The AI Ads Supervisor: Autonomous Google Ads Monitoring & Optimisation
How we built an AI system that monitors Google Ads and GA4, delivers daily insights, and pushes optimisations directly to landing pages and ad campaigns — without human intervention.
Read case study →
Generative AI · Real Estate
AI-Powered Property Listings: Text, Image & Video in One Platform
How we built a full-stack AI solution that turns property addresses into professional listings — complete with AI copy, enhanced photos, and marketing videos.
Read case study →
AI Bot to Go-To-Market Engine
Dulk's Expert Hub: From AI Bot to GTM Engine
How a RAG-powered Expert Hub turned real user questions into a sharper brand and GTM strategy.
Read case study →
Technical Skills & Services
The building blocks we use to create custom Automation & AI solutions for your business.
Agentic AI
AI systems that don't just recommend — they act. Autonomous agents that monitor data, make decisions, and push changes to production on a schedule.
AI AgentsClaude CodeMCPAutonomous Ops
LLM & AI Integration
Production-ready integrations with leading AI models, including prompt engineering, RAG systems, and structured output pipelines.
OpenAIAnthropicRAGEmbeddingsGoogle Gemini
Generative AI
Multi-modal content generation: AI-powered text, image enhancement, and video creation pipelines.
GPT-4oClaude SonnetImage AIVideo AIVeo
Workflow Automation
End-to-end process automation connecting your existing tools into seamless, reliable workflows.
n8nMakeWebhooksCron JobsTelegram Bots
Full-Stack Development
Custom web applications, APIs, and backend systems built for performance and scalability.
PythonFastAPINext.jsPostgreSQLStripe
Data & Analytics
Data pipelines, automated reporting, and dashboards that turn raw data into actionable insights.
Google Ads APIGA4Xero APIDashboardsForecasting
Cloud & DevOps
Secure deployment, containerisation, and infrastructure management for reliable, scalable systems.
DockerLinuxCI/CDGitHub ActionsMinIO/S3
See What We've Built
From AI-powered property listings to automated cash control systems, explore real-world examples of how we've helped businesses streamline operations and unlock new capabilities with Automation & AI.
Real Estate AI
Finance Automation
Inventory Intelligence
RAG Expert Hubs
Agentic Ad Ops
From Idea to Insight, Fast.
1. Discover
We start with your goals. We'll identify the high-impact areas in your business where Automation & AI can deliver a clear, immediate ROI.
2. Build
No theoretical projects. We build and deploy a secure, custom Automation or AI tool, integrating it directly into your workflow.
3. Listen
This is the magic step. We integrate analytics and feedback loops to capture what your customers *actually* want and where your operations *actually* break.
4. Align
We help you use this new, real-time data to make smarter, faster business decisions—aligning your marketing, product, and strategy to real demand.
Blog, Insights & Case Studies
Deep dives into real projects and practical Automation & AI applications, from early experiments to full go-to-market engines.
Generative AI · Real Estate
AI-Powered Property Listings: Text, Image & Video in One Platform
How we built a full-stack AI solution that turns property addresses into professional listings—complete with AI copy, enhanced photos, and marketing videos.
Tags: AI Copy · Image Enhancement · Video Generation · Real Estate
AI Bot to Go-To-Market Engine
Dulk's Expert Hub: From AI Bot to GTM Engine
How a RAG-powered Expert Hub turned real user questions into a sharper brand and GTM strategy.
Tags: RAG · Brand Trust · GTM Insight Engine
Finance Automation
Cash Control Digest with AI: Stay on Top of Cash in 60 Seconds a Day
Automated Daily Cash Control Reports from Xero: instant visibility, zero manual work, better decisions.
Tags: Xero · Cash Flow · Finance Automation
Inventory Automation
AI-Powered Inventory Control: Stop Bleeding
How Loopsu Inventory Intelligence transforms chaotic Excel spreadsheets into real-time stock visibility, AI forecasting, and automatic replenishment.
Tags: Inventory · Stock Management · AI Forecasting
Agentic AI · Ad Ops Automation
The AI Ads Supervisor: Autonomous Google Ads Monitoring & Optimisation
How we built an AI system that monitors Google Ads and GA4 around the clock, delivers daily insights, and pushes optimisations directly to landing pages and ad campaigns — without human intervention.
