IDMP and ePI product data, explained.

IDMP data explained: one medicine's approved facts, written once, in codes every EU system can read

One tablet's facts, from approved documents to a shared EU record

1. The regulator approves the documentsNarvexa's summary and leaflet, version 4, are approved on March 14, 2026. The dossier and carton match the approval, or should.
2. The data steward finds the approved versionsFour documents, 23 pages, owned by labeling, chemistry, manufacturing and controls (CMC), and supply.
3. Each fact becomes coded dataThe steward fills the record, matches each value to its EMA term and settles conflicts. This is where the 98-tablet pack is caught, or copied into the EU record.
4. Someone signs and submits to EMAThe steward signs. The operations lead sends. The record is checked, and anything wrong comes back.
5. The record has to stay trueEvery later change to the product can change the data. Other EU systems read it, and shortages teams rely on it.

Step 3 is the subject of this guide. The people, companies and figures in this example are fictional.

Why the December 2026 deadline surprises IDMP teams: four things newcomers miss

IDMP, PMS, SPOR and ePI in one line each

Fourteen terms are defined in the one-page summary.

The ISO IDMP standards and how EMA turned them into SPOR and the Product Management Service

In shortIDMP is five ISO standards. EMA applies them through four data services, and the rules for filling them in are in the EU IDMP Implementation Guide. Chapter 2 of that guide, version 2.5 from August 2026, is the document that says what each field means.

The five standards fit together like nested boxes. A substance sits inside a pharmaceutical product, which is the form the patient takes. That sits inside a medicinal product, the thing a company is authorized to sell, with its name, packs and manufacturers.2 Articles 25 and 26 of Commission Implementing Regulation (EU) No 520/2012 oblige EU member states, authorization holders and EMA to use them. Submitting data on authorized medicines to EMA has been mandatory since July 2012, in an older format.2

The four SPOR services behind every coded value

How many fields a PMS product record holds

EMA's FAQ puts PMS at around 180 data elements, against around 50 attributes in XEVMPD.3 Its current data-element guide runs to 232 pages, and many elements repeat for every ingredient, package and manufacturer.6 One consultancy counted 250 to 300 fields per product in 2015.11 No regulator publishes an average for a real product, so count yours once you have mapped one.

EMA's PMS data enrichment deadlines: June 2026, December 2026 and June 2027

In shortEnrichment applies to products that are not centrally authorized, meaning they did not go through EMA's single EU-wide route. Critical-list products, those on the EU's Union list of critical medicines, were due at the end of June 2026. All others owe manufacturer data by the end of December 2026 and pack sizes by the end of June 2027. This guide calls each of these deadline groups a lane. EMA moved two of those dates in September 2025 and has asked industry associations for assurance that the new dates will hold.

Who owes enrichment, and who does not

How SPOR data mapping works: free text in a document, one coded term in the record

In shortMost fields do not accept free text. A document's wording has to become a term from RMS, a location from OMS, or a substance from SMS. Often it is a straight match. Sometimes one phrase has to become several terms, and sometimes the list needs changing first.

The film-coated tablet is the easy case. SmPC section 3 says "Film-coated tablet", and RMS has a term with that exact name, code 100000073665.8

The hard cases come from the way documents compress information. A blister described as "aluminium/aluminium" says something a packaging engineer understands and a term list does not. In EMA's own worked examples, the container and its materials are separate values: the container type "Blister" (100000073496), then the materials, such as "Aluminium" (200000003200) and "PVC" (200000003222).8 So one short phrase can become one container and three materials, and only the dossier says which three.

When EMA's list has no term or the wrong address

You do not type around a missing term. You ask EMA to add or correct it, which takes time you should plan for. OMS change requests now go through the EMA Service Desk, with supporting documents, from someone with a SPOR user role.7 RMS change requests moved to their own portal on October 1, 2025.10

EMA's 2018 overview gave an indicative five working days for validating an OMS request, and warned that requests with missing documents or the wrong request type are rejected.18 We found no current service level, so treat five days as a floor and test it with a real request.

Which approved document holds which IDMP field: SmPC, leaflet, carton, Module 3 and the eAF

In shortNo single document holds a whole record. Composition is in the SmPC and Module 3.2.P.1, packaging in SmPC 6.5 and 3.2.P.7, sites in 3.2.P.3.1, 3.2.S.2.1 and the electronic application form (eAF). EMA makes the SmPC the main reference and lets Module 3 harmonize existing data where the documents differ.

