How we get each inbound lead to the right rep
Someone fills in your demo form. Right now they are curious, maybe even comparing you with a rival in the next tab. An hour from now they have moved on. Speed is most of the game with inbound, yet in many teams the lead just waits: for somebody to open the queue, recall which region belongs to whom, and work out who is next in the rotation. Asking a person to do that well, every quarter hour, from nine to six, is asking a lot.
So we gave the job to an agent on Machine. Four times an hour it pulls the new leads out of HubSpot, picks an owner for each, tells that rep in Slack, and sets aside any lead it can't place with confidence so a human can decide. Those are its only moves: it gives leads an owner and sends messages. Deleting or merging records is off the table.
The problem
Our routing logic is spread across three layers. Regions map to territory owners. Certain company sizes and industries go to specialists. Anything left over drops into a round-robin pool. Whoever works the queue has to keep all three straight, apply them without slips, and still move quickly enough that the prospect hasn't gone cold by the time the rep hears about it.
Most teams fall back on a fixed rule chain built inside the CRM. It holds up until real life happens: a territory is carved in two, a rep takes parental leave, a lead straddles two segments. Then leads quietly pile up, land on whatever default owner the builder picked years ago, or get a reply days late when somebody finally spots them. None of that is quick. Worse, the leads that genuinely need a person's call are the ones nobody hears about.
What we built
A schedule wakes an agent on Machine every 15 minutes. Each run gets its own fresh cloud computer with limited access to HubSpot. The agent collects every inbound lead that hasn't been routed yet and scores how keen it looks. Then it checks the territory rules and the segment rules. If neither claims the lead, it goes to whoever in the round-robin pool has had the fewest leads this week.
Once it has an owner, the agent sets that owner in HubSpot, drops a note in the rep's Slack channel, and highlights strong-intent leads so they get called first. When it can't be sure (the territory field is blank, no segment fits, or two reps are dead level) the lead goes to a separate Slack channel where a person assigns it. No lead is ever deleted or merged.
How it works
Wake up every 15 minutes
A cron trigger is just a timer. Each time it fires, Machine starts a new session: a clean cloud computer, or sandbox, that exists only for that run and keeps nothing on disk. Instead of a local list, the agent leans on HubSpot itself. A routing-status field on each lead says "done", so the following run skips anything already handled and only sees leads that are truly new.
Hand it the rulebook
Who owns which region, which specialist covers which segment, and who sits in the rotation are kept as the agent's skills and memory: a playbook it carries into every run, rather than logic hidden in a CRM flow. Redraw a territory or add a new rep to the pool, edit the playbook, and the next run already follows the updated rules. Nothing in HubSpot needs rebuilding.
Plug in HubSpot and Slack
The agent reaches your tools through a connector, a narrow connection whose keys stay on our servers, so the model never handles a raw token. With it, the agent can:
- Look up fresh leads in HubSpot. It sees stage, region, headcount, industry, job title and how engaged the person has been.
- Record an owner and a "routed" flag in HubSpot. Those two fields are all it ever writes. It cannot remove or combine records.
- Message Slack. The rep gets an assignment note, hot leads get their own marker, and anything unplaceable gets posted to an escalation channel.
Draw the lines
Routing follows the written rules and nothing else. Under no circumstances will the agent delete a lead or merge two together. When a lead is unclear or falls outside every rule, it does not pick someone at random; it hands the lead to a person. Credentials are stored encrypted in the Secrets Manager and only handed to the agents you grant them to.
Route, ping, escalate
Put together, each quarter hour brings one sweep of the queue. Every new lead reaches the correct rep along with a short note on why it went to them. Hot ones are flagged so they jump the line. The truly unclear ones sit in their own channel until a person assigns them. A lead can go one cycle at most without an owner or someone knowing the reason.
The pattern
A cron every 15 minutes starts a clean session that reaches HubSpot through a limited connector and talks to Slack through a channel. Territory, segment and rotation rules are the agent's skills and memory. It routes and pings strictly by those rules, and any lead it isn't sure about goes to a person.
Guardrails
Because this agent changes CRM records, we kept what it may write very small and made its route to a human obvious:
- No deleting, no merging. It may fill in a lead's owner and the routing flag. Removing a lead, combining duplicates or editing any other field is outside its reach, however sure the rules make it feel.
- Unclear goes to a person. If no territory fits, no segment fits and the round-robin can't break a tie, the lead is posted in a dedicated Slack channel for someone to assign. The agent never makes up a default owner.
- Locked-down keys. The HubSpot key is stored encrypted in the Secrets Manager and only handed to the agents you grant it to, at the moment the sandbox needs it.
- Rules live in the repo. Territories, segments, the rotation and the intent-scoring criteria are versioned files. Changing them means a reviewed change request, not clicking around a CRM workflow editor.
The outcome
Leads no longer sit around waiting to be noticed. Within 15 minutes each one is with the right rep, with a line explaining why and a flag if it deserves a call first. The agent does the sorting and the pinging. Any lead it can't place with confidence still belongs to a human.
More use cases
How we answer GDPR data-subject requests before the deadline
Our Machine agent checks each GDPR access or deletion request, finds the person's data across our product database, builds the report well within the SLA and hands it to legal before anything is sent.
How we keep our docs matching what the code does
Once a day a Machine agent reads every commit merged since its last pass, finds the doc pages those changes made wrong, rewrites them and sends you one pull request to review.
How we test every pull request without lifting a finger
Our Machine QA agent works with GitHub and our test environment. For each pull request it checks out the branch, runs the tests, tries the change for real, and posts what it found.
Put your first agentto work today.
- SSO
- RBAC
- Governed agents
- Full audit trail