I stopped writing proposals and started shipping demos

A prospect once replied to my outreach with the line every service business dreads and craves in equal measure: "send me something and I'll take a look." I used to spend an evening on that reply — a PDF with a scope, a timeline, some past-work screenshots, a price. It was fine. It was also, I eventually realized, indistinguishable from what any of my five competitors would send in response to the same line. Everyone's PDF looked competent. Competent doesn't win anymore, because competent is now the free, instant output of any model a competitor happens to open.

That's the actual shift worth naming. Proposals used to be a costly signal — spending a night on a document proved you were serious. Now that cost has collapsed to nearly zero for everyone, which means a well-written proposal proves nothing except that you, too, have access to a language model. The only thing left that still separates you from the pile is a proposal that contains work you've actually already done, specific to them.

So that's what I started sending instead: a small, real, working demo, built for that one prospect, attached to a one-page proposal instead of replacing it.

Why this is suddenly cheap

This strategy existed before agents — good salespeople have always known that showing beats telling. What changed is the economics. A custom demo used to mean a few days of a developer's time you couldn't spare for every warm lead, so you rationed it for the biggest deals only. A coding agent turns "build a thin, real prototype against this specific prospect's actual page or data" into a few hours of agent time plus maybe an hour of my review. That's cheap enough to do routinely, but it is not free, and pretending it's free is exactly how you end up doing unpaid client work disguised as sales effort. More on that below.

Scoping it in hours, not days

The playbook has a specific order, and skipping steps is where it goes wrong.

First, a browser agent audits the prospect's public surface — their storefront, their product pages, their content cadence, their reviews, whatever your service actually touches — and produces a findings list. Not vague observations. Specific, verifiable ones: "your product page has no FAQ section," "your most-repeated review complaint is about shipping time and no response addresses it," each with a screenshot or a quote attached, not a guess.

Second, from that findings list, scope exactly one demo buildable in roughly four hours of agent time that makes two or three of those findings tangible using the prospect's actual product, actual copy, actual data. Not a generic template with their logo pasted on. A redesigned mockup of their real page. A working prototype dashboard running on their public numbers. A sample deliverable built from their actual content. The specificity is the entire point — a generic demo is just a fancier PDF.

Third, the proposal itself gets short, because the demo is now doing the persuading: their problem stated in their own numbers, a link to the demo, scope, timeline, price, and one paragraph on how the engagement would actually run. One page, not five.

When not to do this

This only makes sense above a deal-size threshold, because the hours aren't free even if they're cheap. My rule is roughly 20 to 50 times the agent-and-review cost — below that, a solid templated proposal is the right tool, and building a demo for a small deal is just expensive theater that doesn't pay for itself even at a decent close rate. There's also a category of prospect where a demo reads as presumptuous rather than impressive: mocking up someone's brand and product without asking first can land badly, especially where showing up uninvited with "I redid your website" reads as overstepping rather than helpful. I now ask a light permission question in the first reply — something like "mind if I mock something up on your actual page?" — before building anything. It costs nothing, and the answer is itself a useful buying-intent signal: enthusiastic yes, lukewarm yes, or silence all tell you something before you've spent a single agent-hour.

Prospect: [company/URL]. Service: [what I sell]. Step 1: audit
their [store/site/content] and list 5-8 specific, verifiable
findings with evidence (screenshot/quote/URL each — no guesses).
Step 2: propose ONE demo buildable in ~4 hours that makes 2-3
findings tangible using THEIR actual product/data/content.
Step 3: after my approval, build it at [staging URL].
Step 4: draft a one-page proposal: their problem (their numbers),
demo link, scope, timeline, price [range], working process.
Tone: [voice doc]. Do not send anything.

Where it goes wrong

The findings list is where trust gets built or destroyed in the first ten seconds of a prospect's attention, and it is also where an agent's confidence outruns its accuracy fastest. One early audit confidently reported that a prospect "had no email capture" on their site — the capture form existed, sitting behind a popup the agent's browsing session had already dismissed before it looked. Sending that claim to a real prospect would have been an instant credibility loss, not a minor inaccuracy.

The second failure is scope creep, and it's a subtler trap because it feels like generosity in the moment. My second demo, built without a hard stop, quietly turned into something like 60% of the actual proposed project — built for free, before any contract existed. The fix wasn't willpower, it was a mechanical constraint: a four-hour cap on the build, and a framing repeated in the prompt itself — prove capability, don't deliver the value. The demo's job is to make the prospect believe you can do the work. It is not supposed to be the work.

I still read everything before it goes out — the findings list especially, since an agent's wrong claim about a stranger's business is the one failure mode with no good recovery — and I send it personally rather than letting any part of the send itself run unattended. Whatever happens with the deal, win or lose, it goes into the same follow-up tracking I use for everything else, and I ask directly what actually decided it. That answer feeds back into the next findings list and the next demo scope, which is the only reason this keeps getting better instead of just staying a nice trick I did once.

The underlying logic generalizes past sales, and it's worth stating plainly: whenever the artifact you'd normally produce to prove seriousness has gotten cheap for everyone to fake, the only remaining signal is proof that can't be faked cheaply — something that only works because it's actually true. A demo running on someone's real data is one of the few things left in that category.