Paper to pipeline: 15 years of building the boring systems

Every company I have worked in, I ended up doing the same job. Find the boring, repetitive process that slows everyone down, and build a system so nobody has to do it by hand again.

I studied Computer and Digital Systems Technology at Vanier College in Montréal. I was headed to university after that and dropped out, because I was already running a marketing company and wanted to keep running it. Once I was working in business, the technical side turned out to be the useful part. It let me systemize the repetitive work that everyone else did by hand.

Design&Rank: from paper to one pipeline

Design&Rank is a web design, SEO and reputation agency I co-founded in Montréal. We grew it to 40 people and served clients across North America. At its peak it was one of the bigger web design shops in the city.

It started on paper. I mean that literally. Leads were written on paper. Closers called from the paper. Customer service logged each job on paper so the dev team could fulfil it.

The off-the-shelf systems I looked at were not modular enough. We had SDRs, closers, customer service, retention officers, a development team and a networking team, and every client was on a monthly plan. Everyone needed the same client data, and billing had to be tied in.

So the first job was to find a base I could modify enough. The second was to rebuild it around the whole organisation, one stage at a time, with my dev team. By the end, every part of the company ran through it, product fulfilment included.

The result I remember best was not a number. New hires kept telling us it was the easiest system they had ever learned.

Prime CS: minutes are margin

Prime CS sets appointments for car dealerships. It is paid per lead. Dealerships send the leads, and every extra minute we spend on a lead is margin we lose.

When it started, there was no system. Agents dialed by hand from spreadsheets.

I was part of operations, so the feedback loop was right in front of me. The first fix was dialing. Calls go through voicemail detection before an agent picks up. When a call connects, the agent is already talking to a live person, with the right information on screen.

The second fix was formats. Every dealership sent leads in its own format and wanted results back its own way. Some wanted email. Some wanted a file in a specific layout on an SFTP server. The system handles each one.

The third was management. I built dashboards so supervisors and owners could run each campaign to its goal.

Prime CS has been running for about ten years and is still a major player in that industry. It has strong people running the floor, and the system makes their management work easier.

The part that changed

In both companies I designed the system and then hired people to build it: in-house developers, and overseas teams. That worked. It was also slow. Every change went through a spec, a handoff, a build and a review before I could see whether it helped.

The only thing AI changed for me is that loop. I can now build the systems myself and see the result the same day. The business judgment is the same. The time between an idea and a live system is much shorter.

The first solo builds

The first place I built this way was a micro-lending company. No dev team. Four systems:

• A marketing feedback system: reporting across Google Ads and Meta, feeding creative and ad strategy, with an LLM compliance layer over all marketing output.
• A conversion-optimized intake survey, wired into the lending platform's API so agents could act on a client's data right away.
• Augmented underwriting. We kept human underwriters. The AI gives them signals, and it runs on the company's own machine with local models so client data stays there. Approve and decline got faster, and the default rate went down.
• A collections agent that reaches defaulted clients by SMS and email to set up payment plans. Every outbound message gets a compliance check before it goes out.

Then Vinsight, a deal-scouting system for automotive. It pulls sales data and trends. Every new opportunity passes a predictive gate, then a deterministic gate, then an LLM that reads the full context and decides whether it is a real deal. The team stopped scouting and started closing what the pipeline fed them.

Same job, new tools

The Design&Rank pipeline, the Prime CS dialer and the lending systems solve the same problem. People were spending their day on work a system should do, and it cost the company money.

What is different now is scale. Today I run a fleet of agents that builds and operates these systems with me. A strong model plans and reviews, cheaper models do the execution, and every result gets checked before it counts. The other posts in this log cover how that works.