ブログ
Your ERP Has the Answers. Nobody Can Find Them.
The data problem most organisations face is not a lack of information. It is an inability to reach the information they already have.
The system of record is not the system of knowledge
Enterprise software has done its job. Over the past two decades, organisations have invested heavily in ERPs, CRMs, HRIS platforms, ticketing systems, project management tools, and internal wikis. The records are there. Transaction histories, customer interactions, HR policies, product specifications, support ticket resolutions, procurement workflows: all of it is stored, indexed, and backed up across a constellation of purpose-built systems.
The trouble is that storing information and making it useful are two very different things. These platforms were designed as systems of record, optimised for data entry, compliance, and structured reporting. They were not designed to answer the kinds of questions people ask every day. "What did we agree on pricing with that client last quarter?" "Who approved the exception to our procurement policy for that vendor?" "What was the resolution the last time this integration broke?" The answers to these questions exist somewhere in the organisation's software stack. Finding them is another matter entirely.
What tends to happen instead is predictable. Someone messages a colleague on Slack. Someone else schedules a call with the one person who remembers how the system works. A junior analyst spends half a day rebuilding a report that already exists in a different department's folder. A new hire asks a question and gets three different answers, none of which are sourced. The information is there, but for practical purposes it is invisible.
Why enterprise search fails
Most enterprise platforms offer some form of search, but the experience rarely matches what people expect from modern search tools. ERP search interfaces tend to require knowledge of specific field names, module structures, or query syntax. CRM search works well if you know the exact account name, less well if you are looking for a pattern across interactions. HRIS portals bury policy documents behind navigation trees that only HR specialists understand. Ticketing systems return results ranked by recency rather than relevance.
The deeper problem is fragmentation. Even if each system's search worked well in isolation, the answers people need often span multiple systems. Understanding a customer relationship might require data from the CRM, the ticketing system, the contract management tool, and a shared drive. No single system's search can reach across all of those, which means the person asking the question has to know where to look before they can find anything. That prerequisite knowledge, knowing which system holds which type of information, becomes a form of tribal knowledge that lives in people's heads rather than in any documented process.
This is the gap that Fabric's search is designed to close. Rather than replacing existing systems, it works across them, providing a single semantic search layer that understands natural language queries and retrieves relevant results regardless of where the underlying data lives.
The tribal knowledge bottleneck
In most organisations, a small number of people become the de facto knowledge layer. They are the ones who know which ERP module contains the data you need, which Confluence space has the up-to-date process documentation, which Slack channel discussed the decision that led to a policy change. These individuals are enormously valuable, and they are also a bottleneck.
When one of them goes on holiday, decisions slow down. When they leave the organisation, institutional memory leaves with them. When the team grows, their time gets spread thinner, and the queue of "quick questions" gets longer. The organisation has not failed to document its knowledge. It has failed to make that documentation findable.
This pattern is especially costly during onboarding. New team members are expected to absorb years of accumulated context, and the primary mechanism for doing so is asking colleagues. The information exists in various systems, but navigating those systems requires experience that the new hire does not yet have. The result is a slow ramp-up period that could be shortened considerably if the new hire could simply ask a question and get an answer drawn from across the organisation's connected tools.
Connecting systems, not replacing them
The instinct when facing this problem is often to consolidate: migrate everything into one platform, build a single source of truth, standardise on one tool. In practice, consolidation projects are expensive, disruptive, and rarely complete. Organisations end up with a new system sitting alongside the old ones, and the fragmentation problem persists with an additional layer of complexity.
A more practical approach is to add a connective layer that sits on top of existing systems and makes their contents accessible through a single interface. This is the approach Fabric takes. Through connections to dozens of tools and support for the MCP protocol, Fabric can integrate with CRMs, ERPs, project management platforms, cloud storage, communication tools, and custom internal systems. The underlying systems continue to function as they always have. What changes is the ability to query across all of them at once.
What makes this practical rather than theoretical is that Fabric can build connectors to new data sources, not just the ones that ship with pre-built integrations. If your organisation runs a proprietary system or a niche platform that most tools ignore, Fabric can connect to it as long as it has an API. And these connections are not one-time imports. They sync continuously, so the information available through Fabric reflects the current state of every connected system rather than a stale snapshot from last month's export.
For organisations running older or more specialised systems, the company brain for legacy systems approach lets Fabric serve as an accessible front end to data that would otherwise require specialist knowledge to retrieve. A finance team member can ask a question in plain English and get an answer drawn from the ERP, without needing to know SAP transaction codes or Oracle query syntax.
From search to synthesis
Search is the starting point, but the real value emerges when an AI assistant can do more than retrieve documents. It can synthesise information from multiple sources into a coherent answer. Instead of returning a list of links and leaving the user to piece together the answer, Fabric's assistant reads across connected systems and provides a direct response, citing the sources it drew from.
This distinction matters because most enterprise questions are not simple lookups. They are synthesis tasks. "What is our exposure to this vendor across all active contracts?" requires pulling data from procurement, legal, and finance systems, then combining it into a summary. A traditional search would return fragments from each system. An AI assistant that understands the question and has access to all three systems can provide the assembled answer.
