So one letter holds on-label sections and an off-label section, each under its own rules. FDA's 2011 draft is itself limited this way: it "does not address requests for information about approved or cleared indications", so it governs only the off-label part.2

The two sentences that sound alike, on either side of the line

All three sentences come from the guide's fictional letter. The studies behind them are sample text.

FDA's unsolicited-request draft and its scientific information guidance, each with its status in November 2026

In shortTwo FDA documents get confused. The December 2011 draft covers a firm's reply to a question someone else asked, and it is still a draft. The January 2025 guidance covers information a firm decides to share on its own, and it is final with an unclear implementation status. Neither binds anyone by itself, and neither replaced the other. A third, the July 2024 draft on misinformation, is about something else.

"Scientific exchange" is the field's phrase, and neither FDA text on this work uses it

People in medical affairs, the part of a drug company that houses MI, say "scientific exchange" for any non-promotional talk about data, including a field scientist's conversation and a reply letter. A full-text search of the 2011 draft and the 2025 guidance finds the phrase in neither. FDA's own term in 2025 is "scientific information on unapproved uses". We keep FDA's words here, because a reader who thinks the 2025 guidance covers a reply will apply the wrong list.

What the protection amounts to, in FDA's words

The return on all this care is one sentence. If a firm answers a private unsolicited request in the manner described, FDA "does not intend to use such responses as evidence of the firm's intent that its product be used for an unapproved or uncleared use".2

A firm may choose another way. Doing so "would not constitute a per se violation of the law, but could potentially be introduced as evidence of a new intended use". So the draft marks out a protected place to stand, and a company's counsel decides how close to stay to it.

Courts add a limit from the other side. In Caronia (2012) a federal appeals court set aside a conviction resting on truthful speech about off-label use, and in Amarin (2015) a federal district court granted a preliminary injunction on truthful statements. Commentary on both stresses that false or misleading claims stay unprotected.10 Most desks treat these cases as limits on the government and keep their procedures in place.

What makes a question unsolicited, in FDA's own eleven examples

In shortThe test is who started it. Prompting "in any way" by the firm or its representatives makes a request solicited, and FDA says that may be evidence of intent. Three of FDA's examples are unsolicited, and eight show how a firm's own conduct spoils a later question. The fact is known at intake and hard to rebuild later, so it has to be recorded when the question arrives.

Example 11 matters for anyone building self-service. A public page that lets a visitor pull a stored off-label answer from a menu turns the resulting requests into solicited ones.2 A portal can take a question and route it to a person, and a Canadian benchmark suggests few would use it for stored answers anyway: standard responses viewed on company websites came to about 0.5% of annual inquiries.17

The hard case: a question passed on by the company's own field team

A medical science liaison, or MSL, is a scientific field employee who talks with doctors about data. A doctor asks an MSL something. The MSL sends it in. Nothing was prompted, as far as anyone knows, and yet the paper trail shows a company employee's name on the request.

FDA's Example 5 makes the setting decisive, and the draft says nothing specific about relaying a question. Some desks therefore go back to the doctor and ask. That confirmation is a company control, and it gives a later reader a record of whose question it was.

Public questions get a thin public answer

A question posted where others can read it follows different recommendations. A firm should respond only if the question is about its own named product. The public response "should be limited to providing the firm's contact information and should not include any off-label information". The response says the question concerns an unapproved use and gives the medical department's contact details, and the person responding discloses who they work for. The detailed answer follows privately, once the person writes in.2

FDA's Example 13 sets the test: "Can Drug/Device X be used for Condition Y" may get that reply, while "What drug/device can be used for Condition Y", where the firm's product is not approved for it, should get none.

The seven recommendations for a private off-label reply, and what each one means on the page

In shortFDA's draft lists seven recommendations for a private reply: who gets it, how narrow it is, how balanced, what tone, who writes it, what travels with it, and what records stay. Each maps to something a reviewer can check on one letter. A second signature appears nowhere in the list. Where a company requires one, that is the company's own control, and we found no published count of how many do.

Quotations are from FDA's December 2011 draft, section V, which is a draft and nonbinding.2

The five items that travel with an off-label letter

Who may sign, and what the signature records

FDA's draft asks for "medical or scientific personnel" and names no credential. Companies often hire pharmacists, because a drug information pharmacist is trained to answer from the literature, and many use a physician as the medical reviewer.

