Skip to content

AI workflows

Automations we run ourselves, written out step by step with the exact stack, the review points and what the output actually looks like.

Weekly competitive digest delivered to Slack

A sourced summary of competitor moves lands in Slack every Monday at 08:00.

Intermediate~3h saved / month

Stack: n8n + perplexity + claude · Industries: SaaS, Agencies, Consulting

  1. Schedule the runA cron trigger fires every Monday at 07:30 in the team's timezone.
  2. Collect signalsFetch competitor changelogs, pricing pages and news mentions from the last seven days.
  3. Research with citationsSend each signal to an answer engine to confirm the change and capture a source URL.
  4. SummariseA long-context model writes a 300-word digest grouped by competitor, dropping anything unverified.
  5. Deliver and archivePost to Slack with source links and append the digest to a Notion database for trend analysis.

Example output: A Slack message with three competitor sections, each claim linked to its source, plus a one-line 'what changed for us'.

Support ticket triage and routing

Inbound tickets are classified, tagged and routed within 60 seconds of arrival.

Beginner~7h saved / month

Stack: n8n + chatgpt · Industries: SaaS, Ecommerce, Fintech

  1. Watch the inboxA webhook receives every new ticket from the helpdesk.
  2. ClassifyAn LLM assigns category, urgency and product area against a fixed enum, returning UNKNOWN when unsure.
  3. EnrichLook up the customer's plan and recent tickets to attach account context.
  4. RouteAssign to the right queue; anything marked UNKNOWN or high urgency goes to a human first.
  5. Draft a replyGenerate a suggested response for the agent to review — never auto-send.

Example output: A tagged ticket with plan context, correct queue assignment and a draft reply awaiting agent approval.

Course localisation into Hindi and Gujarati

An English course module ships in three languages with narration in under a day.

Advanced~16h saved / month

Stack: claude + elevenlabs + n8n · Industries: EdTech, Enterprise L&D, Publishing

  1. Extract the scriptPull narration text and on-screen copy from the source module.
  2. Translate with glossaryTranslate using a locked terminology glossary so product names stay consistent.
  3. Native reviewRoute each language to a human reviewer with tracked changes before audio generation.
  4. Generate narrationSynthesise approved scripts with a consistent cloned voice per language.
  5. Reassemble and QARebuild the module, check timing drift and publish with localised metadata and hreflang.

Example output: Three published module variants with reviewed subtitles, matched narration and localised page metadata.