If you need the tool basics first, read What is n8n? This piece assumes you can build a workflow and are trying to decide which one.
The MitHub workflow card
Before building anything, fill in six lines. It takes five minutes and kills roughly half of all proposed workflows, which is the point.
| Line | Why it matters |
|---|---|
| Metric it moves | If blank, this is a hobby |
| Trigger | Real-time or scheduled — they fail differently |
| Steps | The shortest version that works |
| Owner | A named person, not a team |
| Failure mode | How it breaks silently |
| Proof artifact | What you will show in a month |
The line people resist is failure mode. Every workflow has a way of breaking that raises no error: a webhook still pointing at a test URL, an enrichment returning empty, a record stuck in an intermediate state. Write it down while you still remember how the thing works.
Build order: measurable before clever
Build in this sequence, not in order of excitement.
- Reporting first. It changes nothing and measures everything. It is also your baseline — without it, no later workflow can be proven.
- Then the biggest leak. Found by tracing a real sale backwards, which is what MitHub's Follow the money chapter teaches.
- Then the workflows that protect the first two: data quality, error alerts, sweeps.
- Then the clever ones.
Teams that invert this end up with an impressive canvas and no answer to "did it work?"
1. Daily revenue snapshot
Metric: all of them — this is the instrument, not the engine. Trigger: Schedule, every morning before the team starts. Steps: Pull counts from the CRM (new leads, leads contacted, time to first touch, stage movements, closed) → append a dated row to a sheet or database → post a short summary to the team channel. Failure mode: it silently reports yesterday's numbers when a query breaks. Include the row count and the date range in the message so a stale report looks stale. Proof: a time series. In six weeks you have a chart, and a chart is what turns work into a case study.
Append; never overwrite. History you did not keep cannot be recovered.
2. Speed-to-lead responder
Metric: median minutes from lead created to first contact attempt. Trigger: Webhook from the form or CRM. n8n's Webhook node gives separate test and production URLs, and lets you respond immediately rather than waiting for the whole workflow to finish (n8n Docs) — respond immediately, so the form never hangs on a slow enrichment step. Steps: Receive → normalize and stamp received_at → deduplicate → create or update the record → assign an owner → fire the first touch (alert, call task, message) → log the elapsed time. Failure mode: the source system is still pointed at the test URL, so leads vanish on every day you are not actively testing. Proof: the distribution of first-touch times before and after, not the average. Averages hide the tail, and the tail is where the money leaks.
Harvard Business Review's 2011 study of online lead response found firms contacting a prospect within an hour were nearly seven times as likely to qualify the lead as those attempting even an hour later (Oldroyd, McElheran and Elkington). The full picture, including what that study does and does not prove today, is in speed to lead.
3. Routing with an SLA escalation
Metric: percentage of leads worked within the SLA window. Trigger: CRM record created or updated. n8n's Salesforce Trigger node, for example, documents events including On Lead Created and On Lead Updated (n8n Docs). Steps: Determine the owner from territory, product or round robin → assign → start an SLA timer → if untouched after N minutes, escalate to a manager → if still untouched, reassign and log the miss. Failure mode: a lead matches no routing rule and sits ownerless. Always define a fallback owner, and make "no owner resolved" raise an error rather than pass through quietly. n8n's Stop and Error node exists to force failure on conditions you choose (n8n Docs). Proof: a weekly count of SLA misses by branch or rep, trending down.
Routing is where most "we need more leads" conversations quietly end. See lead routing.
4. Enrichment round-trip
Metric: percentage of records with the fields needed to act (a reachable phone, a valid email, a size signal). Trigger: New or updated CRM record, or a scheduled batch of incomplete records. Steps: Select records missing key fields → send to the enrichment tool → receive results on a second webhook → write back only non-empty values → mark the record enriched with a timestamp. Failure mode: enrichment returns nothing and the workflow writes blanks over data that was already there. One condition — only write when the value is non-empty — prevents it. Proof: field completeness percentage over time, and the cost per enriched record.
