Ops board
How to report on the system and improve it. This is a different board from the funnel dashboard. That one shows stages. This one shows whether the signals are being acted on.
The principle
Ask the questions first, then go to the data.
There are two ways to use data. One is to check that the thing you built works. The other is to start with "what ten questions should I be asking this week," and then look. The first finds what you expected. The second finds what you didn't.
Some of the ten, to make it concrete:
- Which signal type got the fastest response last week, and which got none?
- Which accounts had two Mediums stack and never crossed into a High?
- Which email got a reply that corrected our number?
- Which rep's queue is empty, and which is backed up?
Four lanes
A mock of this board is at signoz-ops-board.vercel.app. Illustrative data; the structure is the point.
| Lane | What's in it |
|---|---|
| High | One signal that's enough on its own to act. Migration doc read, credit ceiling hit, renewal inside 60 days, SSO asked for, dashboard shared into a new company. |
| Medium | Real, but not enough alone. Integration connected, config snippet copied, former user at a new job. Needs a second Medium and a decision-maker at the account before it fires. |
| Unknown | Something happened and we can't yet say what it means. A visit we can't tie to a person. A path through the site that stopped. A reaction in the community. Watched, not routed. Nobody gets an email from this lane. |
| Fast action | What's already been sent to Slack, sorted by how fast it goes cold. Frustration newest first. Deadlines nearest first. Everything else oldest first, with a floor of five a week. |
What becomes a High. A Medium turns High when a second Medium lands at the same account and a decision-maker is active there. An Unknown turns Medium when we can name the person or the company behind it. Nothing in Unknown goes straight to High. Weak signals never enter any lane.
Tiebreak inside High. Two accounts, one rep: the one with an engineer and a budget-holder both active goes first.
The numbers on it
- Median hours from a High signal to first touch. The one number that says whether any of this works.
- Queue depth per rep.
- How often the signal gets mentioned in the call. If the rep never brings up the migration doc the account hit, the signal was delivered and not used.
- Reply rate by signal type. Cost card on migration hits versus plain follow-up on ceiling hits.
- Percentage of events arriving untagged.
Closed-won fills five fields
Before a deal can be marked won in HubSpot, five fields are required:
- Which signals fired, in what order
- Who replied, and who signed
- Which message got the reply
- Time from the first High signal to close
- What made this account fall outside the profile, if anything
Field 5 is where new account types come from. A deal that the model would have skipped is a template, not an exception.
Each closed deal becomes a lookalike query in Clay. Accounts that match get watched for the same first signal, and get the same sequence with the message that already worked once.
The weights in the scoring rules move only from this data. Never from opinion. Reviewed quarterly.
Honest caveat: SigNoz's close volume is low. The first few templates will be built from one or two deals each. The system is built for when the volume arrives, and filling five fields costs nothing now.
Slack is the lab
A seeded channel where anyone on the team posts articles, Reddit threads, and Medium links. An agent reads the reactions.
Reactions are ranked. "Can we do this?" is worth more than a forward. A forward is worth more than a reply. A reply is worth more than an emoji. The top one is what the agent is really looking for: the moment someone finds out the stack can already do something they didn't know about. That gap, between what the tools provide and what the team knows they provide, is where most of the experiments come from.
The same agent reads the SigNoz community Slack. How customers react to changelog posts and migration guides is a signal source that costs nothing.
Nobody's Slack is being read without their knowledge. The channel is seeded on purpose. The team is the first dataset.
Every reaction produces a candidate. The closed-won loop decides which candidates were real.
Who this is for: someone running marketing across several companies doesn't have the hours to sit in Slack and read how each team reacted. The digest does that. When they want to run an experiment, they read the digest and pick.
Two experiments a month
The lab has an input and an output.
Input is what the team already says. Standups, recorded with the team's agreement, the same way the channel is seeded on purpose. Internal conversations people chose to share. The problems raised when nobody's presenting. The agent reads those the same way it reads the seeded channel, and it's looking for the same thing: a problem someone named that the stack could already answer.
Output is content. Each problem that comes up more than once becomes a post, a guide, or a comparison page. Push it out. Watch who reacts, inside the team and in the community Slack.
From that pool, two experiments a month. Full ones. A hypothesis, a split, a number to read, a date to read it. Not five half-run tests. Two that finish.
The closed-won loop decides which of the two mattered.
Tools
HubSpot, Slack, Clay, Otter.ai for call transcripts.
Source: signoz_assignment_progress.md, Q6 sections from 5 and 6 Sep 2026.