During our Summer Release webinar, you sent us dozens of questions about Pathfinder, automapping, EDIFACT support, data privacy, and more. We’ve grouped every question — and its full answer – by topic below.
Q: Wird Pathfinder nur in der Lobster Cloud verfügbar sein oder auch in den OnPrem Versionen?
A: Pathfinder is available on all systems, Cloud and On-Prem, that are connected to our Lobster API Server (by default, this is true on all instances).
Q: Is Pathfinder only working on your cloud or also on the on Prem installations?
A: Pathfinder works on any instance that is connected to the Lobster API Server (all by default), independent of Cloud or On-prem deployment.
Q: Available from which version and as well available if we run on-premise?
A: Available from 26.3 right now (July 14th) and for both Cloud and On-Prem.
Q: Which version do we need to upgrade to use path finder?
A: Automapping is available from 26.3 (July 14th).
Q: Gibt es funktionale Unterschiede beim Automapping zwischen der Cloud Plattform und der OnPrem Umgebung? Was sind die Grenzen des Automappings (z.B. Edifact)?
A: No differences between Cloud and On-prem. EDIFACT support is in active development and scheduled for the 26.4 release. Mapping is supported for up to 700 fields in either target or source structure.
Q: Ist Pathfinder standard feature oder muss zusätzlich lizenziert/bezahlt werden?
A: Pathfinder is part of any LDP subscription and its new 26.3 functionality comes at no additional cost.
Q: We have installed 26.3 as soon as it was out, but I don’t see Pathfinder mapping in phase 3 of a profile yet. Is there something special I need to do for a profile?
A: Asked 2+. Already available for Update since July 14th.
Topic 2 – How Pathfinder learns & where its knowledge comes from.
Q: How does Pathfinder learn? Based on the data from other models? I noticed that you are using SAP related information during the Demo, are you using X12 related data? What happens if you choose some random Source data and target data?
A: Pathfinder is a hybrid system that continuously learns from anonymized integration knowledge from created and modified mappings (the human-in-the-loop aspect) to improve suggestion quality and also leverages publicly available knowledge like EDIFACT and other industry standard schema specifications within its training data.
Q: Where does Pathfinder get its field mapping recommendations from? Does it also analyze other profiles within the same tenant and check how the field mappings are configured there?
A: Per-tenant learning from existing profiles is in active development and will be a feature in an upcoming release.
Q: For the target and source structure, the pathfinder AI tool will do the mapping based on another profiles or just only the structure (name of fields from source structure and target structure)?
A: Pathfinder currently only uses the structural information (source and target tree and mapping entries) of the specific profile it was invoked in.
Q: Is th learning for pathfinder pecific to my lobster or is it influenced by others, too?
A: Pathfinder is a shared architecture. General level of quality is achieved by anonymized aggregate learning across profiles. The ability to add custom instructions to your specific instance is in active development. These manually created instructions are instance-specific and not shared with any other instance.
Q: Is Pathfinder also considering profile mappings done by other lobster customers?
A: Pathfinder is a hybrid system that continuously learns from anonymized integration knowledge from created and modified mappings (the human-in-the-loop aspect) and also leverages publicly available knowledge like EDIFACT and other industry standard schema specifications within its training data.
Q: Is path finder learning of already existing mappings of other profiles?
A: Per-tenant learning from existing profiles is in active development and will be a feature in an upcoming release.
Q: Does the learning of your automapping work also based on manualy created mappings?
A: Per-tenant learning from existing profiles is in active development and will be a feature in an upcoming release.
Q: Woher kommt das Wissen vom Pathfinder über die Formate von Business Objects aus SAP oder anderen ERP Systemen wie BODs aus Infor LN?
A: Pathfinder is a hybrid system that continuously learns from anonymized integration knowledge from created and modified mappings (the human-in-the-loop aspect) and also leverages publicly available knowledge like EDIFACT and other industry standard schema specifications within its training data.
Q: Q – Pathfinder Automapping: If we do not approve a suggested mapping or an under-review mapping – does pathfinder learn/remember from our choices?
A: Asked 2+. Pathfinder tracks the completion status, so it does receive a signal when an Automapping run is cancelled before it was applied, although we do not ask a reason for cancellation.
Q: When a property needs to be reviewed and a decision to be made: Do you only remember the final property mapping or also the decision (like also the correction)?