The sections a data steward knows by heart

Section numbers follow the current ICH M4Q(R1), the international guideline for the dossier's quality part.20 Its revision, M4Q(R2), is still a draft, and ICH's work plan targets final adoption in June 2027.21

EMA's rule when the SmPC and Module 3 disagree

Chapter 2 says the contents of the eAF, Module 3 and SmPC "shall be aligned, where applicable", and that "the SmPC as authorized/to be authorized is the main referring document for data entry purposes".6 For products already in PMS, where the common content is not aligned, "the information available in the relevant sections in Module 3 can be used to harmonize the values in PMS", as long as confidentiality is kept and no complexity is added.6

Chapter 8 gives the reason, naming four things that happen in older products: different detail across SmPCs, different ways of expressing strength, excipient lists that are not uniform, and different phrasing for national reasons. Its worked example aligns on the most detailed SmPC that matches Module 3.8

Three sums to run on any product: pack size, strength and the deadline

EMA's electronic product information: the EU ePI Common Standard, the roadmap and the pilot

In shortePI is the authorized product information in the FHIR-based EU ePI Common Standard. It starts as a voluntary extra step for centrally authorized medicines, English first, while Word and PDF stay the legal submission. National products follow after a readiness assessment. It becomes mandatory for newly authorized medicines when the new pharmaceutical legislation applies.

How an ePI is built inside the FHIR standard

ePI covers Annex I (the SmPC), Annex II, Annex IIIA (the label) and Annex IIIB (the leaflet).25 EMA's implementation guide, release 1.0.0 of September 17, 2026, built on FHIR R5, describes the shape in the picture below.28 EMA's portal lets a company author in its editor, and its guidance also describes importing content through FHIR.25

What the 2023 to 2024 pilot measured

EMA's pilot took 25 regulatory procedures, and 23 produced an ePI. Of those, 22 ended favorably and all 22 were published.29

Industry participants took 5.2 hours on average to create one, with a range from about 1 hour to 16, against a target of 8. EMA is careful about that number: the ePIs in the pilot "were not notably long or complex".29 The one target only partly met was the editor itself, for adding sections, formatting content pasted from Word, and inserting images and tables.29 EMA's conclusion: no blocking issues, and the network should move toward implementation.29,30

Why approved documents disagree, and the IDMP data errors that follow

In shortEach document is right for its own history. A variation changes the dossier first, a leaflet lags the summary, an address gets abbreviated, and old products were never written to line up. The errors that reach the record are mostly quiet ones: a value copied from the wrong version, a near-match term, a site written two ways.

Nobody signs a document they know to be wrong. The differences build up because a product lives for decades under many hands. A shelf-life extension goes through as a variation, the change procedure EU rules define, and the dossier section changes. The patient documents follow at the next convenient update. A company merges and its addresses are rewritten in one place. The EU's variation guidelines were reissued in September 2025, which will not undo the history already in your files.31

Eight errors that reach an IDMP record, and how each one happens

Chapter 2 asks for discrepancies between the documents to be minimized.6 The record is where you find them. Fixing the document itself happens at its next variation.

What bad product data puts at risk: blocked applications, inspector attention and blind spots in a shortage

In shortEMA's FAQ names the consequences of out-of-date data: the web-based application form stops working for the product, submitted forms can be rejected, and discrepancies go to national inspectors. The bigger stake is what the data is for. In a shortage, EMA needs to know who makes what, and where.

What EMA says follows when product data falls behind

In EMA's wording, NCAs are national competent authorities, the national medicines regulators, and an MAH is a marketing authorization holder.

How ready pharma's IDMP programs are, and what the benchmark found

In shortThe best public benchmark covers 18 organizations in 2024. They named manual data collection and silos as their biggest problems and standardization as the main hurdle to integration. Most see value beyond compliance. Nobody publishes a reliable market size or a cost per product.

In the third quarter of 2024, the Pistoia Alliance, with ACCURIDS and MAIN5, interviewed 81 people at 18 organizations.32,33 The sample is small and leans toward large companies that already care about the subject, so read it as a rough guide.

"The biggest challenges identified are manual data collection, data silos, and the resulting lack of integration across systems."32 89% of companies said IDMP is worth more to them than compliance alone, though the report also says compliance drives the short-term work.32 About 40% said they had a compatible data model.32

Who answers for IDMP data inside a drug company

The people who answer for it are heads of regulatory operations and regulatory information management at authorization holders, usually sharing the work with master data and labeling. A service provider's update in October 2026 says plainly that "there are issues with product data at many companies", and advises a clear data governance process with ownership of the data.13 That matches the benchmark. It is also a seller's view, so weigh it accordingly.