The signature itself is a separate question. Where FDA's electronic records rule, 21 CFR Part 11, applies, a signed record must show "the printed name of the signer", "the date and time" and "the meaning (such as review, approval, responsibility, or authorship)", and those must appear in any readable form of the record, with the signature linked so it cannot be copied to another record.19,20 The same rule asks for validated systems and audit trails on closed systems.21

Does that rule reach a reply letter? Part 11 applies to records kept "under any records requirements set forth in agency regulations", and to records submitted to the agency.22 FDA's records list for a reply is a draft recommendation. So whether your letters are Part 11 records is a question for your counsel. Some desks hold to the same three items in any case, because a signature with no stated meaning leaves nobody able to say what the signer vouched for.

How a standard response library is written, kept current and retired, and why old wording comes back

In shortA standard response is a concise, approved summary of the evidence on one question. Published survey means put updates at more than twice the count of new documents. The slow part is proving each paraphrase still matches its paper. A retired version does not disappear: its sentences live on in every letter that quoted it, ready to be reused.

Do not read that fall as a measured gain. The paper's own Results give 8.3 hours and its Discussion 8.7, the 2018 survey covered 27 US departments, and the two may count different steps and different people. The authors themselves warn of limited external validity, and phactMI, the industry group behind the survey, and Microsoft sponsored the work.

Respondents named the three slowest steps: paraphrasing study content, checking paraphrased text against the source publications, and checking the document's data integrity.7 In other words, the expensive part of a revision is proving that every sentence still matches the paper it came from.

What sits inside one standard response

How a retired sentence gets back into a letter

We found no FDA rule that sets a review interval for a standard response. Some desks set a yearly date at approval and review early when a label changes or a material paper appears.

Retiring a version is the easy half. The harder half is that a letter signed last spring still sits in the archive, and it is the most convenient source for this week's letter.

In the sample used in this guide, a letter signed in March 2026 followed version 4 of a pregnancy response. Version 4 was retired on June 12, 2026. Version 5 removed a 15-week stopping suggestion because the label sets no such period. An October draft picked up the old sentence through the March letter. It has a source and reads well, and its source is no longer valid.

Reading "3 of 11": working through the numbers an off-label letter has to state exactly

In shortA medical information letter is mostly small numbers from small studies. The job is to state each exactly, with its denominator and its limits, and never to carry one a step further than the study did. Here is the guide's fictional letter, worked through, including the tempting sum nobody should send.

The two sample studies, and what each number allows you to say

That width is why "27%" alone misleads. Showing the count with its denominator lets a reader see the size of the study at once. FDA's advertising rule names the error on the other side: presenting a study "in a way that implies that the study represents larger or more general experience with the drug than it actually does".4

The tempting sum a letter must not make

A pharmacist reading "1.3 times the exposure" may do the arithmetic at once. 10 mg gives 1.3 times normal exposure, so 5 mg might give about 0.65 times normal, if exposure scales with dose.

That sum may be sound pharmacology, and the pharmacist is free to make it. The company cannot, because no patient in either study received 5 mg, and no study tested the sum. In the fictional letter, the sentence reporting the limits says so in those words: "no patient in either study received a 5 mg dose". The reader gets the facts and keeps the decision.

Counting the fictional letter, sentence by sentence

Eight sentences, four sources drawn and three references. The label and the two papers go on the reference list. The standard response is an internal source, so it does not. A letter's checks come out as counts like these: sentences with a source, sentences removed, references attached. Clocks, the turnaround targets a desk sets itself, count too. The email arrived on Wednesday, October 7. Five business days later is Wednesday, October 14, and that is the due date a desk with a five-day custom letter clock would set.

Nine ways a medical information letter goes wrong, and why a read-through misses most of them

In shortFew errors are false statements. Most are true sentences in the wrong frame: too strong for the evidence, missing a limit, tied to the wrong reference or to a retired version. A reviewer who reads the whole letter and agrees with it can approve an unsupported sentence without ever disagreeing with anything.

The advertising rule's list works as a reviewer's checklist

21 CFR 202.1(e)(6) is written for advertising, and FDA's draft says a reply made in the manner it describes is not expected to meet advertising disclosure rules. Even so, its list of ways to mislead reads like an editor's marking guide, and reviewers can use it that way.

