Organizations do not communicate in neat, labeled threads. A compliance question might start as a Slack message, migrate to email when a manager is pulled in, generate a ticket when IT is involved, and end with a phone call that is never documented. The issue is one thing. The communications are many things, scattered across systems and time.
BBCO calls the process of connecting these fragments latching. It is not a metaphor. It is the literal mechanism by which disparate messages are joined into coherent issue paths that can be analyzed structurally.
Email clients reconstruct threads using In-Reply-To and References headers. Latching goes further: it connects messages that belong to the same issue even when they span separate email threads, live in different communication systems, and carry unrelated subject lines.
Latching is structural, operating on who is involved, when they communicated, and what organizational relationships connect them. It ignores message content entirely. The question is adjacency and overlap, never what the messages are “about.”
If two messages within a configurable time window share participants, they are candidate links. This is the strongest signal. When the same people are communicating in close temporal proximity, the probability that they are discussing the same issue is high.
Partial overlap is sufficient. If Message A involves Alice, Bob, and Carol, and Message B involves Bob, Carol, and David, the shared participants (Bob and Carol) establish a candidate link.
Time constrains the linkage. Two messages from the same participants separated by three months are unlikely to concern the same issue. The time window is configurable because organizational tempo varies: a trading desk operates on minutes, while a compliance review might span weeks.
For high-frequency communication pairs, participant overlap alone produces too many false links. In these cases, normalized subject-line matching acts as a tiebreaker.
Subject lines are stripped of reply/forward prefixes and compared after normalization. This is deliberately lightweight: a header operation, not a language understanding task. When a subject match is found, the candidate link is further corroborated by checking for participant overlap between the two messages. Links that pass both checks carry higher provenance confidence than either signal alone.
Candidate links are scored and then validated before merging. This separation is deliberate. Automatic merging risks over-clustering: connecting messages that share participants but concern different issues.
Instead, candidate links are validated against depth and participant guards before being accepted into the issue graph.
The quality of latching determines the quality of every downstream metric. When related messages remain unlinked, issue paths are truncated and the organization appears to resolve issues faster than it actually does. When unrelated messages are joined, paths are inflated and the organization appears more complex than it is.
Both errors corrupt variance measurements, and both are invisible to anyone who only looks at the output metrics. For captive insurance programs, where these variance metrics inform capital adequacy and retention confidence, latching accuracy has direct financial consequence. BBCO makes the latching logic open, inspectable, and configurable so that the decisions shaping the evidence base are themselves available for scrutiny.
The most important connections in an organization are often the ones that cross system boundaries. Latching makes those connections visible.
Read more from the BBCO community.