Proactive Knowledge Graph building in SIGNLD: what changes
SIGNLD is changing how the Knowledge Graph gets built. It used to form as a team worked, learning from chat, data connections, and interaction. Now SIGNLD builds the graph proactively from connected systems as soon as they're connected, without waiting for the user to ask anything first.
Proactive Knowledge Graph building is in beta in SIGNLD as of October 2026, with general availability planned for early 2027. This post covers what specifically is different, and what stays the same.
For the wider context, see our explainer on what a proactive knowledge graph is.
In this article
- How the Knowledge Graph used to build
- What's different now
- What SIGNLD proposes vs what you confirm
- Why this doesn't remove the review step
- What connecting a system looks like now
- What doesn't change
- What this means for teams already using SIGNLD
- Comparison: before and after proactive building
- How this affects teams evaluating SIGNLD for the first time
- FAQ
How the Knowledge Graph used to build
Before this change, the Knowledge Graph in SIGNLD grew out of usage. A team connected a system, asked a question, and SIGNLD mapped the tables and fields that question touched. Ask about overdue invoices and the invoice table got mapped. Ask about deal stages the following week and the pipeline table got mapped then, not before.
This worked, and it meant the graph never contained structure nobody cared about. But it also meant coverage lagged behind connection. A team could have five systems connected and still have a graph that only reflected the handful of questions someone had actually typed. Anything not yet asked about stayed effectively invisible to the graph, even though the data was already connected and readable.
What's different now
With proactive building, the trigger moves from the question to the connection. The moment a system is connected with read-only access, SIGNLD reads its schema and starts identifying entities, relationships, and likely cross-system matches, before a single question has been asked.
This means a team that connects its CRM and its accounting system on day one can, in principle, see a broader graph forming that same day, reflecting structure across both systems rather than only the parts a chat question happened to touch. Connecting your first system still takes about 15 minutes and getting a first answer still takes minutes, but the graph behind that first answer now has more context available to it from the start.
What SIGNLD proposes vs what you confirm
Proactive building doesn't mean SIGNLD asserts a finished graph. It means SIGNLD proposes structure and a person confirms it. When two systems are connected, SIGNLD looks for entities that plausibly refer to the same real-world thing, such as a customer name in a CRM and a customer name in an accounting system, and surfaces a proposed match along with a confidence level.
A person then reviews that proposal. They can confirm it, which locks the match in as reliable for future Decision Briefs, or reject it and correct the mapping if the match is wrong. Nothing proposed proactively is treated as ground truth until someone has looked at it. This is a deliberate design choice: proactive discovery is meant to save the time of manually pointing SIGNLD at every table and field, not to remove the judgment call about whether a proposed match is actually correct.
Why this doesn't remove the review step
The honest tradeoff here is real, and it's worth stating plainly: proactive building gets you a broader graph faster, and it means SIGNLD is making more proposals that a human still has to check. A system that reads a schema and infers structure from field names and data patterns is working with less certainty than a system that has watched months of real questions confirm or contradict a mapping.
That's why confidence scores on proactively discovered matches tend to be more conservative early on, and why the evidence behind any Decision Brief still links back to the actual source records the finding is based on. If a proposed entity match is wrong, the review step is where that gets caught before it quietly shapes an answer someone relies on. SIGNLD is not designed to skip that check, even though the initial mapping now happens faster than before.
What connecting a system looks like now
The connection step itself hasn't changed. You authenticate a read-only connection to a system, in about 15 minutes, and SIGNLD pulls the schema. What happens immediately after that connection is what's new: instead of a mostly blank graph waiting for the first question, SIGNLD starts proposing entities and relationships from that schema right away.
For a team connecting a second or third system, this also means cross-system entity matching starts sooner. Previously, a cross-system question might be the first time SIGNLD attempted to match a CRM company name to an accounting customer name. Now that matching attempt can happen as soon as both systems are connected, ahead of any specific question being asked, with the results waiting for review rather than being generated on demand.
