Work

What was stuck. What we built. What happened.

Five builds, from a weekly outreach system to a marketplace audit. Every number is from the work itself, and every case says plainly whether it is live, delivered or built and ready.

CASE 01 · Commercial landscaping and property maintenance

The most valuable thing the company owned was locked in an inbox.

Grounds maintenance and snow removal · property-management and condominium accounts

Live. Running every week.

The problem

This company wins work from property managers who already know them. Years of those relationships sat in two email accounts: quotes sent, jobs won and lost, lunches, people who went quiet and were never chased.

The customer list was a 33-row spreadsheet. Nobody could say which relationships were still warm, who had asked for a price and never got a follow-up, or who to call first on Monday.

The result

Old relationships reopened and turned into booked meetings
Quote requests coming back in from accounts that had gone silent
The customer list rebuilt from 33 contacts to over 120, showing which properties each manager controls, last contact, what was quoted and what happened
A weekly outreach campaign that runs without the owner doing any admin

Read the full case study: how the system works, its guardrails and its results →

~3,000email conversations over three and a half years

945clear facts, each traceable to the email it came from

83ranked opportunities in three priority tiers

How we did it

It read the archive

Every quote, reply, meeting and outcome extracted. One rule made the difference: a quote with no reply is an open opportunity, not a lost deal.

It scores every account out of 100

Relationship quality, quote recency, price competitiveness, signs of unhappiness with their current provider, work controlled, and reachability.

It can’t contact the wrong people

Competitor-linked firms, their staff and properties, and anyone who asked not to be contacted are blocked before scoring. No score can override that.

It writes like the owner

Modelled on his own past emails. Every invitation names the properties that manager controls and offers real dates.

He approves every message

The system can’t send on its own. He approves, edits or holds each one, and every sent message is checked against what he approved.

CASE 02 · Growth playbooks

Not frameworks. The actual words, sequences and numbers.

Applicable to any business that needs customers

Built and in use

The problem

A business that wants to grow has to answer a hundred practical questions, and generic advice answers none of them. What do I say on a cold call? What goes in the first email, word for word? How many follow-ups, how many days apart? How do I raise prices without losing customers? What do I say when someone tells me I’m too expensive?

Most owners either guess, or pay a consultant whose advice is limited to what that one person has personally done.

The result

Getting customers: cold calling, cold email, text, LinkedIn, social groups, door-to-door, direct mail, video outreach
Selling: scripts, objections by type, follow-up sequences, stopping no-shows
Pricing and offers: what to charge, guarantees, raising prices, unit economics
Marketing, partnerships, delivery and strategy: from getting found by AI assistants to when to hire and for what

Applied to one company, they become its growth plan: which tactics fit, which to ignore, in what order, and what to say.

1,745source documents, around 12.4 million words

1,844structured records, each tagged by subject

26operational playbooks, 263,809 words

How we did it

Checked before scaling

A test run over 80 documents recovered five times more useful detail than our own threshold required. Only then did the full run go ahead, recovering over 20,000 additional pieces of detail a normal summary would have thrown away.

Fifteen independent reviewers

They audited the finished playbooks against the raw source. One playbook had dropped material. We repaired it by adding, never deleting, and restored every missing citation.

Fixed rules when applied

Every recommendation cites its source. Where evidence is thin or sources disagree, we say so. Deceptive tactics such as fake referrals and false scarcity are excluded, with legitimate alternatives supplied.

CASE 03 · Customer outreach engine

Lead with something worth reading, not a pitch.

Prospecting system · any sector

Built and tested. Ready to deploy.

The problem

Most owners know roughly who their customer is but have no time to find them, research them and contact them consistently.

And most cold outreach is ignored, because it’s about the seller. “Here’s what we do, can we book a call?” gives the reader no reason to reply.

The result

It builds the target list: the right businesses, with phone numbers, websites, and whether each number is mobile or landline, so it knows whether to call or text
It researches each prospect before writing anything
It leads with something useful: a short audit of their website, an estimate of what a problem is costing them, a map of where they lose time
It writes in your voice, and you approve before anything sends
It runs follow-ups on schedule and reports what worked

The prospect gets something worth reading about their own business. That’s why they reply.

How we did it

The risky part first

Can software act on live accounts safely without someone watching every step? We tested that on real accounts across two platforms before building anything on top.

It can’t send the same thing twice

Each message is keyed to the recipient and the step in the sequence, so even a reworded draft is recognised as already sent and stopped. Proven in testing.

It checks what actually happened

After every action it looks at the screen and confirms the change, rather than trusting a success message. Every run leaves a full record.

CASE 04 · Local-pickup marketplace

The answer to the question he hadn’t asked.

Marketplace concept · independent retail

Complete. Build specification delivered.

The problem

A founder had a marketplace idea and was about to spend real money building it. Shops would list stock, buyers would reserve an item and collect it in person, with a small holding charge to stop no-shows.

He had no way to know whether it could reach the size he was assuming, what it would cost to run, which regulations would break it, or what had to be true for it to work.

The result

A size analysis that contradicted every document he had. Counting the actual qualifying shops showed his revenue targets needed essentially every shop of that type nationally, and one target wasn’t reachable on that model at all
A thirty-day test that settles it, with pass, extend and stop numbers agreed in advance, for a fraction of the cost of building
An audit that disproved two things in his own files: terms recorded as agreed that had never been agreed, and a claim that the product was nearly built
A clickable prototype of eleven screens, including the reservation flow and reputation system

He learned what would kill it before spending on development. That’s what an audit is for.

How we did it

Full technical design

Platform choices with reasons, the data structure, integrations, and every third-party service costed at five different sizes of business.

Payments, rule by rule

Every released hold must be explicitly cancelled, because card networks penalise holds left hanging: invisible while you’re small, expensive by the time you’re not. Reservations expire within five days, not seven.

Regulation as product requirements

One province ruled out because its consumer law forbids advance deposits. Fee disclosure designed around a national enforcement case over hidden charges, so the hold appears as prominently as the price.

CASE 05 · Websites for service businesses

Sites that get found, qualify the lead, and cost nothing to keep running.

Commercial trades and property services

Live

The problem

Service businesses get sold websites that look acceptable and do nothing. No structured data, so search engines can’t properly read what the business does or where. Service pages named after the internal menu instead of what customers type. A contact form that tells the owner nothing. And a monthly subscription for a system the owner never opens.

The result

Readable by search engines and AI assistants: every page carries structured data for the business, its services, its service area and its contact details
Service pages named for what customers search
A qualification form that captures property type, service, site size and timeline before the owner picks up the phone
No subscription, no plugins, nothing to update, and no content system to be hacked

How we did it

Hand-built

No framework and no content management system, which removes the maintenance burden entirely.

Built properly

Structured data typed per page, search and social previews, multi-locality service areas, four responsive breakpoints, a mobile call-and-quote bar, and real accessibility.

Handed over properly

A deployment and operating manual, a go-live checklist with a live form test, dated backups, and a written list of what’s deliberately not included and why.

Your turn

If one of these looks like your problem, let’s talk about it.

The first conversation is short. You’ll leave knowing whether AI is the right tool, and what it would take.

Book a call See services

What you’re trying to fix
Whether AI is genuinely the right tool for it
What it would take