Skip to main content
← Back to Blog
·10 min read·By Alex Chirila

How We Actually Build AI Automation (The Stack Behind SoftSync-AI)

How we actually build AI automation at SoftSync-AI

“What tools do you use?”

I get this question in every demo call. Business owners want to know whether they’re buying into experimental tech or proven infrastructure. Fair question.

So here’s the full stack. No vague “proprietary AI” claims. No black boxes. Every tool we use, why we chose it, and how it fits together.

The Architecture: WAT Framework

We built SoftSync-AI on what we call the WAT framework — Workflows, Agents, Tools. The name is deliberately unglamorous because the architecture is deliberately boring.

Workflows are the instructions. Written in plain English, they define what should happen when a lead comes in, when a job completes, when a review request should fire.

Agents are the decision-makers. This is where Claude AI (Anthropic) operates. The agent reads the workflow, decides what to do, handles edge cases, and escalates when something doesn’t fit the script.

Tools are the execution layer. Python scripts that make API calls, send messages, update calendars, trigger follow-ups. Deterministic code that does exactly the same thing every time.

Why separate them? Because AI is brilliant at reasoning but inconsistent at execution. If you ask an AI to reason through a problem AND execute every step, accuracy drops with each step. At 90% accuracy per step, you’re down to 59% after five steps.

By keeping AI in the reasoning layer and code in the execution layer, we get the best of both: intelligent decisions with reliable execution.

Claude AI: Why Anthropic

45% of businesses now default to Anthropic’s Claude for data analysis, customer service automation, and predictive modelling, according to the Ramp AI Index for March 2026.

We use two Claude models:

Claude Haiku handles high-volume, repetitive tasks. Scoring incoming leads. Categorising enquiry types. Generating quick responses. It’s fast, cheap (fractions of a penny per call), and accurate enough for structured decisions.

Claude Sonnet handles complex reasoning. Writing personalised follow-up sequences. Analysing sentiment in customer responses. Generating review responses that reference specific details from the customer’s feedback.

The key insight from Anthropic’s Economic Index: 77% of API prompts involve users requesting Claude to perform tasks, not just provide advice. That maps exactly to our use case.

n8n: The Orchestration Engine

n8n is the workflow automation platform we use to connect everything. It’s the nervous system — every trigger, every action, every conditional branch runs through n8n.

Some context on why we trust it: n8n raised $180M at a $2.5B valuation in October 2025, led by Accel with participation from Nvidia’s NVentures. It has 230,000+ active users globally and 3,000+ enterprise customers including Vodafone, Delivery Hero, and Microsoft.

What n8n does for us:

  • Routes incoming leads from any channel to the right response workflow
  • Triggers follow-up sequences on a schedule
  • Connects to Google Calendar for real-time availability checks
  • Fires review requests after job completion
  • Logs every interaction to a central dashboard
  • Handles error recovery — if a step fails, it retries or escalates

We run approximately 50 active workflows across our client base.

Twilio: Communications Infrastructure

Every phone call, SMS, and WhatsApp message flows through Twilio. It processes billions of communications globally.

For a UK trades business, this means:

  • A dedicated phone number that AI answers 24/7
  • SMS responses that arrive in under 5 seconds
  • WhatsApp Business API integration
  • Call recordings for quality and compliance

Twilio is invisible to the end customer. They call your number. AI answers. They have a natural conversation. They don’t know — and don’t need to know — that it’s not a human receptionist.

Google Calendar: Scheduling

We connect directly to Google Calendar for real-time availability. When a lead calls and needs a plumber on Thursday afternoon, the AI checks the actual calendar, finds an open slot, and books it.

No double bookings. No back-and-forth. No “let me check and get back to you.”

This single integration eliminates 80% of the evening admin that trades business owners complain about.

How a Real Interaction Flows

Here’s what happens when someone calls a SoftSync-AI client at 9:47pm on a Tuesday:

0s: Phone rings. Twilio routes to AI.
2s: AI answers: “Hi, thanks for calling [Business Name]. How can I help?”
15s: Caller: “My boiler’s making a banging noise and the pressure keeps dropping.”
17s: Claude processes: identifies job type (boiler repair), assesses urgency, prepares qualification questions.
20s: AI: “I can help with that. A few quick questions — what make and model is your boiler, and what’s your postcode?”
45s: Caller provides details.
47s: Claude checks postcode (within service area), checks calendar (Thursday 2-4pm available).
50s: AI: “I’ve got a slot on Thursday between 2 and 4pm. Shall I book that in for you?”
55s: Caller confirms.
57s: n8n books the calendar slot, sends confirmation SMS, adds to CRM.
60s: AI: “You’re all booked in. You’ll get a text confirmation now with all the details.”

Total time: 60 seconds. The tradesperson was watching TV. They’ll see the booking in their calendar tomorrow morning.

What We Don’t Do

Transparency matters, so here’s what AI doesn’t handle:

  • Pricing complex jobs. AI can give ballpark ranges but doesn’t quote custom work.
  • Emergency triage requiring safety judgement. If someone reports a gas leak, AI provides the Gas Emergency number (0800 111 999) and escalates immediately.
  • Relationship management. AI handles the transactional layer. The human relationship stays human.

The goal isn’t to replace the tradesperson’s judgement. It’s to make sure every opportunity reaches them in the first place.

For the Builders Reading This

If you’re technically inclined, here’s how it breaks down:

  • Claude API with explicit max_tokens, system prompts, and JSON schema output
  • n8n Cloud for workflow execution
  • Twilio for telephony and messaging
  • Google Calendar API for scheduling
  • Google Sheets / Airtable / HubSpot for CRM (whatever the client already uses)
  • Python for custom tooling and data processing
  • Cloudflare Pages for web hosting

Every component is replaceable. Every integration is standard API. There’s no vendor lock-in.


SoftSync-AI is built on Claude (Anthropic), n8n, Twilio, and Google Calendar. Every tool chosen for reliability, not novelty. Powering 24/7 automation for UK service businesses.

Ready to automate?

Book a free 15-minute demo and see exactly how SoftSync-AI works for your business.

Book Your Free Demo

Alex Chirila

Founder of SoftSync-AI · Building AI automation for UK service businesses