For teams that need to go further, AI agents can perform multi-step tasks across connected systems. Rather than just answering a question, an agent can gather data, run comparisons, draft summaries, and present findings, all without requiring the user to navigate multiple interfaces or write complex queries.
Making knowledge self-maintaining
One of the quieter problems with enterprise knowledge is that it decays. Documentation goes stale. Process guides describe workflows that have since changed. FAQs answer questions that are no longer being asked. Maintaining documentation is a continuous effort that most teams deprioritise under the pressure of daily work.
Fabric addresses this through docs that write themselves, documentation that stays current because it is generated and updated from the connected systems that contain the source data. Rather than relying on someone to manually update a knowledge base article every time a process changes, the documentation reflects the current state of the systems it draws from.
This approach also helps with retrieval-augmented generation, ensuring that the AI assistant's answers are grounded in current, authoritative data rather than outdated training data or stale cached documents.
Who benefits most
The pain of inaccessible enterprise data is felt across functions, but certain roles feel it more acutely than others. Product managers who need to understand customer feedback patterns across support tickets, CRM notes, and internal discussions spend significant time gathering context before they can make decisions. Sales teams that need to quickly find competitive intelligence, past proposals, or client history waste hours searching through disconnected systems when preparing for meetings. Customer service representatives who need to resolve issues quickly are slowed down by the need to check multiple systems for relevant precedents.
In each case, the cost is not just the time spent searching. It is the decisions made without complete information because finding that information was too slow or too difficult. When the barrier to accessing knowledge is low enough, people use it. When it is too high, they guess, they ask a colleague who might also be guessing, or they proceed without the context they need.
The usability layer
What organisations need is not more data, better databases, or another migration project. They need a usability layer that makes existing investments accessible. The ERP, the CRM, the ticketing system: these are all doing their jobs. The gap is between the data they hold and the people who need to use it.
Fabric fills that gap by providing natural language access to enterprise data, semantic search across all connected systems, and an AI assistant that can synthesise answers from multiple sources. For organisations that want to extend this access programmatically, API access enables custom integrations and workflows that build on the connected data layer.
The result is not a replacement for existing systems. It is the layer that makes them useful to the people who are not specialists in operating them. The data was always there. Now it is findable.
Frequently asked questions
Can Fabric connect to older ERP systems that do not have modern APIs?
Yes. Fabric supports the MCP protocol and can build connectors to new data sources, including older platforms that may not offer standard REST APIs. Once connected, data syncs continuously rather than being imported once, so the AI layer always reflects the current state of the source system. The connections framework is designed to accommodate both modern SaaS tools and legacy enterprise systems.
Does Fabric replace our existing enterprise systems?
No. Fabric sits on top of existing systems as a connective and access layer. Your ERP, CRM, HRIS, and other platforms continue to function as they do today. Fabric provides a unified search and AI interface across all of them.
How does Fabric handle data security when connecting to multiple enterprise systems?
Fabric connects to systems using the same authentication and access controls that are already in place. Users can only access data they are authorised to see in the underlying systems. For more on how Fabric approaches data governance, see AI without the risk.
What kinds of questions can the AI assistant answer?
The assistant can answer any question where the relevant data exists in one or more connected systems. This includes factual lookups, cross-system synthesis, historical queries, and pattern-based questions. It works in natural language, so users do not need to know query syntax or system-specific terminology.
How long does it take to set up connections to enterprise systems?
Many connections can be established in minutes using pre-built integrations. For more complex or custom systems, the MCP protocol and API access provide flexible integration paths. Most organisations can have their core systems connected within days rather than weeks.
Can Fabric help with onboarding new employees?
Yes. By making enterprise knowledge searchable in natural language, Fabric significantly reduces the time new hires spend tracking down information or waiting for colleagues to answer questions. New team members can query across all connected systems from day one.
Does the AI assistant provide source citations for its answers?
Yes. When the assistant synthesises an answer from multiple sources, it cites the specific documents, records, or systems it drew from. This allows users to verify the information and navigate to the original source if they need more detail.
How is Fabric different from enterprise search tools?
Traditional enterprise search tools index content and return ranked lists of documents. Fabric goes further by providing semantic understanding of queries, cross-system synthesis, and AI-powered answers that combine information from multiple sources into coherent responses rather than leaving the user to piece together results from a list of links.
Can teams build custom workflows on top of Fabric's connected data?
Yes. Fabric provides API access for teams that want to build custom integrations, automated workflows, or specialised interfaces on top of the connected data layer. This allows engineering teams to extend Fabric's capabilities to meet organisation-specific needs.
Is Fabric suitable for small teams, or is it designed for large enterprises?
Fabric works for organisations of all sizes. Startups benefit from having a connected knowledge layer from the beginning, avoiding the fragmentation problems that typically accumulate as organisations grow. Larger enterprises benefit from making existing investments in complex systems more accessible.