Inbound lead desk
- HubSpot
- Salesforce
- Apex
- Slack
- lemlist
- Before
- 29.7% touched · 44 h median first touch
- Action
- Qualify, route by state, Slack alert, Day 0/2/4 tasks
- After
- Being measured
- Verified by
- Real-data replay 101/101
The problem
Inbound demo requests were arriving and then mostly sitting. I measured the starting point before building anything:
- 222 inbound leads from April to September.
- 66 of them (29.7%) had any logged human touch from BD.
- Median first touch: 44 hours. The average was 156 hours, about 6.5 days.
- Only 14.0% were reached within 24 hours.
Upstream there was a second gap. HubSpot only syncs a contact to Salesforce after it is marked marketing-qualified, which puts it on the sync list. Of 11,264 contacts created in HubSpot over six months, only 188 (1.7%) were ever marked qualified. The rest were created once and then never updated in either direction, with no error anywhere.
Constraints
- The sync door has an order. HubSpot only syncs changes made after a contact joins the sync list. Write the flag and the enriched details in one call, and the details never reach Salesforce.
- Two minutes to decide. An existing HubSpot workflow routes demo-form leads by US state about two minutes after the form. The engine had to set the state and the qualification first.
- Never take a pass away. The engine only ever writes "qualified = true". It never removes someone a person or a workflow let in.
- Don't overwrite people. Every write fills blanks only, re-read just before writing. A rep's value beats a vendor's.
- The baseline was messy. Much work happened off-system, so it undercounts touches.
How it works
- Catch every new contact. A HubSpot webhook registers each new contact, and an hourly sweep re-checks the last 26 hours, so a dropped delivery loses nothing.
- Look up the person and the company. FullEnrich for the person, and a company-data provider only when it changes the answer. A missing state and country are filled from a fixed order of sources.
- Decide. A contact qualifies if it sits on a target account, or if the company is a US health plan or other US healthcare company. When the company type is unknown, an LLM classifies it and gives a one-line reason.
- Pass first, then values. The desk writes the qualified flag, waits until the contact is on the sync list, then writes the enriched details. Five minutes later it reads Salesforce back to check they landed.
- Route by state. A qualified US lead goes to the BDR who covers its state. The state roster lives in a table, not in code.
- Post to Slack. Each new demo or ROI-calculator lead gets a post built for a five-second read: verdict, owner, form, timing and reason. Only qualified leads tag anyone.
- Run the cadence in Salesforce. Day 0 is a call due one business hour after arrival (9 to 5 ET). Day 2 and Day 4 each need a call and an email. A real touch closes a step on its own; a reply, a meeting or a conversion stops the cadence. If Day 6 passes with no response, the lead closes with a stated reason.
A new inbound contact is enriched and qualified by the lead desk; qualified leads get the sync pass first, then enriched values, a BDR chosen by US state, and a Day 0/2/4 cadence in Salesforce, while every lead gets a Slack post with the reason.
Receipts
The cadence went live on 24 October 2026, with a kill switch. Before that I tested it at every layer in a sandbox:
- Apex unit tests: 9/9, including a daylight-saving week and a 200-lead bulk insert.
- End-to-end: 42/42, on a simulated clock with production-shaped call and email rows.
- Lead stage flow suite: 93/93, to prove existing stage logic still worked.
- Real-data replay: 101/101 real production lead histories.
When I narrowed entry to demo and calculator form fills, the suites grew to Apex 10/10 and end-to-end 48/48.
What broke, and what I changed
- Auto-replies counted as touches. Out-of-office replies looked like first touches in the baseline. I dropped 6. My first filter also caught real rep emails, so I narrowed it.
- A planned BDR was never assigned. On joining the sync list, the connector set a stand-in owner within the minute, and my fill-blanks rule skipped the BDR. Now the stand-in counts as no owner.
- A stale state code blocked routing. A lead kept an IP-based state code under the state the desk wrote, so it got no BDR. Now the code moves with the state.
- Unqualified leads still tagged people. A state employee on a shared government domain was matched to a university, and a non-healthcare lead abroad got a BDR through its company's HQ. Now only qualified leads get a BDR or a tag, and shared government domains are never used to find an employer.
- The form answer was trusted too much. A software company picked "Health Plan" and was posted as healthcare. Now a confident non-healthcare verdict beats the form.
- Salesforce moved tasks silently. An owner change hands open tasks to the new owner without firing task triggers, which stranded steps on integration users. A follow-up job hands them back to the last real BDR.
What I'd do next
- Measure the after. Re-run the same first-touch analysis on the post-launch cohort.
- Settle what counts as a touch. It is still open whether a held, recorded meeting counts as a first touch.
- Revisit calls. A call moves a lead to Working but never to Engaged. That rule is due for review once all calls run through the sequencer.