Anonymised summaries of recent engagements at Digital One Agency. Client names withheld under NDA. All pricing, timelines, tech stacks and outcomes are real and reflect typical results for similar projects in Australia and the UK.
Want to see something more relevant to your industry? Book a free 30-minute discovery call — we’ll walk through 2–3 client engagements that match your use case and share metrics on a call.
Case Study 1 — AI Voice Agent for a Multi-Branch Trades Business (Sydney)
Client: Australian trades franchise with 8 branches across NSW. Anonymised at client request.
Problem: 31% of inbound after-hours calls were going to voicemail and never returned. Customers were going to competitors who answered. The franchise needed every call answered, qualified and booked into the right branch’s calendar — without hiring 24/7 receptionists.
What we built:
- Custom AI voice agent on Vapi + Claude Sonnet, with a fallback to human dispatch for emergencies.
- Branch-aware routing — agent identifies suburb, routes to nearest branch’s calendar (ServiceM8 integration).
- Qualification flow — captures job type, urgency, address, contact, photos via SMS follow-up.
- Multilingual — English and Mandarin (significant Chinese-speaking customer base in target suburbs).
Tech stack: Vapi · Claude Sonnet · Twilio (Australian local numbers) · ServiceM8 API · Twilio SMS for photo capture · custom Node.js orchestration.
Timeline: 6 weeks from kick-off to live.
Investment: AUD $32,000 fixed-scope build + AUD $2,400/month optimisation retainer.
Outcome (90 days post-launch):
- After-hours call abandonment dropped from 31% to 4%.
- Booked-job conversion from after-hours calls increased 38%.
- ~AUD $11,000/month in additional booked revenue attributable to recovered calls.
- ROI on build cost: ~3 months.
Case Study 2 — Two-Sided Marketplace for a UK Property Niche (London)
Client: UK property-services platform connecting landlords with vetted contractors. Anonymised under NDA — UK-VAT registered, London-based founders.
Problem: Founders had validated demand on Sharetribe, hit £80k MRR, then outgrew the platform — couldn’t customise vendor onboarding, KYC, payment splits, or AI matching. Platform fees were eating margin and product velocity was zero.
What we built:
- Bespoke marketplace with custom KYC, GDPR-aware vendor onboarding, multi-step verification.
- Stripe Connect (UK) Custom Connected Accounts with escrow, instant payouts, refund flows.
- AI matching — vector-search-driven contractor recommendation using job description embeddings.
- Native mobile apps (iOS + Android, Flutter) for contractor side; web for landlord side.
- Admin dashboard with dispute resolution, fraud signals, vendor performance scoring.
Tech stack: Next.js · NestJS · PostgreSQL · pgvector · Stripe Connect UK · Cloudflare Workers · Flutter · Claude (for matching reasoning) · Twilio for SMS.
Timeline: 19 weeks discovery-to-launch, including data migration from Sharetribe.
Investment: £128,000 fixed-scope build + £6,500/month optimisation retainer.
Outcome (6 months post-launch):
- Vendor onboarding time cut from 9 days to under 24 hours.
- AI matching lifted job-to-quote conversion 27%.
- GMV grew ~3.4× year-on-year, attributable to product velocity post-migration.
- Platform fees eliminated — direct margin impact ~£14k/month.
Case Study 3 — Custom AI Agent for a Brisbane Healthcare Group
Client: Australian multi-clinic healthcare group, ~40 practitioners. Anonymised — health-data sensitive.
Problem: Clinical admin staff were spending 18+ hours per week on routine document review — referrals, results, prior-care summaries. Practitioners were waiting for admin to surface key info before consults, slowing throughput.
What we built:
- Self-hosted AI document agent (no third-party LLM cloud, given health-data sensitivity).
- RAG pipeline over the practice’s historical referral and results corpus.
- Pre-consult summary generation — surfaces relevant prior care, allergies, flags exceptions.
- Practice-management-system integration (Cliniko) for in-context display.
- Audit logging and OAIC-compliant audit trail for every AI inference.
Tech stack: Self-hosted Llama 3.3 70B (NVIDIA H100 inference) · Qdrant vector DB · LangGraph orchestration · Cliniko API · custom Python FastAPI · audit-logging via PostgreSQL.
Timeline: 14 weeks build + 4 weeks soft launch in parallel with manual process.
Investment: AUD $78,000 build + AUD $3,800/month inference & retainer.
Outcome (4 months post-launch):
- Admin time on document review reduced from 18 hours/week to 4 hours/week.
- Pre-consult summaries available in < 90 seconds vs previous 4–6 hour wait.
- Practitioners reported ~12 minutes saved per consult, enabling ~2 additional consults per practitioner per day.
- Estimated annualised value: AUD $640,000 in recovered practitioner capacity.
See If Your Project Is a Fit — Free 30-Minute Discovery Call
The case studies above are typical of how we work — fixed-scope, milestone-billed, senior-engineer delivery. If you’re considering a similar build, the 30-minute call is free, you’ll get an architectural recommendation in writing, and you’ll know whether we’re the right team for it.
- Senior engineer on the call from minute one
- Live references on request after fit confirmed
- Written architectural recommendation, not a sales deck
- Fixed-scope quote — no hourly billing