AI in everyday work
AI in practices and pharmacies: three workflows that give care time back
A report for the payer, a hygiene plan, patient information in four languages: in practices and pharmacies from Kiel to Málaga, paperwork eats the time meant for people. Three workflows show what an AI colleague takes on and what it explicitly does not do here.
In brief
- In Germany, 91% of doctors in private practice feel overloaded by administrative tasks, and around 61 days a year go into bureaucracy.1
- On average, every pharmacy in Germany spends more than 20 hours a week managing supply shortages.10
- Kisuno works with your own documents. Templates, quality manual, hygiene and emergency plans and your country’s billing rules become a knowledge base you can query in your own words. Every answer names its source.
- No diagnosis, no triage, no medication check, no substitution suggestion. That line is not caution, it is product law.
Starting point: paperwork eats the time meant for care
The finding is the same everywhere, only the figures carry different names. In Germany, 91% of doctors in private practice feel administratively overloaded, around 61 days a year go into bureaucracy, and more than 5,000 GP positions are unfilled.1 In France, according to a 2019 survey, general practitioners work 54 hours a week on average, 5.5 of them on administration and coordination.4
In the pharmacy it is no better. On average, every German pharmacy spends more than 20 hours a week managing supply shortages.10 In Norway, only 13% of pharmacy managers handle administrative work within regular working hours.7 In Spain the network counts 22,273 community pharmacies, 86.9% of them individually owned and 64.5% outside the provincial capitals.6
A single crisis story would still be wrong. In Germany the number of pharmacy premises fell by 440 to 16,601 in 2025.2 In Austria it has risen since 2016 and stood at 1,445 public pharmacies at the end of 2025.3 In France, 260 pharmacies closed in 2024.5 What these markets share is not the crisis, but the writing load on ever fewer shoulders.
Where the line runs
In this field the line belongs at the beginning. Kisuno makes no diagnosis and no tentative diagnosis, does no triage, no urgency rating and no appointment scheduling by symptom. It gives no treatment recommendation, no dosage statement, no interaction or medication check, no substitution suggestion and no aut idem suggestion, meaning no proposal to swap a prescribed product for one with the same active ingredient, and it evaluates no findings, images or lab values.
The reason is product law, not caution. As soon as a system contributes to a diagnosis or a treatment decision, it is a medical device. If that product has to be assessed by a notified body, meaning a conformity assessment body appointed by the state, it is at the same time high-risk AI under Article 6(1) and Annex I of the AI Act, Regulation (EU) 2024/1689. Since the amendment by Regulation (EU) 2026/1744, the obligations for that group apply from 2 August 2028.11 Pharmaceutical substitution remains a decision for the pharmacist, based on verified databases, and the diagnosis remains a decision for the doctor. And nothing leaves the practice or the pharmacy without human approval.
Use case 1: reports and applications from your own practice knowledge
The payer asks for a statement, the insurer requests a progress report, an authority sends a questionnaire. The AI colleague knows your templates, the earlier letters to the same office and your team’s wording blocks. It gathers what your documents already say and files a draft in your usual tone for review.
In the pharmacy the same workflow carries the clawback appeal, the objection to a claim the health insurer has cut after the fact: summarise the facts, name the evidence, structure the objection. In France, administration and coordination account for 5.5 hours a week.4 Repetitive, text-heavy, rule-bound: exactly where a draft from your own documents pays off.
- The payer’s request
- Templates, earlier letters, wording blocks
- A draft in your usual tone, waiting for a signature
What deliberately does not happen: no medical assessment, no estimate of an appeal’s chances, no automatic sending. A language model can produce errors; every text is a draft for review.
Use case 2: the team’s knowledge becomes queryable
What is the reprocessing procedure for this instrument? Which cover arrangement applies during the summer holidays? Today such questions interrupt the management. Your quality manual, hygiene and emergency plans, work instructions, device documentation and your country’s billing rules become a knowledge base. The AI colleague answers from it and names the source, checkable in seconds.
The benefit is greatest where knowledge sits with single people. In Spain, 86.9% of pharmacies are individually owned.6 If the owner is away, half the operating knowledge is away too. New pharmacy technicians and stand-ins ask in their own language and get answers from the same documents. Whoever may not see the personnel folder does not see it through the AI either.
- Quality manual, hygiene plan and billing rules
- Access governed per tool, every call logged
- An answer with its source, checkable in seconds
The flip side belongs in the picture: an outdated filing system produces outdated answers. For guidelines and standards, the licence decides whether they may go in.