A: Asked 2+. Pathfinder tracks full provenance from original suggestion per field to any modifications or selections/deselections made by the human expert for continuous learning.
Q: What’s the model underneath Pathfinder? – can you describe special logic/rules in natural language to the AI?
A: Asked 2+. Pathfinder Automapping is a hybrid system that leverages foundation models from AWS Bedrock for parts of its processing chain. Natural language features to allow you to interact with Pathfinder via a conversational window are in active discovery and will likely be part of a future release. No ETA yet.
Q: When it comes to automapping, can we provide our own destination structure, how would it learn mapping on self made structures?
A: Pathfinder will attempt to map to any target structure you create in Phase 3. If your structure has semantically meaningful field- and node-names (in a generally understood language), it will suggest mappings for you. Quality is case-specific but can be improved over time via manual expert-correction and confirmation on your end.
Q: Frage 1: Habe ich das richtig verstanden, dass der Pathfinder von bestehenden Profilen die Strukturen lernt? Hintergrund: Wir haben XML Strukturen, bspw. vom TMS System, wo Knoten und Felder nicht immer sprechende Namen haben und meiner Meinung nach die KI daher diese im ersten Step nicht autom. zuordnen kann. Die Inhalte wiederholen sich aber in der Zielstruktur bei unterschiedlichen Kunden aber immer wieder und somit sollte Pathfinder später wissen „Info X kommt immer in das OptionField1“. Long story short: Wie kann Pathfinder customized Zielstrukturen beibringen?
A: Per-tenant learning from existing profiles is in active development. Currently, Pathfinder learns from an anonymized aggregate set of existing integration configurations.
Q: Für die Pathfinder Zuweisung der Quell- und Zielstruktur verwendet Pathfinder eine Semantische oder eine nominelle Zuordnung? Also könnte Pathfinder mit einem Inhouseformat, welches Felder wie D-0045 verwendet, auch umgehen, wenn dieses als Vorlage definiert wurde mit entsprechenden Beschreibungsfeldern? Oder wie kann ich ein Inhouseformat mit nichtsprechenden Feldern
A: Pathfinder works semantically, so indeed, as long as the inhouse structures have meaningful names and/or descriptions, Pathfinder will be able to suggest meaningful mappings.
Q: So to understand, that the tool can start as soon as you already defined a basic manual data mapping. Is the AI tool also capable to start from the whole data basis of the source system or is a pre-restriction by manual basic mapping mandatory?
A: Pathfinder is currently not able to create or modify source or target structures or create profiles. It requires a pre-existing source and target structure to execute against.
Q: Classical use case: an Xml or JSON source structure was implemented. Some time later the Xml or JSON structure is enhanced. To make use of the new fields they have to be added to the existing source structures. Can Pathfinder AI help here?
A: This functionality is in active development and will be added to Pathfinder over the next releases, when it’s ready.
Topic 4 – EDIFACT / EDI guidelines & format support.
Q: Does auto-mapping also work with EDIFACT structures? EDIFACT structures use additional qualifiers in the segments. Can such constants also be set automatically in the target structure or is a predefined target structure required?
A: EDIFACT support is in active development and planned for the 26.4 release.
Q: Wie kommt der KI-Pathfinder mit komplexeren Eingangsstrukturen wie Edifact-Daten klar? Speziell wenn nur eine Ziel-Datenbank vorhanden ist und das jeweilige EDI-Segment vorab mehrfach via Pfad betreten werden muss, um bspw. diverse MEA-Gewichte wie Netto und Brutto Gewichte abzuholen. MEA+AAZ+AAB+KGM:20.508′ / MEA+AAZ+AAA+KGM:15.008′ – Aktuell speichern wir diese dynamisch in Variablen und nutzen diese später per copy Variable in der Zieldatenbank.
A: EDIFACT support is in active development and planned for the 26.4 release.
Q: Are there ideas to improve the auto-mapping in a way that I could upload customer specific EDI Specs / guidelines and have pathfinder create the mappings based on that? For example create an order response mapping from Orders04 idoc to XML?
A: This functionality is in active development and will be added to Pathfinder over the next releases, when it’s ready.
Q: Can a guideline, like e.g., an EDIFACT guideline be provided to Pathfinder as context?
