The presales engine

The problem: nobody owned the request

When I stepped into presales, requests from sales landed in a shared chat space with every design lead and project manager across four studios. Whoever felt they had bandwidth picked the work up.

It looked flexible. In practice, nobody owned anything. Work landed unevenly across leads and locations. Quality depended on who happened to grab a request, not on who was best suited to it. No one owned a timeline, so the sales team ended up acting as project managers, chasing each lead individually several times a week instead of selling.

Estimates took about a week. Proofs-of-concept took two to three weeks. And the market had just shifted underneath us: lead volume across the industry reduced dramatically, and every pitch had more agencies in it, most of them trying to undercut in this competitive number battle.

Slowness was the symptom. The real problem was ownership.

What I built

One owner, and a map of who's good at what. I took single ownership of all design-side presales, so sales had one point of accountability instead of a chat thread. Then I ran a workshop where the design leads mapped their own strengths: domains, websites versus applications, comfort in front of a client. Combined with my own read of the team, that became a cross-studio roster. For any request, we knew who should take it.

An estimation model, encoded rather than replaced. We already had a parameter-based estimation model: screen counts, UX and visual complexity for each screen, platforms, discovery workshops, buffers. The model was sound, but it was applied by instinct. Every lead estimated differently, and instinct pads for uncertainty. I encoded the entire model into an AI skill. Requirements calls are recorded with the prospect's consent, and the skill turns the conversation into a draft estimate within hours.

Two human checkpoints stay in the loop. The first is design judgment: a trained designer verifies every parameter before anything is generated. The second is commercial judgment: every estimate is checked against sellability benchmarks for each project type, which I built with the sales team.

Proofs-of-concept prospects can touch. We stopped sending static decks. For each POC we agree a deliberate scope with the prospect (a homepage, or a screen that shows real design thinking, never a login flow) and deliver a hosted, interactive prototype they open as a link.

AI handles the production. Designers own the ideas. Directions start as their own sketches, and designers decide what the AI over-produced, what to cut, and every visual choice from typography to motion. The speed comes from the machine. The originality comes from the people.

How adoption happened

I didn't train the bench in a classroom. I took designers onto live presales calls with me, so they learned by doing the work alongside me and, just as importantly, found out they enjoyed it. People volunteer for work they like. They resist work they're assigned.

When I was stretched thin, I put the people I trusted in my place and watched how they did. Once I was confident in them, I nudged the sales team to go to them directly. They do. Today roughly a third of all requests, close to 40% in a strong month, go straight to designers I trained, without me involved.

I never set a tighter deadline. I changed the method, and the timeline followed.

What changed

• Estimates: from about a week to a day.
• Proofs-of-concept: from two to three weeks to five working days.
• Conversion: lead volume halved, and monthly closures held, at premium pricing against cheaper competitors.
• Accuracy: the tighter estimates held up in delivery. Where projects ran over, the cause was scope the client added after signing, not the estimate.

One pitch made the point better than any metric. A major Indian industrial group was launching electric trucks and buses and needed a launch website built around scroll-driven video storytelling. Instead of presenting an approach, we built a working homepage, with AI-generated video turned into a live scroll experience, and put it in front of their senior leadership about a week after our first conversation.

We won the project. Only afterwards did I understand why so clearly: they had six weeks to design, build, test, and launch. Our speed wasn't a nice-to-have. It was the unspoken criterion they were evaluating us on.

What I'd do again, anywhere

1. Fix ownership before speed. You can't accelerate a process nobody owns.
2. Encode your best judgment; don't replace it. Our estimation model already existed. AI made it consistent and fast.
3. Keep humans at the judgment points, design and commercial, and automate everything between them.
4. Build the bench on the job. Presales capability transfers by doing the work together, not through documentation.
5. Speed is a client experience. Prospects evaluate things they never tell you. A working prototype in five days answers questions no deck can.