Use case 3: information in several languages
A product is unavailable, the practice closes for two weeks, the prescription process is switched over. From one version come the notice on the door, the short briefing for the counter and the message to patients, each in the languages of your catchment area. In a rural pharmacy with two staff, that is the difference between a notice and none.
The line is especially sharp here, because the text reaches people. Every version is approved by the team first. Package leaflets, consent forms and anything legally binding remain authoritative in their official version; the translation is an aid to understanding.
- One version, written by your own team
- Language versions from the same source text
- Professional approval by the team
If you open a chat for patients, you must disclose that a machine is answering. Even there, the assistant must not rate symptoms.
Connectors: what can really be linked
An AI colleague is only as useful as what it is allowed to reach. A connector is the link to another program. MCP is an open standard for such links. Through it, programs can be addressed that have a documented interface, meaning an official route for exchanging data. Each tool has its own permission, and every call is logged. For practices and pharmacies the field is closed. The verified state of play for AI connections in August 2026:
| Software (country) | Interface | Connection to Kisuno |
|---|---|---|
| Kaddio (SE) | GraphQL API, publicly documented; access with an account and the add-on | Connection via MCP possible, the only verified case in the sector |
| Dentalink (ES) | REST API documented; administrators create the access | Connection via MCP possible, through the practice administrator |
| CGM PROFIM and Wingesfar (IT) | No public API, only the vendor’s own add-on modules | No connector, the route runs via exports and documents |
| Sistema Tessera Sanitaria (IT) | Specifications public, access via credentials and a PIN | An authority channel. Preparing yes, sending no |
| medatixx, IXOS and awinta (DE) | No public developer portal | Only through the vendor’s partner channel |
| Telematics infrastructure, ePA and e-prescription (DE) | gematik specifications are open, access only through approved components | No connector, and we are not holding one out either |
| Fortnox (SE) and Tripletex (NO) | Open REST API, documented | Connection via MCP possible: invoices, receipts, contacts |
| Microsoft 365 (EU-wide) | Graph API, openly documented | Connection via MCP possible: email, calendar and files |
France, Austria and Liechtenstein are missing on purpose, and so are CGM MEDISTAR, TURBOMED and WINAPO: we have no verified finding on them, and a promise that breaks in the first conversation costs more than it brings. More under integrations.
The benefit here does not hang on the connector anyway. Reports, applications, notices and quality documents are documents, and documents need no system connection. Anyone waiting for their practice management system to be connected is waiting for the wrong thing.
Data protection: health data
Health data are special categories of personal data under Article 9 GDPR. Above them sits a professional secrecy duty backed by criminal law in every country: in France the secret professionnel under Article 226-13 of the Code pénal, in Italy the segreto professionale, in Spain the secreto profesional, in Germany medical and pharmaceutical confidentiality. A data processing agreement does not lift that. This article is not legal advice.
Two more points belong here. First the European Health Data Space, Regulation (EU) 2025/327, published on 5 March 2025 and applicable in stages from 26 March 2027.12 It sets requirements for electronic health record systems. Kisuno is not such a system and keeps no patient record. Second, retention: how long a patient record has to be kept is for each country to decide, and the periods differ widely. The knowledge base does not take the place of the record, and the retention periods stay where the record is kept.
Kisuno runs on your own server or, in the managed variant, in ISO 27001 certified data centres. The certification sits with the data centre operator, not with us. The models from OpenAI (ChatGPT) and Anthropic (Claude) run in the AWS data centre in Frankfurt. AWS is a US company, so the US CLOUD Act applies even with servers in Frankfurt, and we say so openly instead of leaving it out. For higher data-protection requirements, open frontier models such as GLM, Kimi and DeepSeek are available, operated by European companies on servers in Europe; the CLOUD Act does not apply there. Whatever model you choose: inputs and uploaded data are neither stored by the model provider nor used for training. What we do not say: that your data never leaves your business. In the managed variant it does, under control and within Europe. Anyone who does not want that for health data operates the platform themselves. The data processing agreement and the subprocessor list are public.
In France there is one particularity. Anyone hosting health data for third parties needs HDS certification, the French approval for hosting health data, under Article L1111-8 of the Code de la santé publique.8 Institutions that operate their own information system themselves are exempt. Plainly: self-hosting is the shorter route, because then nobody is hosting for third parties. For the managed variant we settle the HDS question with you.