A: Asked 2×. This functionality is in active development and will be added to Pathfinder over the next releases, when it’s ready.
Q: Auto-Mapping: Welche Formate für Beispieldateien können verwendet werden? Kann auch eine schriftliche Schnittstellendokumentation in Form eines Word- oder PDF-Dokuments zur Unterstützung des auto-mappings hochgeladen werden?
A: This functionality is in active development and will be added to Pathfinder over the next releases, when it’s ready.
Q: Does this also take Qualifiers into consideration where we would now set an „equals“-function on the node. For example to map data from AG from SAP-Structure to BP in Edifact
A: Pathfinder currently only suggests structural mappings, i.e., setting links/paths between nodes and fields. To achieve this, it receives any metadata associated to this field, i.e., a field description, min and max values etc.
Q: Waren in ersten Beispiel die Pfade der Zielknoten bereits gesetzt oder hat Pathfinder diese gesetzt?
A: Ein Knotenpfad war beispielhaft gesetzt und wird von Pathfinder als „Mapped by you“ identifiziert, die verbleibenden Knotenpfade wurden durch Pathfinder vorgeschlagen. Pathfinder arbeitet sowohl auf Knoten- als auch Feldebene.
Q: Prüft das Pathfinder Automapping immer den gesamten Quell- und Zielpfad oder kann man auch auf bestimmte Unterknoten einschränken? Ich gehe davon aus, dass mit Einschränkung der zu prüfenden Pfade das Reviewing noch weniger aufwändig werden kann.
A: Asked 2+. Benutzerdefiniertes „Section-mapping“ ist in aktiver Entwicklung für die 26.4 Release.
Q: How does Pathfinder Automapping recognize or handle note loops?
A: Pathfinder Automapping in 26.3 works on the static structure and proposes node and field mappings as if a user would set path manually. A source node mapped to a target node leads to repeating entries in the output based on the number of incoming entries in the source structure. Same as existing functionality.
Q: Wie erhebt Pathfinder seine Confidence über die Entscheidung? Ab welchem Confidence-Level wird es „Grün-Gemappt“, ab wann braucht es Review und ab wann wird es gar nicht gemappt?
A: Currently, high confidence is >= 85 percent, human review is > 55 and < 85, below that is no match. These values will change over time as our confidence scoring gets optimized.
Q: Can I see example data (if picked from test-file) in the pathfinder mapping window?
A: Yes, the testfile data is displayed in the Pathfinder Automapping modal next to the target tree.
Q: Isn’t there an option to also provide an expected output file to pathfinder? It could help a lot to increase the confidence of the mapping.
A: Absolutely, we agree! This functionality is in active development and will be added to Pathfinder over the next releases, when it’s ready.
Q: Kann das Automapping auch Wertausprägungslisten für bestimmte Felder, wie zB Verpackungen durchführen innerhalb des Profils? zB eingehend ktn auf KT im Zielsystem.
A: Pathfinder arbeitet auf Basis der verfügbaren semantischen Informationen sowie Metadaten in der Quell- und Zielstruktur des Profils sowie, falls bereitgestellt, auf Basis der Testdaten. Solange sich Beziehungen anhand struktureller Eigenschaften ableiten lassen, erstellt Pathfinder einen geeigneten Vorschlag. Zusätzliche Kontextinformationen sind in aktiver Entwicklung, geplant für einen späteren Release.
Q: Ist der Autmapping-Pathfinder denn auch in der Lage, JSON-Objekte, die mittels der Funktion „http json-lookup()“ in der Zielstruktur erzeugt wurden zu verarbeiten?
A: Pathfinder works on the static source and target structure that exists on disk in Phase 3 of a profile, if available. Dynamic structures are not yet supported. But you can create a static export of the expected JSON structure and run Pathfinder on top of that to suggest a mapping.
Q: Could Pathfinder be used in segmented batches of 50 attributes with their corresponding picklists (32000 picklist values to be mapped and 500+ attributes) able to do this task?
A: Customer-defined „section mapping“ is in active development for the 26.4 release.
Q: Does Pathfinder has automatically the context of existing Profiles or do I have to let Pathfinder „run over“ it one time after I Update the Platform?
A: Pathfinder is invoked on one specific profile at a time. No additional steps are required. You can open any profile once your instance is updated to 26.3 and the Pathfinder Automapping button will be available.
