Automation case studies, problem to result

Challenge

Leads from listing portals, the website and Facebook ads landed in a shared inbox and were picked up whenever an agent had a free moment. Average first response was close to three hours, and evening and weekend leads often waited until the next morning. By then many buyers had already spoken to a competing brokerage, and nobody could say how many leads were lost.

What we built

A Make scenario collects leads from every source and logs them in GoHighLevel. Within 90 seconds, a voice agent built on Retell AI calls the buyer, confirms budget, area and timeline, and offers viewing slots from the listing agent's Google Calendar. If the call goes unanswered, a Twilio text follows up. Qualified leads trigger an instant Slack alert to the right agent, with a short call summary attached.

Challenge

Around 180 small business clients sent receipts, bills and bank statements by email, WhatsApp and shared folders, in every format imaginable. Junior staff spent the first week of each month renaming files, typing totals into QuickBooks and Xero, and chasing missing documents. Reconciliation started late, errors surfaced during review, and the firm could not take on new clients without hiring more people for data entry.

What we built

An n8n workflow pulls every incoming file from email, the WhatsApp Business API and shared folders into one queue. Google Gemini reads each document, extracts vendor, date, amount and tax, and files it under the right client. Clean items are pushed to QuickBooks or Xero as draft transactions ready for bank reconciliation, while low-confidence extractions and missing statements land on an Airtable review board for a bookkeeper to check.

Challenge

Customers messaged on WhatsApp, Instagram and email, and more than half of all questions were some version of 'where is my order?'. A three-person support team copied tracking numbers from Shopify into courier websites by hand. During sale periods replies slipped to a full day, refund and exchange requests got buried, and negative reviews started to mention slow support.

What we built

A chatbot on the WhatsApp Business API, orchestrated in n8n, answers in English and Arabic using OpenAI models grounded in the brand's own policies and size guides. It checks live order status in Shopify and the courier's tracking API, and answers returns questions within set rules. Refunds, damaged items and unhappy customers are handed to a person with the full conversation attached. Weekly support reports run in Looker Studio.

Challenge

Drivers photographed signed proof of delivery (POD) notes and sent them by WhatsApp or email, or handed in paper copies at the end of the week. Two admin staff matched each POD to a job in the transport system before an invoice could go out in Xero. Invoicing ran five to eight days behind deliveries, and missing PODs regularly delayed payment from large customers.

What we built

Drivers upload POD photos through a simple AppSheet form or by WhatsApp. An n8n workflow uses Anthropic Claude to read the job reference, consignee name, signature and any damage notes, then matches each POD to its job in the transport system. Clean matches create a draft invoice in Xero with the POD attached. Missing signatures, mismatched references and damage claims go to an admin review queue in Microsoft Teams.

Challenge

Each nursing or care role attracted 80 to 200 applications through job boards and the agency's website. Recruiters skimmed CVs by hand, checked registrations and right-to-work details, then spent days trading emails to set up phone screens. Strong candidates were often hired elsewhere before anyone called them, and recruiters had little time left for the client relationships that actually win placements.

What we built

A screening agent built with LangGraph and Anthropic Claude reads each CV against the role's must-have criteria, such as registration type, shift availability and location, and writes a short scored summary into the agency's Pipedrive pipeline. Recruiters review and approve every shortlist. Approved candidates get a text and email via Twilio with a link to book a phone screen in the recruiter's Microsoft 365 calendar, plus automatic reminders.

Challenge

Front desk teams across four practices were juggling check-ins, insurance questions and a steady stream of calls. Roughly one in four calls went unanswered during busy mornings and lunch breaks, and after-hours callers reached a voicemail that few people returned. New patient calls were the most costly to miss, since those callers usually booked with whichever practice picked up first.

What we built

An AI receptionist built on Vapi with an OpenAI model answers overflow and after-hours calls in a natural voice. It handles common questions, books, moves and cancels appointments through the practice management system's scheduling API, and texts confirmations via Twilio. Insurance and clinical questions go to staff with a call summary. The setup is HIPAA-aware, with transcripts stored in the group's own Supabase account and call reports in Looker Studio.