What connected data looks like in practice

"Connected" is easy to say and hard to picture. In licensing terms, it means the same piece of work can be followed from one tool to the next, with the same people showing up at each step. Here is what that looks like in a hypothetical company, before and after anonymization, and how to tell whether your own data has it.

Imagine a 60-person logistics software company. On a Tuesday morning, a customer emails their account manager: shipping labels are printing with the wrong warehouse address. The account manager forwards the email to support, and a support agent opens a ticket.

The agent posts in the support channel in Slack, tags an engineer and pastes the ticket link. The engineer reproduces the problem, opens an issue in Jira and links it back to the support ticket. A product manager joins the thread and asks whether the same bug affects two other customers. Someone writes a short doc comparing two possible fixes.

Two days later, the engineer opens a pull request whose title includes the issue key. A reviewer asks for a test. The fix merges, a deploy notification lands in Slack, the support agent closes the ticket, and the account manager emails the customer. A week later, the account manager logs a note in the CRM that the renewal conversation went well, partly because the fix was fast.

That single bug touched email, the support desk, chat, the issue tracker, a doc, the code host and the CRM. Every step references the one before it.

After anonymization, the people and companies are gone, but the trail is intact. The customer is COMPANY_087 everywhere. The account manager is PERSON_022 in the email, the CRM note and the Slack thread. The engineer is PERSON_014 in chat, in Jira and on the pull request. The support agent, product manager and reviewer each keep one token across every tool.

Email addresses, phone numbers and the customer's warehouse addresses become tokens too. If someone had pasted an API key into the thread while debugging, it would be removed outright. What is left reads like a case file: who noticed what, who decided what, and how the problem moved from complaint to fix to renewal.

This only works if the tokens are consistent across systems and across years. We explain why in why the same person must get the same token everywhere.

AI labs training agents for workplace tasks are not short of isolated text. What is scarce is the full sequence of a task as it unfolds across tools, with the context each person had at each step. A connected trail like the one above can become a training trajectory, an evaluation task with a known outcome, or part of an environment where an agent has to do the work itself.

A chat export on its own shows the conversation but not the code. A code repository on its own shows the fix but not why it was prioritized. Together, with consistent tokens, they show the job. That is why connection depth is one of the main drivers of value, alongside years of history, headcount and number of systems, as described in how company data is valued.

Most companies have more connection than they realize, because their teams link things out of habit. Look for:

• Ticket or issue keys in commit messages, branch names and pull request titles
• Links to tickets, docs and pull requests pasted into chat threads
• Integrations that post tracker, repository or deploy events into chat channels
• Support tickets that reference CRM accounts or link to engineering issues
• Consistent use of work email, so the same person can be followed across tools
• Docs and wikis that link out to the tickets and decisions they describe

You do not need all of these. Two or three habits practiced consistently over several years can produce a well-connected dataset.

A quick way to check is to pick a memorable incident from a few years ago and try to retrace it yourself. If you can get from the first message to the final fix by following links and names rather than memory, the data is probably well connected. If the trail goes cold after one tool, that tells you something too.

Connection also has common failure points. Tool migrations are the big one: if the company moved from one tracker to another in its third year and did not carry the history over, older links point at nothing. Accounts deleted when people left can orphan their messages. Teams that worked mainly on calls or in personal channels leave fewer written trails.

None of these make data worthless. They change what can be connected and for which years, which is exactly what a read-only evaluation is designed to measure. It takes under an hour of your time, and no firm number is given before the data has been counted.

There are also things you choose to break on purpose. If a particular client, channel or folder is excluded from scope, its part of the trail disappears. That is a reasonable trade, and the rest of the record usually still connects around it.

The quickest way to know whether your company's history connects is to have it counted. The evaluation looks at which systems overlap, how far back each goes and how often work crosses between them, and the result shapes what a licence could be worth. Typical licences range from $20K to $400K+.

If you think your company's tools tell one story together, book a 15-minute call and we will tell you what we see.