Tags: Google Ads · GA4 · Agentic AI · Autonomous Optimisation
Generative AI · Text Generation · Image Enhancement · Video Creation
Building an AI-Powered Property Listing Platform: Where Text, Image, and Video Generation Converge
Real estate agents spend 3+ hours creating each property listing—writing descriptions, editing photos,
and coordinating marketing materials. We built a platform that uses three distinct AI capabilities
to compress this into minutes. Here's how the technology came together.
The Hidden Time Tax in Property Marketing
Every property listing requires a surprisingly similar workflow: gather details, write compelling copy,
enhance photos to look professional, and increasingly, create video content for social media.
For busy agents juggling multiple listings, this repetitive work becomes a significant bottleneck.
3+ hrs
Per listing creation
$500+
Cost if outsourced
80%
Time on non-selling tasks
The question wasn't whether AI could help—it was whether we could integrate
text, image, and video generation into a single, coherent workflow
that non-technical users could actually use. The result is
Realestate.ai,
a platform built specifically for Australian real estate agents.
Three AI Engines, One Workflow
The platform orchestrates three distinct AI capabilities, each chosen for its specific strength:
Text Generation (OpenAI GPT-4o)
Carefully crafted prompts incorporate property context, selling vibe, and agent tone
to generate headlines, descriptions, and social captions.
Image Enhancement (Google Nano Banana AI)
Room-specific enhancement that corrects lighting and colour—without fabricating
features that don't exist. Ethics built in.
Video Generation (Google Veo AI)
Two-stage process: generate a lifestyle scene from property photos, then animate
into social-ready video clips.
The Platform in Action
Step 1: AI Property Lookup — auto-fills property details from Domain & REA.
Step 2: AI Copy — headlines, descriptions, and social captions.
Step 3: Image Enhancement — professional lighting correction.
Step 4: AI Video — lifestyle videos for social media.
Technical Architecture Note
The platform runs on a FastAPI backend with n8n workflow automation orchestrating the AI calls.
This separation proved essential: long-running AI tasks (especially video generation) can take minutes,
so we use webhook callbacks rather than blocking API calls.
▸Backend: FastAPI + PostgreSQL
▸Orchestration: n8n workflows
▸Storage: MinIO (S3-compatible)
▸Payments: Stripe subscriptions
What We Learned Building This
Prompt engineering is product design
The difference between "good enough" and "actually useful" AI output came down to prompt
refinement. We iterated through dozens of prompt versions, testing with real property data,
before the copy generation felt right for Australian real estate conventions.
Async is non-negotiable for generative AI
Video generation can take 60+ seconds. Image enhancement, 10-30 seconds. Building a responsive
UI meant embracing webhook callbacks and background job processing from day one—not retrofitting it.
Ethical constraints improve the product
Our "no fabrication" rule for image enhancement initially felt like a limitation. In practice,
it became a selling point. Agents don't want to mislead buyers—they want photos that look
professional while remaining honest.
The Outcome: From Hours to Minutes
Before
✗3+ hours researching, writing, editing per listing
✗$50-150 for professional photo editing
✗$300-500+ for marketing video production
✗Days of back-and-forth with vendors
After
✓10-15 minutes from address to full package
✓Unlimited image enhancements included
✓AI videos ready for social platforms
✓Same-day turnaround, every time
The Bigger Picture: Multi-Modal AI in Practice
This project demonstrated something we're seeing across industries: the real power of generative AI
isn't in any single capability—it's in combining text, image, and video
generation into workflows that previously required multiple specialists, tools, and handoffs.
Real estate was a good proving ground because the workflow is well-defined and the pain point is
universal. But the pattern—orchestrating multiple AI engines behind a simple interface—applies
far beyond property listings. See it in action at
realestatekit.ai.
Work with Loopsu
Exploring something similar for your business?
If you're thinking about combining multiple AI capabilities into a cohesive product or workflow,
we'd be happy to share what we learned and explore whether our approach could work for your context.
Dulk's Expert Hub: How a 6-week panel turned an AI bot into a Go-to-Market strategy engine
Loopsu designed a RAG-powered “Expert Hub” for Dulk, a new food venture, connecting trusted nutrition sources with real user conversations, and used that signal to reshape the brand’s GTM.