The rule names a "quote or paraphrase out of context", literature "that purport to support" a claim "but in fact do not support the claim", and "favorable information or opinions about a drug previously regarded as valid but which have been rendered invalid by contrary and more credible recent information".4 None of these needs a false statement, which is why the review has to move from the letter to the sentence.

A person misses them for ordinary reasons. Letters are built from earlier letters, and a sentence approved once arrives looking settled. Volume pushes the reviewer to trust what looks familiar. And checking, as the survey respondents said, is among the slowest parts of the work.7

From one advisory sentence to intended-use evidence, untitled letters and False Claims Act cases

In shortA reply that strays loses the draft's protection and can be read as evidence of a new intended use. FDA's letters on promotion jumped in 2025, and past off-label settlements ran into billions; the 11 that the Government Accountability Office (GAO) counted for 2003 to 2007 began with sources other than FDA. We found no public FDA letter aimed at a private reply. The practical cost is more often the rework of a desk that cannot show what it sent.

These letters concern promotion. They matter here for two reasons. They show FDA's attention on how companies describe their drugs, across a large volume: GAO counted over 68,000 final promotional submissions in 2007, and FDA reports over 70,000 in 2024.24,25 And they let you read the agency's recent reasoning on promotional claims, in its own words.

We found no public FDA letter aimed at a private medical information reply. None of the 119 untitled letters on FDA's table, dated 2017 through September 28, 2026, contains the words "unsolicited", "medical information" or "standard response", and two web searches for warning letters and older letters found none. We did not read FDA's warning letters one by one, so this is no proof none exists, and these replies are private.

How the record protects a desk, or fails to

GAO's finding that other sources started those cases points to the file. In enforcement, the internal record is what gets read: who asked, what they asked, who passed it on, what was sent and who approved it.

Companies under integrity agreements, the compliance terms attached to some federal settlements, have had this spelled out. Novartis's 2010 agreement with the inspector general of the US health department (HHS) required a database of inquiries with six named items: date, form, the name of the requesting professional or institution, the nature and topic including "exact language" if written, the nature and form of the response, and the name of the Novartis representative who interacted with the requester, if known.18 That list is a good starting point for anyone designing intake today.

What a wrong or late letter means for the pharmacist and the patient on Monday

The legal language serves a plain purpose. A pharmacist reading a letter is about to do something for a patient on Monday. A wrong number, a drifted claim or a stale recommendation reaches a decision the company will never see. A slow letter fails too: the question gets answered somewhere else, without the label beside it.

In a survey of professionals who had just used these services, 246 responded. Of the 222 who answered on patient care, 39% said the reply helped them educate patients and 32% said it enhanced treatment efficacy.31 Those are self-reports to a survey whose authors work for or are paid by drug companies and their industry group, and they point to what a reply is for.

Answering an off-label question in the UK, the EU and Canada: the same reply by different legal routes

In shortEach of these treats a reply to a specific question as something other than advertising, on conditions. The UK says activity designed to solicit questions is likely promotional. Canada says a response on off-label uses from sales or marketing staff may be promotional. So a global desk writes one letter per country, each citing that country's label, with one discipline underneath.

The label differs by country: US Prescribing Information, the UK and EU Summary of Product Characteristics (SmPC), the Canadian Product Monograph. A use that is on-label in one may be off-label in another, so the status of a part depends on where the requester practices. Letters need a template per country, and in Canada often a version in French.

Patient details in a question also travel differently. In the US, a drug maker is generally outside the definition of a covered entity, the kind of organization the health privacy law HIPAA regulates directly, though the hospital is inside it (our reading of 45 CFR 160.103).36 European desks work under data protection law. Ask your counsel, and collect no more about a patient than the question needs.

How large medical information teams are, what else they do and how questions reach them

In shortPublic numbers on this work are few, and they come from the industry's own surveys. A 2025 survey of 35 US companies gives staffing and services. A Canadian benchmark gives channels. We found no current published figure for inquiry volumes per company, cost per letter or market size, so we show what you can measure instead.

That chart explains why the desk's time is under pressure. Answering questions is one job among many: the same teams review promotional pieces, train field teams and support labeling. The survey gives median hours of about 16 a week reviewing promotional materials before meetings and 6 in review committee meetings.8

Numbers we could not find published, and how to get yours

The trend is easier to state. In the 2025 survey, authors reported that "the majority are evaluating or in the process of adopting" AI, without percentages, and they named oversight of AI in the insights process as one of three priorities.8 Vendors now sell response automation as a category. Treat every speed claim as a question to test on your own letters.

