Your business runs on manual work it should not.

// the problem

Eight agents were running roughly 40 Zoom Phone calls a week. The recordings piled up in Zoom's cloud and nobody had time to listen back, so follow-ups depended on whoever remembered what was promised. Deals were quietly dying in the gap between the call and the CRM.

// what I built

An n8n workflow fires the moment Zoom marks a recording complete. It pulls the audio, transcribes it, then has an AI model extract the commitments — what was promised, who owns it, when it is due. Each one becomes an assigned Planka card, a summary lands in the team channel, and the full transcript is archived and named so any call can be found by client later. Failed runs retry, then alert a human instead of disappearing.

// the result

• 40calls a week, processed unattended
• 0missed follow-ups since launch
• ~6 hrsof admin returned each week

// the problem

Cold email was spread across 19 Gmail accounts to protect deliverability, and two assistants checked them by hand. Interested replies routinely sat for two days before anyone saw them, and some were never found at all — the cost of every missed one was a booked call that never happened.

// what I built

n8n polls all 19 mailboxes, de-duplicates threads, and runs each reply through an AI classifier: interested, not now, out of office, auto-reply, or unsubscribe. Each one is enriched with the campaign and prospect context, then posted into a single Discord channel as a colour-coded card the team can action immediately. Unsubscribes are suppressed automatically across every account, and any failure posts itself to an ops channel rather than failing silently.

// the result

• 2 days → 20 minmedian reply time
• 100%of replies captured and classified
• 2assistants freed for real conversations

// the problem

Around 1,200 people a month finished the skin-type quiz and then hit a generic thank-you page. The answers sat in Jotform, Klaviyo received nothing, and every one of those hard-won responses was a paid lead thrown away.

// what I built

A Jotform webhook hands each submission to n8n, which normalises the answers and maps them onto one of seven skin profiles. The matching Klaviyo profile is created or updated with the quiz properties, the right flow is triggered, the Shopify customer record is tagged, and three products are recommended based on that profile. Replenishment and back-in-stock messaging then follow the same segmentation instead of going out to everyone.

// the result

• 7personalised flows running live
• 1,200/moquiz takers now entering a sequence
• Month 1quiz revenue passed a full quarter of newsletter revenue

// the problem

New patients were emailed a fourteen-page PDF. About 40% never came back completed, staff re-keyed the rest into the practice management system by hand, and protected health information was travelling as email attachments — the part that kept the practice manager awake.

// what I built

HIPAA-enabled Jotform intake with real conditional logic, so a patient only ever sees the sections that apply to them — roughly a 60% cut in visible length. Consent is captured with Jotform Sign, submissions stay in encrypted storage, and n8n routes each one into the practice management system, notifies the assigned clinician, and chases anyone who abandons partway. No PHI ever touches an email body.

// the result

• 60% → 94%intake completion rate
• ~5 hrsof weekly re-keying eliminated
• 100%consent signed before first session

// the problem

Forty-plus Zaps were pushing around 380,000 tasks a month and the bill had climbed past $1,100. Multi-step Zaps were timing out under load, there was no version control and no staging — every change was made live and hoped for. They asked me to fix a broken Zap; the honest answer was that Zapier had become the wrong tool at their volume.

// what I built

I audited all forty Zaps and consolidated them into twelve n8n workflows, using branching where Zapier had forced duplication. n8n now runs self-hosted on their own infrastructure in queue mode with a Postgres backend, split into staging and production, with every workflow exported to git so changes are reviewable. Failed executions raise alerts instead of being noticed a week later.

// the result

• ~83%lower monthly cost ($1,100 → ~$190)
• 40 → 12workflows to maintain
• 0timeout failures since migration

// the problem

The WordPress site was slow enough that visitors left before it finished painting. Thirty-eight plugins, unoptimised hero images and render-blocking scripts were the cause, and the firm did not appear on the first page even for its own name — prospects handed the name by a referral simply could not find them.

// what I built

A rebuilt theme with twenty-one plugins removed, images converted to WebP with responsive sources, critical CSS inlined, non-essential JavaScript deferred, fonts self-hosted, and caching plus a CDN in front. On the search side I fixed the canonical and schema errors, rebuilt the sitemap, and mapped redirects for every old URL so nothing lost its history. Everything was handed over documented.

// the result

• 96 / 100Lighthouse performance, desktop
• 1.1 slargest contentful paint, desktop
• 0.02cumulative layout shift, mobile