On which tool should own which half of this, see Clay vs n8n.
5. Data-quality quarantine
Metric: wasted contact attempts (calls to dead numbers, emails to bounced addresses). Trigger: A failure signal — a bounce, an invalid number, a disconnected call result. Steps: Flag the record → remove it from automated outreach → route to re-enrichment or human review → return it to circulation only when the field is fixed. Failure mode: the quarantine has no exit. Records go in and never come back, and nobody notices the list shrinking. Report the quarantine size weekly. Proof: bounce and bad-number rate falling, plus a quarantine queue that empties.
This is the least popular workflow on the list and often the highest return, because bad data makes every other system look broken.
6. After-conversation processing
Metric: percentage of conversations with a logged outcome and a scheduled next step. Trigger: Webhook from the calling platform, meeting recorder or email tool when a conversation ends. Steps: Receive the transcript or result → one AI step extracts outcome, objection and next step into fixed fields → update the CRM → create the follow-up task or booking → notify the owner. Failure mode: the model returns a value outside your allowed list and it flows into the CRM unchallenged. Validate the output against your allowed values and route anything unexpected to a human. Proof: the share of conversations with a next step booked, before and after.
This is the one place on the list where a language model clearly earns its keep: the input is unstructured human speech and the output is four fields. Keep it to one AI step. See n8n for AI automation for how to structure it.
7. Stale pipeline sweep
Metric: number of records sitting in a stage past its expected duration. Trigger: Schedule, every morning. Steps: Query records with no activity in N days, by stage → apply rules (nudge the owner, move to nurture, flag for review) → post one digest to the team channel, not one message per record. Failure mode: notification fatigue. If the digest lists forty items every day, people stop reading it and you have built noise. Cap it, sort by value, and tune the thresholds until the list is short enough to act on. Proof: average age of open records trending down.
The sweep has a second job: it catches records that got stuck because another workflow failed without raising an error. It is the safety net under everything else on this list.
8. Reactivation trigger
Metric: revenue from previously closed or dormant records. Trigger: A defined event that makes a past contact relevant again — a contract anniversary, a loan paid off, a renewal window opening, a role change. Steps: Detect the event → verify eligibility and contact permission → enrich if the data is old → queue the outreach → route any response to a human immediately. Failure mode: contacting people who asked not to be contacted. Check consent and suppression lists before the outreach step, never after, and confirm what the rules require in your market and channel before you build. Proof: conversion rate of the reactivated segment against a comparable cold segment.
For anything irreversible or customer-facing, put a person in the path. n8n's human-review pattern pauses a workflow and sends an approval request through a channel such as Slack or email, running the action only if approved (n8n Docs).
Do not build these yet
- A multi-agent system before a single workflow runs reliably for a month.
- Anything sending external messages before consent, suppression and channel rules are checked by a workflow, not by memory.
- A second workflow that writes the same field as an existing one, until you have written the field ownership rules. See CRM automation with n8n.
- A replacement for a report someone already trusts. Add to it instead. Trust is expensive to rebuild.
What scale looks like
MitHub's pioneers have built AI voice campaigns that ran across 28 live branches of a multi-location lending business, including a 10-branch pilot with 13,159 AI calls. At that volume, none of the eight workflows above is optional — the sweep, the quarantine and the daily snapshot stop being nice practice and become the only way anyone knows what is happening.
The one rule to carry over
A green execution is not a business result. It means n8n finished running. Open the CRM record, look at the field, confirm the human got the message. Every workflow on this list should end with something a non-builder can read: a log row, a digest, a dated line in a sheet.
That habit is also what makes your work provable. "I built eight workflows" is a list. "First-touch time went from this to that, and here is the chart" is a case study — which is how MitHub's learn, build, prove, earn path actually converts into work. Start free at the Faculty of Revenue Reverse Engineering.