AI on drug questions: chatbot test scores, invented references and how far the rules reach a reply letter

In shortGeneral chatbots answering drug questions on their own have mixed published results, poor in the earlier tests, and their references have often been invented or wrong. A tool that writes only from retrieved sources fails more quietly, with a real paper attached to a sentence it never says. No regulator has written a rule for AI in this job. The sound design has software draft and check, and has named people decide, sign and answer for the result.

Why "it searches before it writes" is only part of the answer

Newer tools retrieve first, and they do better on invented references. They still fail. Stanford and Yale researchers found that three legal research tools, from two providers whose claims they tested, each gave a false or wrongly sourced answer 17% to 33% of the time.39

On drug questions, one 2026 study put 30 questions answered by drug information faculty to four clinically focused chatbots and one general one. The best scored 17.82 out of 25, and all were statistically below the faculty's answers. "Lack of false information" scored highest and "evidence supporting the content" lowest.40

A second 2026 study, at one hospital, found two tools, GPT-4o and Medical Chat, with composite scores comparable with pharmacists', two others lower, and low agreement among the raters (kappa 0.015, on a scale where 0 means chance agreement and 1 means perfect), which itself tells you how hard this is to judge.41

Read together, those studies disagree on how good the answers are. The 2026 faculty comparison does show where to look: its raters scored the evidence behind an answer lowest, and only the source can settle that. These are hospital drug information tests, a neighbor of manufacturer letters and no measure of them. A company's reply has added duties those studies never tested: label status, balance, tailoring and enclosures.

Where software fits in the work, and where people stay

What regulators and standards say, and how far each reaches

How a custom letter gets written by hand today: the people, the hand-offs and the days in each step

In shortA custom letter passes through a specialist, sometimes a contact center first, a reviewer when it is off-label, and a standard response owner when the library falls short. On the guide's sample desk, most of the elapsed time goes to searching, drafting and checking, and to waiting. Teams run this in house, outsourced or in a mix, on a case system that holds the record.

A specialist, a medical reviewer, a manager and a response owner: who answers for what

In house, outsourced or with software: three ways desks are built

Martin Kessler's dialysis question, from the inbox to the medical reviewer's signature

In shortThe question splits into an on-label part and an off-label part. The letter assembles paragraph by paragraph, each tagged with its source, and one click opens the passage. The off-label guard strikes a sentence that advises an unapproved dose. Priya Ranvelle, the specialist, signs, Dr. Omar Haddad, the medical reviewer, adds the second signature, and the letter goes to one person with its references and the label.

1

The question, split into two parts with a status on each

On the left, Dr. Kessler's own words, with each question highlighted and numbered. Part 1, severe renal impairment and dialysis, is marked on-label. Part 2, a 5 mg dose after dialysis, is marked off-label. Below that sit the requester ("HCP verified"), the route ("Custom letter: no standard response covers every part") and the request ("Unsolicited, from the requester directly").

In the center is the letter itself, on paper, with a tag in the margin beside every paragraph: Label, Standard response, Publication or Template. On the right, the off-label guard's checklist and the sources drawn.

2

"Replay the agents": the letter assembling in view

Priya can watch the run again at any time. Each agent, a software worker with one narrow job, lights in turn along the top of the page: inquiry reader, request check, evidence finder, letter writer, off-label guard, citation checker. Sources light up as they are drawn. Paragraphs appear one by one, each with its tag.

The replay shows that nothing happened out of sight. The same steps are on the case's audit trail afterward.

3

One click from a paragraph to the passage behind it

Priya clicks the paragraph on the single-dose study. The right panel changes to that paragraph's source: Varga 2024, a sample publication, with the passage highlighted. Exposure 1.3 times higher in 8 patients with eGFR below 30, and less than 5% of the dose removed by a 4-hour session. The numbers in the sentence and the passage match, digit by digit.

This is the change a reviewer feels most: the letter becomes a set of sentences to confirm.

4

The off-label guard strikes one sentence, with its reason and origin

Section 2 opens with the approved dose, 10 mg once daily or 2.5 mg with a strong CYP3A4 inhibitor (a medicine that slows the body's breakdown of certain drugs), and says the asked dose and timing are not in the approved labeling. It reports the two studies, then their limits: both small, one retrospective and uncontrolled, and no patient given 5 mg. Then the label's warning on hypotension.

The letter writer had carried over a sentence from a draft of letter MI-23977. The guard struck it and says why: "Advises an unapproved dose and has no published source. An off-label section may state the evidence; it may not recommend." It stays on screen, struck through, and comes back only with the medical reviewer's signature.

Check your own template against FDA's draft. This sample section states the approved dose and points to the enclosed label for full risk information. The draft's list also names a prominent statement of the approved indications and of important risk information.2

5

Priya signs with a stated meaning, and the route is on the screen

The checks sit on one line: 8 of 8 sentences cited, off-label section built from scientific evidence, 1 sentence removed by the guard, 3 references attached. The route has three stops: Priya signs now, Dr. Haddad adds the second signature, then send and file.

The meaning of her signature is written out: "I reviewed this response and approve it as accurate, balanced and specific to the question asked." That wording, her name and the time go on the record, the three items 21 CFR 11.50 names.19

6

Dr. Haddad signs against a list, for the off-label section

Because part of the letter is off-label, it waits. His screen lists what he is attesting to: specific to the question asked, disclosure, balance, references and label, and kept apart from promotion, sent privately to the requester only. Priya's first signature, October 7 at 10:24 a.m., sits beside it.

The second signature is your company's control, set in Settings. It lines up with FDA's recommendations and comes from none of them.2

7

Sent to one person, with the case closed and archived

The letter's PDF is built with its 3 references and the Prescribing Information enclosed, then sent by secure email. The case closes and the letter, sources, decisions and signatures are archived. The library keeper notes 4 dialysis dosing questions this quarter and suggests an off-label standard response to its owner.

A retired pregnancy sentence, a field-submitted question, the library and the manager's dashboard

In shortThe same discipline runs across the desk. The citation checker catches a sentence following a retired standard response, and a person decides. A question sent in by a medical science liaison is confirmed with the doctor before anyone drafts. The library shows what is due and what keeps coming back. The manager sees volume, turnaround and citation coverage on one page.

1

A sentence from a retired standard response, caught and explained

Dr. Helena Marsh, a dermatologist, asks about Veltrimab and pregnancy planning, among other things. Ten of the eleven sentences check out. The eleventh reads: "Northwind Bio suggests considering stopping Veltrimab 15 weeks before a planned conception."

The citation checker explains: it was reused from a letter signed in March 2026, which followed SRD-VLT-014 v4. Version 4 was retired on June 12, 2026, and v5 removed the advice because the Prescribing Information sets no such period. It offers the current wording. Priya can use it, keep the original with a note the reviewer will see, or compare the two versions. The checker never accepts its own finding.

2

A question a field team sent in, confirmed before drafting

Dr. Nadia Petrov, a cardiologist in Toronto, asked a liaison about Dravenzo with clarithromycin at the end of a meeting. The request check keeps the liaison's note of how it came up, records that the question was confirmed with her as her own, and shows who reviewed that confirmation, all before drafting. The clock for this is one business day, and it is locked. In the sample's last 30 days, all 41 field-submitted questions were confirmed first.

The same check gives an answer when the origin is unclear: it asks the requester. A question prompted by a sales representative gets no off-label section.

3

Every question in one list, with its route, label status and clock

All open and recent inquiries, filterable by product, channel and label status, each showing its route, stage and time to due. "Field-submitted · confirmed" sits beside the route where it applies.

4

The standard response library, with versions, owners and use

186 current standard responses across three products. Each row shows its version, approval date, next review, how often it was used in 30 days, and its status. Two are due for review. One of those, storage and temperature excursions, is the most used of the nine shown, at 58 uses in 30 days. Only current versions can be cited, and an owner gets a reminder 30 days before a review.

5

The questions that keep coming back without an answer on file

Three questions drew 4 or more custom letters in 90 days and have no standard response: giving Dravenzo through a feeding tube (6), a washout when switching from another biologic (5), and Talvoren with untreated brain metastases (4). One click drafts a standard response from the signed letters and their sources. It is a draft. Its owner approves it.

6

The desk at a glance, from the door to the letter

Dana Okafor, the MI manager, sees 1,184 questions answered in 30 days and 236 custom letters. All 2,124 sentences checked have a source, on this sample. A flow shows today's 48 questions from how they arrived to how they were answered and where they are now. Below that, turnaround against the by-hand baseline, and the topics people ask about.