Q: Kann KI auch mit Funktionen helfen? Oder nur Matching der Felder?
A: Asked 2+. In active development and currently scheduled for the 26.4 release.
Q: Kann die KI auch beim Erstellen einer Zielstruktur und Hinzufügen von Funktionen unterstützen? Oder übernimmt Pathfinder ausschließlich die Zuordnungen in eine bestehende Zielstruktur?
A: Aktuell ausschließlich die Zuordnung in existierenden Strukturen. Funktionalität zur Erstellung der Quell- und Zielstrukturen sind in aktiver Entwicklung.
Q: That sounds really cool. Is there going to be an AI that I can tell what output structure I want and that will automatically create it?
A: This functionality is in active development and will be added to Pathfinder over the next releases, when it’s ready.
Q: May pathfinder suggest also hard-coded values or frequently used functions for a certain node or field?
A: This functionality is in active development and will be added to Pathfinder over the next releases, when it’s ready.
Q: Is it planned as well as functions over the Mapping KI abzubilden in dem man test eingang und test ausgang zur verfügung stellt?
A: Asked 2+. Transformation function (chains) are in active development and currently scheduled for the 26.4 release.
Q: Can AI compare two versions of profile include functions used?
A: Not yet. Such a feature is in active discovery but we do not have an ETA yet.
Q: Kann ich über das Ergebnis des Mappings/Pathfinder-Vorschlags eine Dokumentation (z.B. in Form eines Excels) runterladen?
A: Documentation of both existing and newly suggested Pathfinder mappings/profiles is in active discovery and will be included in a future release.
Q: When you apply the new profile, is it still possible to back to the previous profile. Is that stored?
A: Yes, same as if a human colleague would have made changes to a profile. It works exactly the same as LDP always has.
Topic 7 – Data privacy, security & GDPR
Q: Where does the data used by Pathfinder Automapping go? Let’s say we build a process for HR with sensitive HR data such as names, etc. The question here is where the LLM runs, as we could quickly run into GDPR issues.
A: Pathfinder Automapping only works on the static source and target structure from Phase 3. The only way to expose/leak potentially sensitive data via a testfile is by selecting a testfile with confidential inputs. To avoid this, the user always has the explicit option to run Automapping without any testfile by toggling the checkbox in the Pathfinder Automapping window.
Q: Whenever Pathfinder learns to work better off of our mappings, do these informations remain within our system or are there chances information bleeds into other systems?
A: Pathfinder is a shared architecture. General level of quality is achieved by anonymized aggregate learning across profiles. The ability to add custom instructions to your specific instance is in active development. These manually created instructions are instance-specific and not shared with any other instance.
Q: Greift Pathfinder auch auf die Daten aus den Profilen zu? Es geht um sensible Daten die in cfg-Files oder angehängte Datenbanken liegen. Und wie geht der MCP-Server damit um?
A: Pathfinder liest ausschließlich die statische Quell- und Zielstruktur sowie die Mappingeinträge und dazugehörige Metadaten wie Feldbeschreibungen.
Q: You are supply chain specialists. What about other use cases that deal more with personnel data?
A: As with any LDP functionality, Pathfinder Automapping isn’t exclusive to the Supply Chain Domain but learns from all interactions happening on your instance. Quality might differ, as we primarily benchmark against core Supply Chain and Logistics cases.
Q: GS1 use case: How does the hallucination percentage compare to other LLMs, when memory requirements (KV cache, weights + context) would be between 180-350GB at 8bit quant without using a RAG?
A: Answered live by Jan: Pathfinder runs on shared Lobster infrastructure, not per instance, and communicates via HTTPS across the Lobster API Server.
Topic 8 – Platform API, Claude code & MCP integration.
Q: Pathfinder is an exciting step forward. Given the rapid pace of AI development, have you considered opening the platform to external AI agents such as Claude Code or Codex? Allowing customers to use the AI model of their choice for mapping creation and maintenance could be a strong advantage over a closed ecosystem. Is that part of your long-term vision?
A: Yes. Answered by later stages of the presentation, esp. Platform API.
Q: Will the former lobster pro functionality be also available as api? I really would love to use claude to build forms, behaviors and events handlers). If yes … what time line are you envisioning .. this year, next year?
A: This functionality is in active development and will be added to Pathfinder over the next releases, when it’s ready.