IDMP data extraction from the SmPC: what AI does well, where it falters, and what regulators expect

In shortPublished research shows language models pull flat facts from an SmPC well and nested packaging data badly. The authors say human validation is mandatory. No regulator has written a rule for AI in product data entry, so general AI, data integrity and computer system expectations apply, and the signature stays with a person.

A 2025 study tested two large language models on 81 EMA SmPCs, against 5,224 reference elements. It reports a score of up to 0.98 on one similarity measure for the medicinal product section, and concludes that the method allows accurate and scalable extraction.34

On a stricter word-by-word similarity measure, scores fell sharply as data got more structured, and packaged product, the part next June's deadline depends on, came last. The deepest nested data averaged around 15%.34 The authors kept to SmPCs with one product and one dose form, built their reference set from model output corrected by experts, and state that "human validation is mandatory".34

What each authority expects of AI in regulated product data

Four limits of any AI that reads approved documents

How regulatory data teams build a product record today: who does what, and where three weeks go

In shortA data steward gathers documents from four teams, copies values into a spreadsheet and then into PMS, looks up each term, chases the owners of any conflict, and signs. Nobody has published how long this takes. Our estimate is about three weeks of elapsed time per record, most of it spent copying, looking up and settling.

Data steward to labeling manager: four people who answer for one record

Spreadsheet, RIM module, service provider or document extraction: four ways to do it

Narvexa 50 mg: four approved documents become a signed IDMP record with 79 days to spare

In shortThis is the example from the top of the page, inside the product. Five agents read Narvexa's approved documents and fill 61 of 64 fields in 7.3 seconds, each tied to its sentence. Ines Kovač, the regulatory data steward, makes the three decisions they leave her: a container term, a site and a pack size. Then she signs.

1

Four approved documents, one press of a button

It is October 13, 2026. Ines opens VLT-118, Narvexa 50 mg film-coated tablets, authorized through decentralized procedure XX/H/0000/001/DC in 9 countries. The record shows 0 of 64 fields. Beside it are the documents that hold the answers: SmPC and leaflet version 4, both approved on March 14, the 28-tablet carton, and Module 3 sequence 0007.

The record is the home screen, with the source document on the left and the IDMP tree on the right. She presses "Read approved documents".

2

Sixty-one fields, each with a line to its sentence

The field extractor reads all 23 pages and fills the four parts of the record: substance, pharmaceutical product, packaged product and manufacturer. A thin line runs from each field to the sentence behind it. SmPC section 3 says "Film-coated tablet.", and the SPOR term matcher locks it to the RMS term of that name, a step the screens call a snap. In all, 38 values snap to EMA terms.

Sections 4.3 to 5.3 of the SmPC are clinical text and stay out of the record. The readiness scorer reports 61 of 64, ready after three decisions by the steward.

3

"Aluminium/aluminium blisters" matches no term

The container material glows amber. Below it are the three nearest terms with the evidence for each. Cold-form scores 91%, because Module 3.2.P.7 describes cold-formed polyamide (OPA) 25 µm, aluminium (Al) 45 µm and PVC 60 µm. Aluminium and PVC alone scores 62%, because it misses the polyamide layer. Strip scores 40%, because the dossier says blister.

Ines picks the first. It takes one click where a request or a long search used to. The agents never pick a value below the 0.95 threshold she has set, and they never invent a term.

4

One address written two ways, and a pack only the SmPC lists

The leaflet gives the batch release site as "Voorbeeldweg 41, 2399 SS Leiden". EMA's organization list holds Veltara Pharma B.V. at "Voorbeeldweg 41-43", the address Module 3.2.P.3.1 uses. The matcher offers that location at 89%, and Ines chooses it. The screen adds that 8 other products name the site the same way.

Then the cross-document checker shows the 98-tablet pack. Ines excludes it from the record, and follow-up FU-0218 goes to Sofie Vandermeer, the labeling manager, to align SmPC section 6.5 at the next variation. The agent shows both passages and leaves the decision to her.

5

Ines signs, and the record becomes version 5

All 64 fields are filled. She signs with her password and a stated meaning: "Approved for submission to EMA's Product Management Service". The record becomes version 5, and the 9 Narvexa products join batch PMS-0427.

