Inter Pharm represents close to 30 global medtech brands across orthopaedics, cardiology, neuroradiology, microbiology, anaesthesia, surgery, and transfusion, supplying the Clinical Centre of Montenegro, Montefarm, all seven regional general hospitals, and a growing set of private clinics. In August and September 2026, our commercial and regulatory team worked with Coinis, a Claude/Anthropic partner, on hands-on training to build an AI-assisted process for finding and qualifying tenders and grants. This is what we built, how it runs day to day, and what changed in the first weeks.
What problem were we solving?
Nearly all public medical-device spend in Montenegro runs through the CeJN e-procurement portal. Some procedures are simple and close in three to five days — too fast to catch by checking the portal occasionally. The portal indexes titles, not specifications, so a relevant tender can hide behind generic wording. A procedure can sit marked “in progress” well after its deadline has actually passed, which means every lead has to be checked by hand to confirm it is still live.
Funding opportunities were a separate, messier problem. EU and national SME grant calls are scattered across aggregators with inconsistent eligibility rules, and unlike tenders, we had no standing process for tracking them at all. Underneath both problems was the same gap: the institutional knowledge for judging which tenders and grants were worth our time lived in a handful of people’s heads, not in a documented, repeatable procedure.
What did we actually build?
Working with Coinis over August and September 2026, we built five things:
- A structured business profile of Inter Pharm as shared AI context — licences, portfolio, customers, financials, SME status — with every fact marked confirmed or to-verify.
- Written decision rules for qualifying and ranking tenders against our actual product categories and buyer relationships.
- A CeJN tender-monitoring procedure that scans open procedures, verifies real deadlines, filters by product category and key buyers, and produces a ranked Excel report with lots, values, and procurement contacts.
- A grant and funding screening procedure against eufondovikonkursi.com, scoring each call 0–100 for relevance with an apply, check, or monitor recommendation.
- Practice using Claude for tender document preparation — translations, certificates, proof of authorised distribution, and manufacturer authorisation letters.
The accuracy discipline we trained into every step matters as much as the procedures themselves: never invent conditions, mark unknowns clearly, and always link back to the original tender or call so a person verifies before we act on it.
How does this run day to day now?
The profile, decision rules, and procedures live in a shared Claude Project, so anyone on the team works from the same instructions. Tender monitoring runs a quick scan every one to two days, plus a full weekly review. Grant screening runs weekly or monthly and maintains a “to monitor” list for calls that are not an immediate fit but might become one. Output is always a ranked Excel report with direct tender links, lot values, and procurement contacts — not a dashboard, something our team can open and act on directly. Four to five people currently use Claude daily for this work.
What were the measured results?
These are figures from our own tracking over the program’s first two weeks — not an independent audit, and we’re presenting them as an early read, not a settled benchmark.
- Tender research time fell from roughly 10–15 hours a week to 1–2 hours.
- We found three to five relevant tenders that would otherwise have been missed.
- Tender-shortlist preparation dropped from two to three days to under an hour.
- Relevant tenders identified went from around five a month to around 15 a month at the current pace.
- Scans now cover all 1,000-plus open procedures on CeJN, instead of manual title searches on a fraction of them.
- We have screened roughly 20 funding calls so far and shortlisted two to three — a category we did not track at all before.
Is this AI making regulatory decisions?
No. This is worth being explicit about, given that we operate in a regulated, medical-device market. Claude does research, filtering, and first-pass ranking against rules our own commercial and regulatory team wrote and owns. It does not decide which tender we bid on, what goes into a submission, or any question touching regulatory compliance or product claims — those decisions stay with qualified people at Inter Pharm, and every lead is checked against its original source before we act.
What’s next?
We are scaling up AI-assisted preparation of tender documents — translations, certificates, distribution proof, authorisation letters — so the time we’ve saved finding tenders carries through to submitting bids, not just identifying the opportunity. For the procurement framework this process runs against, see Montenegro public-tender process for medical devices.
Frequently asked questions
- What problem were you solving?
- Almost all public medical-device spend in Montenegro moves through the CeJN e-procurement portal. Simple procedures can close in three to five days, the portal only indexes titles rather than specifications, and “in progress” status can persist even after a deadline has passed. Grant and funding calls are scattered across separate aggregators with inconsistent eligibility rules. The knowledge for judging which tenders and grants were worth chasing lived in people’s heads, not in a shared, checkable process.
- What does Claude actually do in this process?
- Claude works from a written business profile of Inter Pharm — licences, portfolio, customers, SME status, every fact marked confirmed or to-verify — plus written decision rules for qualifying and ranking opportunities. On that basis it scans open CeJN procedures, checks real deadlines, filters by product category and key buyers, and produces a ranked Excel report with lots, values, and procurement contacts. A parallel procedure screens funding calls on eufondovikonkursi.com and scores each one 0–100 for relevance, with an apply, check, or monitor recommendation.
- Is any of this automated end-to-end, or does a person still check?
- A person always checks before anything is acted on. We trained the process around one non-negotiable rule: never invent conditions, mark unknowns clearly, and always link back to the original tender or call so someone on our team verifies the source before we submit or apply. Claude accelerates research and first-pass filtering; people make the commercial and regulatory decisions.
- What were the measured results?
- This is our own tracking from the program’s first two weeks, not an independent audit. Weekly tender research time dropped from roughly 10–15 hours to 1–2 hours. We found three to five relevant tenders we would otherwise have missed. Shortlist preparation dropped from two to three days to under an hour. At the current pace we are identifying around 15 relevant tenders a month, up from about five, because scans now cover all 1,000-plus open procedures instead of manual title searches. On the grants side we have screened roughly 20 funding calls so far and shortlisted two to three — a category we did not track systematically before.
- Is this AI making regulatory decisions?
- No. Claude does research, filtering, and first-pass ranking against rules our commercial and regulatory team wrote and owns. It does not decide which tender we bid on, what we submit, or any regulatory or medical question — those stay with qualified people at Inter Pharm, checking against the original source every time.
- What’s next?
- We are scaling up AI-assisted preparation of tender documents — translations, certificates, distribution proof, authorisation letters — so the time we’ve saved finding tenders carries through to submitting bids, not just identifying opportunities.