7

Turnaround against the by-hand baseline, and what people ask

Custom letter turnaround, with the by-hand 2025 baseline beside it, and the top topics. Pregnancy and lactation lead, followed by dosing and administration, and drug interactions. Off-label use is its own topic, at 71. Topics are how a manager decides which standard response to write next.

8

Signers, clocks and sources are settings your manager owns

Who may sign custom letters, who gives the second signature, whether exact-match standard responses go without a second look, and where patient questions go. Clocks for custom letters (5 business days) and standard response letters (2) can step up or down. The field-submitted clock is locked. Rules added in plain words sit here too, all on the audit trail.

9

What the agents may cite, and which template each country uses

Current approved standard responses and current labels are switched on and locked. Data on file counts only if approved for external use. Peer-reviewed literature can be cited, and abstracts from medical congresses can be allowed with a note. Each letter template shows its version and who approved it, and US, Canadian and UK versions sit side by side.

From inquiry reader to library keeper: the letter's seven agents, and the specialist and medical reviewer who sign

In shortSeven agents each do one narrow job, and each has things it never does. None of them signs. A medical information specialist signs every letter, and a medical reviewer adds a second signature to any letter with an off-label part. Each finding is decided by a person.

Four people who answer for a letter, and what each signs or owns

Roles follow your organization. On the screens: Priya Ranvelle, PharmD, specialist; Omar Haddad, MD, medical reviewer; Dana Okafor, manager; Dr. Maya Chen, Veltrimab standard response owner. The controls here are meant to support your compliance; the product is never itself "compliant" or "certified". Botminds' own credentials are SOC 2 Type II audited and information security managed to ISO 27001.

Inquiry records and standard responses going in, signed letters and a full record coming out

In shortYour case records, standard responses, labels, literature and country templates go in. A letter with a source on each paragraph, a status per part, a reference list and an audit record come out. Your case system stays the system of record. Design targets are labeled as targets.

The case system, library, literature and directory you keep

Medical review barely moves in that target, and it should barely move. Faster review would raise a fair question about what the reviewer is doing. The gain is in everything before it.

What slow letters, drifted wording and re-checking cost your medical information desk

In shortYour cost sits in skilled hours, spread across drafting, checking and keeping a library current. The public figures give the shape only. Your case system and one week of sampling will give you the real numbers. Keep rare and large costs, such as an enforcement matter, in your risk discussion and out of your business case.

Finding your own letter, library and review numbers in a week

What a late or inconsistent letter costs beyond your hours

Your return on medical information automation: custom letters, library upkeep and review time

In shortSix things create value here, and three are easy to price. The calculator counts specialist hours on custom letters, library hours and reviewer hours. It counts nothing for risk avoided or for clinicians answered sooner, so weigh those separately, and count only the hours you would redeploy.

Your hours back on custom letters, standard responses and review

The formulas: letter hours = letters × (today − with help). Library hours = (new × hours + updated × hours) × share saved. Reviewer value = letters × off-label share × (today − with help) × cost. Default letters are the sample desk's 236 in 30 days × 12. Default 3 hours sits within the sample desk's 2 to 4, and 1.5 within its estimated 1 to 2. The 25% library share and the review minutes are assumptions, and small companies should expect much smaller library numbers than these means. Risk avoided counts for nothing here, and hours count only if you would redeploy them.

Scoring your readiness for letter automation: request records, standard response versions and who signs

In shortEight questions score your records, your library, your rules and your people, 0 to 2 each. A ninth is for reflection. Nothing you choose leaves this page. Your band suggests the next step.

Fifteen questions for any medical information AI vendor, on citations, advice and signatures (ask us too)

In shortA demonstration on a vendor's own letters proves little. Test on your label, your questions and your awkward past letters. Ask for evidence on these five themes, and watch for the warning signs beside each answer.

Piloting letter drafting on one product: signed letters first, then live questions in shadow

In shortStart with one product. Have medical review and compliance pick past signed letters and plant known errors before anyone sees the tool. Then run live questions in shadow, with specialists still writing their own, and only then let drafts reach a signer. Decide on missed advice and wrong citations first, speed second, and set your pass rules before you see any result.

The letter pilot's scoreboard: missed advice, wrong citations, signer trust

Medical information letters and unsolicited requests: what teams ask most

In shortEach answer stands alone, in plain words, with status as of November 2026. This is general information, so ask your own counsel before relying on it.