The agents do not send anything to EMA. Martin Aldred, the regulatory operations lead, sends the batch.

The solution's design target is 2 days per product record, against our estimate of about 3 weeks by hand, and 96% of fields filled from approved documents. They are targets, to be measured on your own products.

From the portfolio board to the phone preview: nine more screens of IDMP and ePI work

In shortOne product is the story. These screens show the rest of the work: the board of 412 products, fields filling as the documents are read, the pack size decision, the signed record, repeated wordings across the portfolio, conflicts, submissions, your settings, and the leaflet as ePI.

Narvexa is a nationally authorized product, and EMA's draft roadmap reaches those after centrally authorized medicines, with national implementation in 2027/2028.23 On these screens its ePI is prepared and approved ahead of that rollout, in English, Danish and German.

Reading and matching by agent, with every record, ePI and batch signed by a named person

In shortFive agents each do one job within firm limits. None of them chooses between documents, rewords approved text or sends anything to EMA. A steward signs each record, a labeling manager each ePI and an operations lead each batch. These are controls that support your compliance with electronic-records rules; the responsibility for compliance stays with you.

Field extractor to readiness scorer: five agents for one product record

Steward signs the record, labeling the ePI, operations the batch

Sentence-linked, approved versions only, every change a version

Botminds is SOC 2 Type II audited and manages information security to ISO 27001. A product cannot be certified for GxP, 21 CFR Part 11 or EU Annex 11, so we describe controls: role-based access, maker and checker separation, electronic signatures with meaning, and an audit trail you can export.

Approved SmPCs, Module 3 and SPOR lists in; signed IDMP records and ePI out

In shortThe solution works from documents you already hold and the lists EMA already publishes. Your regulatory system, PMS and the portal stay where they are. Decisions, signatures and evidence sit with the record, ready for export.

Your regulatory information (RIM) system, PMS access and EMA's ePI portal stay in place

What a hand-keyed IDMP record costs you in steward weeks, returned data and a second round for ePI

In shortStart from your own costs, because no regulator or study publishes them. Three published figures frame them: how far from done the first lane was four weeks out, what an ePI took in EMA's pilot, and what companies call their main hurdle.

Timing five product records to find your IDMP baseline

Launches held up, audits and a repeat project for ePI

What faster product records return: backlog hours in year one, upkeep and ePI after that

In shortPut in your own numbers. Most of the value in an enrichment push arrives in year one, when the backlog clears. Later years rest on upkeep, ePI and reuse of the data. Only the ePI hours today come from a published source; every default marked illustrative is an assumption to replace.

Keying, term lookups and repeated wordings: where IDMP automation saves

Products in your December lane, hours per record and your hourly cost

A company with hundreds of national products, sold in many countries with repeated wording, clears this easily. A company with ten centrally authorized products owes no enrichment push at all, and should say so before buying anything.

How ready your IDMP program is: portfolio, documents, term decisions, sites and sign-off

In shortPick the answer closest to your team today. Eight questions score 0 to 2; the last is for reflection. The meter adds up your score as you go, and each question says why it matters and what a strong answer sounds like.

What a score of 0 to 16 means for your IDMP program

What to ask IDMP software or an enrichment service about sources, SPOR terms, conflicts and ePI

In shortAsk every vendor the same questions, us included. The answers show how the record is built: what it reads, how it picks terms, what happens when documents disagree, what evidence you keep, and how you would measure it on your own products.

Test IDMP extraction on 40 of your own products with planted errors, then decide on three numbers

In shortMeasure first, test on your own documents with errors you planted, then decide with three numbers. With manufacturer data due at the end of December, run the quick steps at once and keep the pilot on real work, so nothing it produces is thrown away.

1. Time five records, one simple and four with several countries, by step. That is your baseline.
2. Count your lanes: products owing manufacturer data, pack sizes and ePI, with your rate of work against each date.
3. Gather 40 products with approved documents, your source order and an approval path to share them. Copy ten and plant known errors in them.
4. Run on live products, with your stewards signing every record that goes to EMA.
5. Decide with three numbers: catch rate by error type, wrong values by field family, and staff minutes to a signed record.

Catch rate, wrong values and minutes per signed IDMP record

IDMP compliance, PMS enrichment, SPOR and ePI: the questions regulatory teams ask most

In shortEach answer stands on its own, with status as of November 2026.