Flexiwork: product discovery, a build-vs-buy call, and a 45-minute task cut to 30 seconds

Context

Flexiwork is a B2B platform for office space rental and acquisition. I have worked with them twice, four years apart. In 2022 and 2023 I was the front-end engineer on the platform rebuild, in a two-person team. In April 2026 they brought me back, this time as a product and technical consultant.

Being hired back by a former client is the part of this story I am proudest of, and it is the reason both missions belong in the same case study.

First mission: the decision not to write the code

Problem

The platform needed a rebuild. I owned the product specifications and the tech stack decision, chose Next.js and Tailwind, and led the full UI/UX redesign in Figma.

Then I reached a conclusion that went against my own interest: the team would not be able to maintain a custom codebase after I left. There was no one to own it. A well-built custom platform that nobody can change is worth less than a modest one the team can actually operate.

What I did about it

I decided to rebuild the platform on no-code instead. That meant scoring every feature for feasibility on the new stack, deferring the ones that could wait to a V2, and being explicit about what would be lost. Then I shipped the rebuild solo before transitioning out.

The lesson I took from it is the one I still apply: the right architecture is the one the team can live with once you are gone.

Second mission: product consulting, end to end

Problem

Four years later, the company had grown and the constraint had moved. Growth, operations, data and conversion all needed work, and there was no sequencing to say what came first or what was worth building at all.

Discovery and the PRD

I owned the discovery for a four-track product roadmap: organic acquisition, internal tooling, data infrastructure, and commercial conversion. It was formalised in a complete PRD and presented to the founders, so the plan was something they could challenge and own rather than a list of tickets.

Sequencing was the substance of the recommendation: acquisition, then operations, then data, then conversion. Each stage feeds the next, and doing them in a different order wastes the work.

Build versus buy, with numbers

For each track I made build-versus-buy decisions with benchmarked ROI. This is where a consultant earns their fee, and the honest answer is often "buy it, or do not do it yet". Deciding what not to build is worth more than building fast.

What I am shipping

I am executing across all four tracks:

• An AI-powered commercial PPT generator. Claude APIs orchestrated via n8n, generating the commercial decks the sales team used to assemble by hand.
• A technical SEO rebuild to make organic acquisition work.
• Automated multi-source offer aggregation, so listings do not have to be collected manually.
• A prospect-facing deal room with engagement tracking and HubSpot sync, so the sales team can see what a prospect actually looked at.

Technical approach

The PPT generator is the piece that best shows how I use AI in production. The interesting problem is not calling a model, it is grounding it: the deck has to contain the company's real offers, real numbers and real formatting, which means the LLM is one step inside a pipeline that collects and shapes the data first. n8n orchestrates that pipeline; the model does the part it is genuinely good at.

The same discipline runs through the deal room, where HubSpot is the system of record and the product syncs to it rather than duplicating it. Integrate with the tools a team already lives in, rather than asking them to move.

Challenges

The hardest part of the second mission is not technical. It is holding a four-track roadmap in sequence when every track has someone who wants it first. The PRD exists for exactly that reason: to make the trade-offs visible so the decision belongs to the founders, and stays made.

Result

The AI-powered PPT generator took a sales workflow from 45 minutes to 30 seconds. The roadmap is sequenced, formalised and being executed track by track.

[ADD RESULT: outcomes on the acquisition and conversion tracks, once they can be measured and shared.]