Automation systems,
documented properly.

The client was paying enterprise CRM prices for something that behaved like a shared list. Contacts and deals lived in HubSpot and nothing happened to them on their own: no routing, no follow-up, no alerting. Every action in the pipeline depended on a person remembering to take it.

Records were updated by hand. Enquiries got worked by whoever happened to notice them, follow-up ran on memory, and there was no mechanism that made an untouched lead visible to anyone. Nothing was automated, so the CRM was effectively a spreadsheet with a licence fee attached.

• Full database migration off HubSpot into GoHighLevel, roughly 1,600 contacts and 1,000 deals, with pipeline structure and history carried across intact
• Pipeline stages that move on real events rather than manual updates
• Acknowledgement and quote-chasing sequences that run without anyone starting them
• Missed-call text-back, so a call nobody could answer gets a reply inside a minute
• Reactivation, review and referral sequences on the back of completed jobs
• Voice AI answering after-hours calls, and a chatbot across web, SMS and social
• Two-way Xero sync, so paying an invoice moves the deal stage, with no one re-keying it
• Persistent lead alerts into Slack that keep firing until the contact is marked as contacted
• Architecture aligned to their NDIS compliance requirements

• GoHighLevel
• n8n
• Xero
• Meta Lead Ads
• Slack
• HubSpot (migrated from)

The database moved across intact and the CRM stopped being a passive record store. Follow-up now happens whether or not anyone remembers it. The alerting repeats until a lead has been contacted, so enquiries no longer sit untouched because nobody saw them. Calls that arrive out of hours get answered or texted back rather than lost. The system has been running in production ever since, and is still supported today.

The business was losing every single customer who clicked cancel. Nobody caught them, nobody called, and nobody offered them anything. The cancellation just went through. Every one of those customers had already been paid for once, and was walking out without a conversation.

Cancellations were handled entirely by the billing system. There was no trigger, no queue and no owner, so a save depended on somebody happening to notice a churn report after the fact, which is far too late to do anything about it.

• A cancellation event that fires the moment someone clicks cancel, rather than showing up in a report weeks later
• Automated outreach to the leaving customer while the decision is still fresh
• Retention offers routed by the reason given, so the response fits the reason for leaving
• Lifecycle email sequences running behind the flow

• n8n
• Lifecycle email

The flow now recovers two to three customers every week. Before it existed the same process saved none of them. The difference is not a better offer, it is that somebody reaches the customer at all, automatically, at the moment they leave.

The account contained 153 inherited workflows and no map of what they were for. Some were live, some were half-finished, and nobody could say which was which. Meanwhile the booking funnel had stopped producing bookings, silently, with no error anywhere.

Nobody touched anything, because nobody knew what would break. An account in that state stops being an asset and becomes a thing the business works around, and the cost of it is invisible precisely because nothing is throwing an error.

• A complete audit of all 153 workflows: what each one does, whether it runs, and what depends on it
• 11 real findings, each ranked by what it was costing the business
• Root cause found on the booking funnel: a configuration fault that was showing no available times, so visitors who tried to book could not
• A prioritised order of work, so the expensive problems got fixed first

• GoHighLevel
• Reporting

The silent failure was the important one. The booking funnel had been turning away everyone who tried to use it, and nothing in the account reported a problem because technically nothing had failed. The audit found it, ranked it against the other ten findings by cost, and fixed it.

The business email was silently dependent on DNS records the migration was about to change. Nothing in the plan accounted for it, and nobody had noticed. It is the kind of dependency that is invisible right up until the moment every inbox stops receiving.

The migration was ready to proceed on the assumption that moving a website only affects the website. It is a reasonable assumption and a wrong one, and the failure it produces is both total and hard to reverse quickly.

• Migration halted before cutover, rather than after the discovery
• Every email dependency on the existing DNS traced and verified first
• Cutover sequenced so mail delivery never had a gap
• The old administrator kept reachable throughout, so nothing was stranded behind the change

• DNS
• WordPress
• GoHighLevel

The migration completed with email continuity intact and no interruption to delivery. The valuable part was not the migration. It was stopping to check, because this failure is unrecoverable in the moment and completely avoidable an hour earlier.

Identity data needed real isolation rather than a column flag, and nothing client-facing could be allowed to ship on an automated pipeline alone. The usual startup defaults, one database and deploy on merge, were the wrong defaults here.

The requirement was not "move faster". It was to make the dangerous actions impossible to take by accident, without making ordinary development miserable.

• Identity data held in a physically separate database with its own isolated credentials
• A human approval gate in the release state machine, so nothing client-facing ships unreviewed
• A crisis check that runs before payment is taken, not after
• An end-to-end integration test suite covering the paths that matter

• NestJS
• Next.js
• PostgreSQL
• Prisma

Built, proven end to end by the integration suite, and running. Included here because it shows the same instinct as the automation work: find the failure that nobody would notice until it was expensive, and design so it cannot happen quietly.