How does SIGNLD work

In this article

• Step one: connecting your systems
• Step two: resolving entities into a Knowledge Graph
• Step three: asking a question inside a Topic
• Step four: how the Decision Brief gets composed
• Where does the AI inference actually run
• How is evidence traced back to source data

Step one: connecting your systems

SIGNLD connects read-only to more than 800 business systems, including CRMs, accounting platforms, job scheduling tools, and e-commerce systems. Read-only means SIGNLD can pull data out to analyze it, but it cannot write back to or change anything in your source systems. Connecting the first system typically takes about 15 minutes, and every connection is encrypted with TLS 1.2 or higher in transit.

For the wider context, see our overview of what SIGNLD is.

There is no data migration and no separate staging environment to configure. Once access is granted, SIGNLD begins reading records immediately and moves to the next step of the pipeline on its own. Because access is read-only, connecting a new system carries little risk for IT teams evaluating the request, since there is no path for SIGNLD to alter production data.

Step two: resolving entities into a Knowledge Graph

This is the step that replaces a traditional data modelling phase. As records come in from each connected system, SIGNLD identifies when a customer, job, or invoice in one system is the same record as a customer, job, or invoice in another system, even if the identifiers, names, or formats do not match exactly. Those matched records get linked together inside a private Knowledge Graph.

This matters because most operators never see the full picture of a customer or a job, because the story is scattered across five or ten disconnected tools. The Knowledge Graph puts the full story in one place automatically, without a consultant building a schema first. No mapping spreadsheet or integration consultant is required to tell SIGNLD that a customer in the CRM is the same customer in the accounting system. The resolution happens as part of the pipeline, not as a separate project with its own timeline. For a deeper look at how this graph gets constructed, see how the Knowledge Graph is built in SIGNLD. Data at rest inside the graph is encrypted with AES-256.

Step three: asking a question inside a Topic

Once the Knowledge Graph exists, you ask a question in plain language, such as "which accounts are past due by more than 60 days" or "which jobs ran over budget last quarter." Inside SIGNLD, that question and its related evidence live inside a Topic, sometimes called a Decision Topic.

A Topic keeps the systems, entities, and evidence relevant to a specific business question organized together, so follow-up questions in the same Topic stay consistent with earlier answers. You do not need to specify which systems to search or how records connect. That work already happened in step two. A Topic can also be shared with other people on the team, so a finance leader and an operations leader looking at the same overdue receivables question see the same evidence and the same recommended action. The full definition of a Topic and other core terms are covered in concepts.

Step four: how the Decision Brief gets composed

Once SIGNLD identifies the relevant part of the Knowledge Graph for your question, it composes a Decision Brief. The Brief has four parts: the finding, which states the answer directly; the evidence, which lists the specific records that support the finding; a confidence signal, which indicates how strong the underlying evidence is; and a recommended action, which suggests a next step based on the finding.

This structure means the answer is never just a number on a chart. It comes with a trail back to the rows that produced it, so anyone reading the Brief can verify it rather than take it on faith. You can find the full pipeline described visually on the how it works page.

Where does the AI inference actually run

The language model that reads your question and drafts the Decision Brief runs on a single-tenant AWS Bedrock instance. This is a private large language model powered by AWS Bedrock, dedicated to your organization rather than shared across customers. No third-party AI APIs are involved in generating the Brief, and the model is never trained on your data.

This single-tenant setup matters for companies handling sensitive financial or customer information, since it keeps inference isolated to your own instance rather than routing through a shared, general-purpose service. Combined with read-only connections and full query logging, this design keeps every step of the pipeline auditable, from the moment a system connects to the moment a Decision Brief is delivered.

How is evidence traced back to source data

Every finding in a Decision Brief links to the specific records in the Knowledge Graph that support it, and those records trace back to the original rows in your source systems. Every query run against SIGNLD is logged and timestamped, so there is a complete audit trail of what was asked, when, and what data was used to answer it.

This traceability is what separates a Decision Brief from a typical AI-generated summary. You can open the evidence behind any finding and see exactly which invoice, job, or customer record produced it.

Key takeaways

• SIGNLD connects read-only to more than 800 business systems, including CRMs, accounting platforms, job scheduling tools, and e-commerce systems.
• The language model that reads your question and drafts the Decision Brief runs on a single-tenant AWS Bedrock instance.
• Most connections can be set up by a business user in about 15 minutes, since SIGNLD uses read-only access rather than custom integrations.
• SIGNLD resolves conflicting or mismatched identifiers as part of building the Knowledge Graph, matching the same customer, job, or invoice across systems even when names or formats differ.
• Now that you know the pipeline, from read-only connection to Knowledge Graph to Decision Brief, the fastest way to understand it is to see it running on your own data.

FAQ

Does SIGNLD need IT to set up the connections?

Most connections can be set up by a business user in about 15 minutes, since SIGNLD uses read-only access rather than custom integrations. IT involvement depends on your internal approval process for granting system access.

What happens if two systems disagree about a record?

SIGNLD resolves conflicting or mismatched identifiers as part of building the Knowledge Graph, matching the same customer, job, or invoice across systems even when names or formats differ. The evidence attached to a Decision Brief shows which source each data point came from.

Can I ask a new question without reconnecting systems?

Yes. Once your systems are connected and the Knowledge Graph is built, you can ask new questions inside a Topic at any time. There is no need to reconnect systems or rebuild anything for each new question.

Is the Decision Brief a static report or does it update?

A Decision Brief reflects the data available in the Knowledge Graph at the time the question is asked. Asking the same question again later pulls current data, so the Brief updates as your underlying systems change.

Does the private LLM ever see other customers' data?

No. Inference runs on a single-tenant AWS Bedrock instance dedicated to your organization, so your data and the model's behavior stay isolated from other SIGNLD customers.

Related reading in this series: How long does SIGNLD take to set up and How much does SIGNLD cost.

Try SIGNLD free

Now that you know the pipeline, from read-only connection to Knowledge Graph to Decision Brief, the fastest way to understand it is to see it running on your own data.

Try SIGNLD free or explore the concepts behind the platform.