Award: Digital Game-Changer Award 2024, st place jury and 1st place audience · Customer: Schott Pharma Schweiz AG · Time period: 2023-2024 · Status: in productive further development, roadmap 2026 for broader use within the group
The problem you probably know
In every technical organization, knowledge accumulates over years, distributed on local drives, in SharePoint, in emails and attachments, in tickets, in manuals, in Excel files, in wikis, in old project folders, and in the minds of individuals. A question about a component, machine, or specification typically starts a chain of phone calls, tickets, meetings, and supplier inquiries, although the information usually already exists somewhere.
The result: the same questions are asked multiple times, documents are created anew even though the knowledge is already there, and employees bind each other in research chains. At Schott Pharma, in the pitch minutes to 4 hours of time loss per employee per day quantified.
What KOBAI solves
KOBAI is a semantic search and answer layer over the distributed data stock of an organization. It does not replace existing systems. It makes the existing knowledge discoverable, combinable, and verifiable.
An employee asks a question in natural language. KOBAI searches the released sources meaningfully (not just keyword-based), provides a comprehensible answer, and links the original sources. The answer is not made up but derived from actually found evidence, with source cards, evidence status (well documented, partially documented, no sufficient source) and version information.
A concrete example from practice
Initial situation at Schott Pharma: Kamera 4065 an der Prüfmaschine AK2000 wird eingestellt. Ein Ersatz muss für bestehende und neue Maschinen evaluiert werden.
Without KOBAI: Der Maschinenbau beantragt die Beschaffung von Hardware. Die Suche in alten und neuen Ordnerstrukturen bleibt unvollständig. SAP, KnowHow Crew, ServiceNow, Archive und Projektordner werden separat überprüft. Mehrere Personen sind beteiligt, ein Ticket wird weitergegeben, in Meetings besprochen, Lieferanten kontaktiert. Am Ende kann neue Dokumentation erstellt werden, obwohl Teile des Wissens bereits verfügbar waren.
With KOBAI, question:
Is there an alternative to camera 4065 for AK2000 and what documents support that?
KOBAI points to similar projects, displays SAP order information, finds archive and SAT documents, links relevant tickets or change requests, names open points for pre-study and approval, and marks which sources are old or not approved. From half a working week becomes a solid pre-study on the same day.
What employees experience
- Search by meaning instead of just keyword-based. In Jira there might be Tilt angle, in the manual Alignment gauge, in the test protocol Tilting. KOBAI recognizes the semantic proximity and delivers all three sources.
- See sources, don’t guess. Every answer links to the original documents. Those who want more details can click through.
- Status of evidence at a glance. Answers are marked as well documented, partially documented or no sufficient source , those who need to prepare decisions in a substantiated manner know immediately what they’re dealing with.
- Version and validity notice. Veraltete oder unveröffentlichte Dokumente werden als solche gekennzeichnet, anstatt ohne Vorwarnung zitiert zu werden.
- History and favorites. Find previously asked questions again, save important answers or documents, even for the team.
Interfaces that we typically connect
- Local drives and shared network drives (even very old ones)
- SharePoint, OneDrive, Teams channel storage
- Email archives (Outlook PST, Exchange)
- Wikis, Confluence, internal documentation portals
- Ticket systems: Jira, ServiceNow
- SAP and ERP exports (CSV, JSON, REST)
- DMS and quality document storage
- PDF, Office, and scanned documents with OCR
- Active Directory or Entra ID for groups and permissions
New sources are added through clearly defined connectors. Pragmatically first (the main thing is that the pipeline runs) and later cleaner API-wise, when the source allows it.
Usable modules
- Chat and semantic search. Ask a question, receive an answer plus source cards, run parallel search tasks.
- Document viewer. Preview for Word, PDF, Excel, email attachments, images. Hit locations are highlighted, deep link to the original system.
- Q-manual. Qualitätsrelevante Dokumentation gefiltert und durchsuchbar, Freigabestatus aus dem Quellsystem übernommen, Original bleibt führend.
- History and favorites. Personal history and team favorites, including a note when sources have been updated since the last answer.
- Usage statistics. Search time, response latency, document status, hit quality, top topics, performance and optionally API costs in a dashboard.
- User and permission management. Groups, roles (User, Power User, Source Owner, Admin), hierarchical inheritance with overrides on individual elements.
- Document generator. Structured drafts from found sources for summaries, lessons learned, project profiles or ticket responses. Approval process remains outside.
- Feedback loop. Thumbs up/down, comment, report incorrect source, Source Owner is notified.
What reassures the IT department
- No migration. Data stays where it is.
- No system conversion. No existing software needs to be deactivated or changed.
- Your own local server. Web interface, indexing, vector search and retrieval run on your hardware. External cloud only where you explicitly want it.
- Groups and inheritance. Sources are assigned to groups, users too, KOBAI adheres to this.
- Audit log and traceability. Every answer has sources, every action is traceable.
How the time savings add up
In the pitch, a time saving of minutes to 4 hours per employee and day scheduled. Conservatively calculated:
- daily users per business unit
- hours savings per user and day
- 2 working days per month
- results in about 20 hours per month, approximately ,640 hours per year
- that corresponds to about 10 working days per year at 8.5 hours per day
With multiple business units, the effect scales accordingly. The numbers are an estimate; the actual effect will be measured during pilot implementation per source and team, via KOBAI’s own usage statistics.
Expansion to proactive actions
KOBAI starts as a search and answer layer. Those who want more can expand depending on the settings:
- Preliminary studies and project templates automatically generated as soon as an event arises (discontinued component, new SOP requirement, audit preparation).
- Ticket drafts prepare with source list, the editor checks and sends.
- Update reminders for outdated documents, the source owner receives a notice, not the seeker.
- Onboarding paths for new employees (curated entry sources per role/machine).
- knowledge gap detection: conflicting sources are marked and reported to the responsible parties.
award
KOBAI has won internally at Schott Digital Game-Changer Award 2024 both the 1st place of the jury and the 1st place of the audience. The solution has since been further developed productively, with a concept update in 2026 for broader corporate use.
Do you recognize your organization in that?
If yes: We will also build something similar for you. We start with a single pilot data source, often just a network drive or a database. We show measurable time savings. Then we expand step by step, at your pace, with your data sources, without having to stop or replace existing systems.