What doesn't change
What this means for teams already using SIGNLD
Teams that were on SIGNLD before this change don't lose anything from how their existing graph was built. Entities and relationships that were already confirmed through the reactive process stay confirmed. Proactive building applies going forward, to newly connected systems and to gaps in existing connections that hadn't been touched by a question yet.
In practice, this means an existing account might see new proposed entities and cross-system matches appear for systems they connected months ago, simply because SIGNLD is now looking at that schema proactively rather than only in response to a question. Those proposals go through the same confidence scoring and review process as anything else. Nothing gets silently added to the graph without a chance to check it first.
For a brand-new account, the difference is more visible from day one. Instead of a graph that only reflects a handful of early questions, a new account connecting two or three systems in its first day can expect a broader set of proposed entities and relationships waiting for review almost immediately after each connection finishes.
The output of SIGNLD, a Decision Brief with a finding, evidence linked to source records, a confidence score, and a recommended action, is unchanged. Read-only access to connected systems is unchanged. The 15-minute first connection and minutes-to-first-answer pattern is unchanged. Spreadsheets remain first-class sources, mapped the same way any other connected system is.
What's changed is purely when and how the underlying graph structure gets proposed: proactively from the connection, rather than reactively from usage. The confirmation step that keeps the graph accurate is still there, and still matters just as much as it did before.
Comparison: before and after proactive building
| Before (reactive) | After (proactive) | |
|---|---|---|
| Graph mapping trigger | A question touches the data | The system gets connected |
| Coverage right after connecting | Limited to what's been asked | Broader, based on discovered schema |
| Cross-system matching | Starts when a cross-system question is asked | Starts as soon as two systems are connected |
| Review step | Still required | Still required |
| First connection time | About 15 minutes | About 15 minutes |
How this affects teams evaluating SIGNLD for the first time
A team evaluating SIGNLD for the first time will notice this change mostly in how quickly a connected system starts looking mapped. Where a trial in the reactive model might have shown a mostly empty graph until several questions had been asked, a trial today shows proposed entities and relationships appearing shortly after each connection completes, even before a first Decision Brief is requested.
This doesn't change what the trial itself involves. Connecting a system, asking a question, and getting a Decision Brief back all work the same way. What's different is that the structure behind that first answer has more of the connected system's schema already accounted for, rather than only the narrow slice a single question would have touched under the old approach.
Related reading in this series: What you still control when the graph builds itself and How proactive discovery finds the entities you never mapped.
Key takeaways
- Before this change, the Knowledge Graph in SIGNLD grew out of usage.
- Proactive building doesn't mean SIGNLD asserts a finished graph.
- The honest tradeoff here is real, and it's worth stating plainly: proactive building gets you a broader graph faster, and it means SIGNLD is making more proposals that a human still has to check.
- A team evaluating SIGNLD for the first time will notice this change mostly in how quickly a connected system starts looking mapped.
- Not if the review step is used as intended.
FAQ
Does proactive building change how long it takes to connect a system?
No. Connecting a system still takes about 15 minutes for the read-only authentication and schema read. What changes is what happens immediately after that connection, since SIGNLD now starts proposing graph structure right away instead of waiting for a question.
Will proactive building make the graph less accurate?
Not if the review step is used as intended. Proactively discovered matches come with a confidence score and require confirmation before they're treated as reliable. The graph should end up broader sooner, without skipping the human check that keeps it accurate.
Do I still need to confirm entity matches manually?
Yes. SIGNLD proposes matches based on schema and pattern matching, but a person still reviews and confirms cross-system entity matches, particularly ones with lower confidence scores, before they shape a Decision Brief.
Does this change what a Decision Brief looks like?
No. A Decision Brief still has a finding, evidence linked to source records, a confidence score, and a recommended action. Proactive building only affects how the underlying knowledge graph forms, not the structure of the answer you receive.
Is this available to all SIGNLD accounts today?
It's in beta as of October 2026, with general availability planned for early 2027. Check the roadmap for the latest status.
Read the category overview in what is a proactive knowledge graph, or see more answers in the proactive knowledge graph FAQ. For the full mechanics, see how it works.