Finance operations
Forward-deployed engineering
Training
Glox Group: one finance process across four markets
Client: Glox Group, a trading group of around 20 staff across China, Hong Kong, the UK and the UAE. When: September 2025 to March 2026, with ongoing support.
The problem
One small finance team served every entity, and each market had its own VAT, audit and bookkeeping rules. Invoices were keyed in by hand, supplier spend sat in inboxes, expense claims arrived as phone photos and direct debits were chased manually.
What we did
We started with on-site whiteboard sessions with the finance team to map every workflow and handoff and record a baseline. The roadmap ranked each opportunity by payback. Then we built it entity by entity: invoice capture, coding and matching into Xero and QuickBooks; VAT and audit rules applied per market; expense intake with approval routing; direct debit collection and reconciliation. The team was trained on each workflow as it went live, and failure alerts tell us when something breaks.
Results (figures supplied by Glox Group)
- 50% less manual finance admin each month
- 90% fewer coding and matching errors
- Four markets on one process
Marketing operations
AI build
Live handover
Elematch: a full marketing function, run by one founder in about three hours a week
Client: Open the Door Company, which runs Elematch, a team-matching SaaS product. A one-person business with no marketing hire. When: built May to July 2026; running since 12 July 2026.
The problem
A solo founder still needs social content, design, publishing, SEO, paid ads and market research. Every time marketing needed doing, product work stopped.
What we did
We built an AI-assisted marketing system across content, design, publishing, SEO, paid ads and market research, then handed it over live so the founder runs the whole function in about three hours a week — nearly all of it spent approving and steering rather than making.
Result
- A complete marketing function kept running on roughly three owner-hours a week
- Product work no longer stops whenever marketing needs doing
Forward-deployed engineering
AI implementation
Automation care
Training
Governance
A Hong Kong university's Executive MBA: training, engineering, automation, care and governance in one engagement
Client: the Executive MBA programme of a Hong Kong university. Scope: the AI, data and automation layer behind a 12-month recruitment campaign, working alongside the programme's media agency.
The brief
Fill a small, senior intake: people with seven-plus years' experience, a degree and strong English, across Hong Kong and mainland China. Every lead had to reach the admissions team quickly, scored, and with nothing lost on the way. Volume can mislead here. One right candidate is worth more than a hundred clicks.
Forward-deployed engineering
Before any ad went live we wrote the conversion framework: the events that count at each stage, from first click to a booked Coffee Chat. We then linked the lead pipeline to the programme's Salesforce through its API, tested in the sandbox first and upserting on an agreed ID with approved field mapping. A secure CSV route ran from day one alongside it, so lead delivery never depended on one integration.
AI implementation and automation
Every lead from Meta, LinkedIn and Google was queued, cleaned, de-duplicated and scored. The score is explainable: 100 points from seniority (40), experience (35) and organisation (25), with small bounded adjustments. AI reads messy job titles and company names into that model; the rules themselves are fixed and were approved by admissions. Anything unclear goes to a person. New and existing CRM records were tagged, classified and enriched. Confirmations and reminders for Coffee Chats and Info Sessions went out on WhatsApp through the official Meta Business API, on the university's own account, using four approved templates. Each send checked consent, opt-outs, quiet hours and frequency first.
Automation care
A daily self-checking report reconciles lead counts between every source and destination, so a failed transfer shows up the same morning. It runs whether or not Salesforce is connected. A monthly report covers performance by channel and lead volume by tier. AI drafts the narrative from fixed figures only, and a person signs it off. Maintenance and support cover the full term.
AI training
The admissions team got a user manual for the scoring and reporting automations and two 60-minute online training sessions. The marketing team had a workshop on using AI tools for ad layouts and content, covering prompt writing and tool selection, with written materials to keep.
AI governance
The score prioritises follow-up only. Admissions keeps every eligibility and admissions decision, and no model rejects a candidate. AI can't invent statistics, fees or programme facts. No personal data goes into public consumer AI tools. A tool register records each vendor, its purpose, the data it may touch and its retention. Every AI-assisted output records the tool, prompt version and reviewer.
Results
- 100% of AI output approved by a named person; no candidate decision made by a model
- Full campaign results to follow after the first admissions cycle.