From Books and Podcasts to Breakthrough
Loopsu's RAG system, fed with trusted expert sources (books and podcasts), didn't just build brand confidence—it uncovered the real performance needs of Dulk's audience and shaped a sharper, more aligned Go-to-Market plan.
The "Revealed" Insight
The panel's top questions weren't about having a perfectly
"balanced meal-replacement", as initially framed.
They centred on "sustained energy" and
"showing up at their best every day", without compromising on food quality.
The Brand GTM Shift
Dulk shifted its core message from "a nutritionally balanced meal-replacement" to
"performance nutrition that powers your day, so you can give your best with zero compromise on food."
This immediately resonated with test users.
The Product GTM Shift
The performance and energy angle was elevated from a supporting benefit to the
core launch promise, prioritising steady energy, focus, and convenience—so customers can perform at their best every day while still eating food they trust.
Strategic Takeaways
The "Revealed" Insight
Panel's top questions weren't about having a perfectly
"balanced meal-replacement" on paper.
They focused on "sustained energy" and
"performing at their best every day" without compromising on food quality.
The Brand GTM Shift
Core message evolved from "a nutritionally balanced meal-replacement" to
"Performance nutrition that powers your day, so you can give your best with zero compromise on food."
The Product GTM Shift
The performance and energy dimension was elevated from a supporting benefit to the
core launch promise, prioritising steady energy, focus, and convenience so customers can perform at their best every day.
Work with Loopsu
Exploring something similar for your business?
If you’re thinking about using automation or AI in a similar way, we’re always happy to compare notes
and see whether our approach could fit your context.
Cash Control Digest with AI: How One Team Got a Daily View of Cash in 60 Seconds
In this project, we worked with a growing services business using Xero that wanted a calm,
reliable way to understand cash every morning without logging into multiple systems or
running custom reports.
The Starting Point: Cash Questions Arriving Too Late
Before we started, the leadership team only discovered cash issues when they were already
painful. The patterns were familiar:
•Bank balances felt “off”, but no one had an easy way to see why.
•Large bills or payroll were looming without a shared view of upcoming commitments.
•Overdue customer invoices were picked up late, often when the cash squeeze was already there.
•Answering “Can we afford this right now?” meant someone disappearing into Xero and spreadsheets for an hour.
The team didn’t need a new accounting system. They needed a simple, shared signal that
made cash health visible every day.
Our Approach: A Daily Cash Control Digest from Xero
Together, we designed a small automation around Xero: an
automated Daily Cash Control Digest sent to the team each morning.
The goal was not to replace their accountant, but to give operators a quick, shared view
they could trust.
In one page, the digest brings together:
Today's Cash Position
Consolidated view across the main operating accounts.
Recent Revenue & Spend
Last 7 and 30 days, so trends are visible at a glance.
Upcoming Bills & Payroll
Commitments over the next 7 and 30 days, straight from Xero.
Expected Customer Payments
Invoices due in the next 7 and 30 days, highlighting key payers.
Layer of Interpretation
On top of the raw numbers, we added a simple AI layer that flags patterns such as
“watch next Wednesday, large bills + low expected cash in” or “these overdue invoices
are now critical”.
The digest is delivered to email and Slack at the same time every weekday, so it naturally
becomes part of the team’s rhythm.
How It Works Behind the Scenes
1
Connect to Xero
We securely connect to Xero and agree on which accounts and data to use.
2
Nightly Processing
Each night, an automation pulls balances, invoices, bills and recent
transactions, then computes the key indicators.
3
AI Highlighting
An AI layer adds commentary on risk (tight weeks) and attention points
(late payers, unusually high spend, etc.).
4
Delivery to the Team
The digest is formatted into a single visual report and sent to the agreed
channels (email, Slack, Teams).
What the Team Sees
Example Daily Cash Control Digest – all key signals in one glance.
What Changed for the Client
Shared Visibility
Everyone now starts the day with the same picture of cash and upcoming movements,
instead of relying on individual spreadsheets or gut feel.
Cleaner Decisions
Hiring, marketing spend and supplier negotiations are now taken with a much
clearer sense of “where we stand this week and next”.
Fewer Surprises
Cash squeezes are spotted weeks earlier, giving time to act rather than react.
Minimal Extra Work
Once deployed, the system runs from existing Xero data. For the client, it’s just
another email or Slack message in the morning – but a very useful one.
Next Experiment: A Conversational Financial Assistant
Building on this daily digest, we’re now exploring a
financial assistant chat interface that can answer questions
like “What if we bring forward this hire?” or “Which overdue invoices matter most this week?”
using the same underlying data.
The aim is simple: keep humans in charge of decisions, but make it much easier for them to
get a clear, timely view of the numbers.
Work with Loopsu
Exploring something similar for your business?
If you’re thinking about using automation or AI in a similar way, we’re always happy to compare notes
and see whether our approach could fit your context.
Inventory Automation · AI Forecasting · Stock Management
AI-Powered Inventory That Thinks Ahead: From Excel Chaos to a Single, Shared View
This case study comes from a multi-location retailer who had outgrown spreadsheet-based
inventory and wanted a more reliable way to decide what to order, when, and for which site.
The Hidden Cost of Spreadsheet Inventory
The client was running most of their stock control in Excel, with exports from POS and
invoicing systems stitched together by hand. It worked – until it didn’t. The pain points:
•Frequent stockouts on top sellers that directly impacted revenue and customer trust.
•Overstock on slow movers, tying up tens of thousands of dollars in inventory.
•Manual, error-prone updates with no single source of truth across locations.
•No practical forecasting: reorders were often based on intuition rather than data.
•Discrepancies between physical counts and system numbers that only surfaced during big audits.
The team wasn’t looking for a large ERP replacement. They wanted a lighter-weight “brain”
that could sit across the tools they already used and help them think ahead.
Our Approach: An Inventory “Intelligence Layer” on Top of Existing Systems
We built an automation + AI layer that connects Excel, POS, ecommerce platforms and
invoicing systems into a single inventory view. The focus was less on replacing tools and
more on joining them up.
From there, we added a set of capabilities designed around the way their team actually
works in week-to-week operations.
What the System Now Does for Them
1
Real-Time Stock Sync & Alerts
Stock levels from POS, ecommerce and warehouse spreadsheets are consolidated.
Low-stock alerts are raised before shelves are empty.
2
AI-Powered Forecasting
Forecasts combine sales velocity, seasonality and upcoming events to
suggest “order-by” quantities per SKU and per site.
3
Draft Purchase Orders
Based on agreed rules, the system prepares draft POs for suppliers. The
team keeps control by reviewing and approving before sending.
4
Discrepancy Detection
Differences between system records and physical counts are flagged quickly,
prompting targeted checks instead of full-scale audits.
5
Digital Stock-Taking Support
During audits, staff use a simple interface on phones or tablets. Variances
are summarised automatically for follow-up.
6
Monthly Inventory Review
A monthly “inventory story” report highlights slow movers, overstock risks
and where cash is tied up, along with suggested actions.
The Inventory Command Center
One place where the team now checks stock levels, alerts and suggestions.
Why This Changed Day-to-Day Work
Better Use of Cash
With overstock visible SKU-by-SKU, the team could plan promotions and buying
pauses, progressively unlocking cash that had been sitting on shelves.
Fewer “We’re Out” Moments
Stockout alerts and forecasting meant bestsellers were rarely unavailable, which
helped protect both revenue and customer experience.
Less Time Wrestling Spreadsheets
Manual consolidation time dropped significantly. Weekly conversations shifted from
“What’s the correct number?” to “Given these numbers, what should we do?”.
Pilot Example: Coffee Roastery
Before
✗Regular stockouts of a flagship espresso blend.
✗Substantial cash locked in niche single-origin SKUs.
✗Many hours each week spent reconciling POS exports with Excel.
✗Stock discrepancies only uncovered during quarterly stocktakes.
✗Seasonal demand (e.g. cold brew) often underestimated.
✓Slow-moving lines identified and progressively cleared, freeing up cash.
✓POs generated automatically, with staff mostly reviewing and approving.
✓Variance alerts pointed to potential theft or recording errors much earlier.
✓Seasonal demand spikes were anticipated using historical patterns.
Exact numbers vary by client, but across pilots we consistently see a mix of freed-up
cash, fewer stockouts, and meaningful time savings for the operations team.
How We Structured This Kind of Project
Rather than a big-bang implementation, we typically structure inventory intelligence work
in layers, starting with visibility and only then adding automation.
• Draft purchase orders and supplier notifications.
• Variance monitoring between counts and records.
• Monthly “inventory story” and continuous refinement.
Scope, timelines and depth depend on each client’s starting point. We usually begin with a narrow pilot to validate impact before scaling.
If your team is wrestling with similar inventory questions and you’d like to see how an
“intelligence layer” could sit on top of the tools you already have, we’re happy to talk
through what we did in this and other projects.
Work with Loopsu
Exploring something similar for your business?
If you’re thinking about using automation or AI in a similar way, we’re always happy to compare notes
and see whether our approach could fit your context.
Agentic AI · Google Ads · GA4 Analytics · Autonomous Optimisation
The AI Ads Supervisor: How We Built a System That Monitors, Analyses, and Optimises Google Ads — While You Sleep
Running paid ads for a SaaS product means checking dashboards twice a day, cross-referencing
data across Google Ads and GA4, and turning numbers into decisions. We replaced that entire
loop with an agentic AI system: an automated pipeline that pulls live campaign data, generates
structured performance analysis, stores everything for historical tracking, and delivers
prioritised recommendations to your phone every morning and evening. But the real game-changer?
The AI agents don't just recommend changes — they implement them
directly on landing pages and ad campaigns, pushing optimisations to production
without waiting for a human to act.
The Problem: Ad Optimisation Is a Daily Grind That Never Ends
When you're running Google Ads for a SaaS product, the feedback loop between spend and
outcome isn't always obvious. You're juggling multiple data sources — Google Ads for cost
and click data, GA4 for on-site behaviour — and trying to connect them into a coherent
picture, twice a day. Then you need to act on what you find: update ad copy,
add negative keywords, tweak landing page sections, adjust bids. By the time you've done
the analysis, there's barely time left for the implementation.
2×
Daily reviews needed
6+
Data streams to cross-reference
0
Connection between ad spend & on-site behaviour
The real question wasn't whether to automate the data pull — it was whether we could build
a system that thinks and acts like a performance marketer:
one that connects ad spend to landing page behaviour, identifies waste, spots opportunities,
delivers specific actions — and then executes those changes itself.
The Autonomous Loop
The Supervisor runs a continuous collect → analyse → act → deliver cycle,
triggered automatically at 8 am and 6 pm every day. No manual intervention required.
1. Collect — Google Ads + GA4
Pulls campaign metrics (impressions, clicks, CTR, CPC, conversions, cost per conversion),
search terms that triggered ads, and keyword quality scores from Google Ads.
From GA4: paid traffic sessions, bounce rate, session duration, landing page performance,
and custom homepage engagement signals — scroll depth, CTA clicks,
time-on-page milestones, and exit intent.
2. Analyse — AI Performance Analyst
The combined dataset is sent to an AI engine configured to act as a senior Google Ads
analyst who understands the specific business context. The output is a structured report:
performance summary, what's working, what needs attention, and 3–5 concrete actions
prioritised by expected impact — using the actual numbers from the data.
3. Act — AI Agents Push Changes to Production
This is the step that separates the Supervisor from a reporting tool.
AI agents take the analysis and implement the recommendations
directly: adding negative keywords in Google Ads, adjusting bid strategies,
reordering landing page sections, updating CTA copy, or tweaking headlines —
and pushing those changes live. No ticket, no handoff, no waiting.
4. Store & Deliver
All raw metrics and AI analysis are stored for historical trend tracking.
The analysis is pushed to Telegram for instant reading. A web dashboard provides
trend charts, search term tables, engagement funnels, and the latest AI report —
so you always have the full picture at a glance.
How It Fits Together
Automated — 8 am + 6 pm daily
Google Ads+GA4 Analytics
│
Data Pipeline
│
AI Analysis Engine
│
AI Agents — push changes live
│
TelegramWeb Dashboard
↻ fully automated
What the agents can change autonomously
▸Add/remove negative keywords
▸Adjust bid strategies
▸Update landing page copy
▸Reorder page sections
▸Tweak CTA wording & placement
▸Push changes to production
The Dashboard in Action
Every data point the Supervisor collects is stored and served through a web dashboard
that provides the trend visibility no single Telegram message can: period-over-period
summary cards with change arrows, granularity controls, sortable search term tables,
section engagement funnels, CTA click heatmaps, scroll depth and time milestone charts,
and the latest AI analysis rendered inline.
The live Ads Supervisor dashboard — all sections populated with campaign data and AI-generated analysis.
Not Just Reporting — Thinking and Acting
Most "AI-powered" ad tools summarise numbers you can already see.
The Ads Supervisor goes further because it sees both sides
of the funnel — what you're paying for in Google Ads and what
those visitors actually do on your site — and then acts on
what it finds.
Typical "AI" Ad Tools
✗"Your CTR dropped 12% this week"
✗"Consider reviewing your ad copy"
✗No visibility into what happens after the click
✗You still have to implement every change manually
The Ads Supervisor
✓"CTR dropped 12% — correlated with 68% bounce on /pricing, suggesting landing page mismatch" → reorders the pricing page and pushes live
✓"'free' keyword wasting $110/week" → adds it as a negative keyword automatically
✓Connects ad spend to scroll depth, CTA clicks, and exit intent
✓Implements changes, then measures the impact next cycle
What We Learned Building This
Context is everything for AI quality
A generic "analyse this data" instruction produces generic advice. We spent significant time giving the AI engine context about the specific business, the campaign type, and what "good" looks like. That context is the difference between useful recommendations and noise — and it's what allows the agents to make safe, informed changes autonomously.
Connecting the data is the hard part
Google Ads and GA4 are completely different systems with different data structures and quirks. The AI analysis is only as good as the data you feed it — getting clean, unified data from both platforms was the majority of the engineering effort. The AI analysis and the agentic implementation layer came together relatively quickly once the data was solid.
Autonomous doesn't mean unsupervised
The agents can push changes to production, but every action is logged, reported, and reversible. The Telegram notifications include what was changed and why, so the human stays informed. Think of it as a team member who works overnight, takes initiative, but always leaves clear notes about what they did.
The Outcome: From Reactive to Fully Autonomous
Before
✗Manual login to Google Ads + GA4 twice daily
✗No connection between ad spend and on-site engagement
✗Insights lived in someone's head, not in a system
✗Even when issues were found, changes took days to implement
After
✓Fully autonomous — analysis and optimisation twice daily
✓AI agents push changes to ads and landing pages automatically
✓Every change logged, reported, and reversible
✓Continuous improvement loop — each cycle measures the impact of the last
The Bigger Picture: From Dashboards to Autonomous Operations
This project is a blueprint for what we call "agentic supervision" —
AI systems that don't just report on your business, they run parts of it.
The pattern goes beyond monitoring: collect data, analyse it, decide what to change, implement
the change, measure the result, repeat. It applies far beyond advertising — inventory
management, financial monitoring, customer support triage, or any domain where a human
currently reviews dashboards and makes incremental decisions.
The key insight: the gap between "insight" and "action" is where
most value is lost. Plenty of tools can tell you what's wrong. The real leverage comes
from closing that gap — letting AI agents act on what they find, immediately, every day,
while you focus on the work that actually requires a human.
Work with Loopsu
Want AI agents optimising your business while you sleep?
Whether it's ad spend, inventory, cash flow, or operational KPIs — if you're currently
reviewing dashboards and making changes manually, there's a good chance we can build AI
agents that do both for you, around the clock.
I'm Eric Amiel, founder of Loopsu, an entrepreneur first, not just a consultant. I built Loopsu with a simple mission: to create the practical, effective AI tools I wished I'd had in my own businesses.
I thrive in fast-paced projects, moving quickly from idea to a functional, value-driven tool. My focus has always been on practical application—developing automations and AI systems that solve real-world operational bottlenecks and unlock new insights.
This "build-for-my-own-business" mindset means I understand your challenges. You don't need theoretical AI; you need a reliable system that delivers a clear ROI. That's what we build at Loopsu.
I'm a dynamic leader with a strong background in technology, operations, and finance.
At We Wander,
I led Operations, Tech, and Finance and remain one of the company's directors.
I'm also a director of
Dulk.
Earlier in my career, I held key roles at Thales, Airbus, and the French Space Agency.
My core expertise sits at the intersection of project management, software engineering, and entrepreneurship.