That leaves the objection that you already have an AI tool. Usually it is off-the-shelf software: in Norway, 71% of AI-using companies deploy ready-made software, in Sweden 62%.9 A tool like that does not know your quality manual. That is the gap a knowledge base from your own documents closes.
Frequently asked questions from practices and pharmacies
Can the AI help with diagnosis, triage or medication?
No. Kisuno makes no diagnosis, does no triage, gives no treatment or dosage statement, checks no interactions and suggests no substitution or generic swap. As soon as a system contributes to a treatment decision, it is a medical device and high-risk AI under Annex I of the AI Act, a different product with different obligations.
Can you connect Kisuno to our practice management system, such as medatixx or IXOS, or to the electronic health record?
The systems in our markets mostly have no publicly documented interface. Where one exists, the route runs through the vendor. We do not connect to the German telematics infrastructure with the electronic health record and e-prescription: the gematik specifications are open, but access runs only through approved components. Anyone promising you a connection there is promising too much.
Does our use fall under the AI Act?
Two things apply to you as a deployer: the AI literacy duty since 2 February 2025 and the transparency duty since 2 August 2026, when people interact with an AI system. An internal assistant for writing and looking things up is not a high-risk system. It becomes high-risk where decisions are made about people, for instance when pre-sorting job applications. This answer is not legal advice.
We are in France. What does HDS certification mean for us?
Anyone hosting health data for third parties needs HDS certification in France, under Article L1111-8 of the Code de la santé publique. If you operate Kisuno yourself, the exemption for institutions that run their own information system applies. No hosting for third parties arises. For the managed variant we settle the HDS question beforehand.
May patient data go into the knowledge base at all?
For the benefits this article describes, it does not need to: quality manual, hygiene and emergency plans, work instructions, templates and billing rules contain no patient data. Where health data is processed after all, Article 9 GDPR, professional secrecy and the access rights in the system apply. You settle the legal basis before the first file.
What does it cost for a team of ten?
Testing is free and not limited in time. Every new account starts with a starting credit of CHF 50 (EUR 55) for model usage, and no payment details are required. The managed instance costs CHF 7 (EUR 7.50) per person per month, through the end of the current month with 50 percent off on a twelve-month term, that is CHF 3.50 (EUR 3.75). Self-hosting stays free permanently: the LibreChat base is MIT licensed, and the repository with our own extensions follows shortly. All details on the pricing page.
See it on a case of your own
Bring a letter to a payer or a hygiene plan, and we will walk through the workflow together. Or sign up and be among the first to start.
Sources
All figures come from the following sources, last checked in August 2026.
- Kassenärztliche Bundesvereinigung, dossier on cutting bureaucracy, Germany, as of 17 June 2026: kbv.de
- ABDA, press release on the pharmacy count, Germany, values as at the end of 2025: abda.de
- Austrian Chamber of Pharmacists, statistics, Austria, as of 31 December 2025: apothekerkammer.at
- DREES, working hours of general practitioners in private practice, France, published 7 May 2019, sample of 3,300 doctors: drees.solidarites-sante.gouv.fr
- Conseil national de l’Ordre des pharmaciens, pharmacist demographics 2024, France, as of 1 January 2025: ordre.pharmacien.fr
- Consejo General de Colegios Oficiales de Farmacéuticos, community pharmacy statistics 2025, Spain, as of 31 December 2025: farmaceuticos.com
- Norsk Farmaceutisk Tidsskrift, 28 March 2026, on the survey by Rambøll for Norges Farmaceutiske Forening, Norway, response rate 57%: farmatid.no
- Agence du Numérique en Santé, hosting of health data, France, reference document per the arrêté of 26 April 2024: esante.gouv.fr
- Eurostat, dataset isoc_eb_ai, survey year 2025, enterprises with 10 or more employees: ec.europa.eu
- ABDA, German Pharmacies: Figures, Data, Facts 2026, statistical yearbook, Germany, values as at the end of 2025: abda.de (PDF)
- AI Act, Regulation (EU) 2024/1689, amended by Regulation (EU) 2026/1744 of 8 July 2026, Official Journal of 24 July 2026, on the deadlines for high-risk systems: eur-lex.europa.eu
- Regulation (EU) 2025/327 on the European Health Data Space, Official Journal of 5 March 2025: eur-lex.europa.eu
Last reviewed: