What Early PMF Looks Like for a Utility

People ask what product-market fit looks like before you have a sales team. For a utility, it is not a NPS screenshot. It is a loop we can watch in logs: someone lands from search, the tool works in one sitting, they come back or they share the URL. We watched that loop on Deskzy (a no-signup file and link toolkit) and it taught us more than any positioning doc.

The loop, in numbers we actually saw

Deskzy is taking 1M+ requests across 20+ countries. One busy day cleared 120k requests. In a quieter seven-day window we saw about 43k requests against 4.3k visits. Those numbers are not a valuation. They are a description of behavior: people do not browse Deskzy. They arrive with a job (compress this PDF, shorten this link) and leave.

That ratio (many requests per visit) is the tell. A brochure site has one request per pageview. A tool that works has the user hitting the same origin again while they iterate: merge, split, retry, copy the short link. If you only count "users," you miss the product.

Constraint first, features second

Most free PDF sites are an upload form plus a server. That architecture is easy to build and impossible to trust. We inverted it: compress, merge, split, resize, and convert run in the browser. The file never touches our disk. That constraint killed a class of roadmap items (server-side OCR, "email me the result") and made the remaining items honest.

• No account required to finish the job, signup is a conversion event, not a gate
• No file on our servers means no "we got breached, here is your NDA" story
• Client-side work has a cost ceiling we like: CPU on the user's machine, not our GPU bill
• The Pro tier has to sell something the browser cannot do, edge analytics on links, not a paywall on merge

If your paid plan is the same tool with the free badge removed, you do not have a business. You have a popup. We priced Pro at ₹399/month for colo-level analytics on short links, because that is a different job: marketers who already finished the free job and now need to know which country clicked.

Search is the sales team

Utilities do not get discovered at conferences. They get typed: "compress pdf no upload," "utm builder," "whatsapp link generator." We treated each tool as a landing page with one action, not a dashboard with twelve. SEO here is not a content calendar. It is matching intent to a working control in under five seconds.

What we do not do on those pages

• A hero video before the file picker
• A feature grid of other tools competing with the job they searched
• A mandatory newsletter to download the result
• A dark-pattern "your file is ready, create an account"

This is the same discipline we use on scoped client MVPs: one workflow, one done state. See what funded startups get wrong about AI MVPs. A utility just happens to have millions of tiny, identical workflows instead of one enterprise buyer.

Edge as product, not as resume line

Link analytics only matters if it is fast where the click happens. We route at the Cloudflare edge and expose colo-level demand (BOM, IAD, SIN, FRA, and the rest) because "global" is not a slogan when a short link is opened in São Paulo at 2am. The architecture is boring on purpose: the interesting part is that the free PDF path never shares a queue with the paid analytics path.

Building on the internet is weird

We had already shipped things that did not get this loop. After a few of those, the rule became: build the thing we personally open on a Tuesday. Deskzy is a collection of everyday tools (PDFs, images, links, QR, JSON formatters), not a platform pitch. The platform temptation is always there ("workspace! teams! API!"). It is how you lose the search-to-done loop.

The honest version of our learning: failed attempts were usually over-scoped. The version that moved was under-explained and over-shipped. That is the opposite of how we talk about legal OS work, and that is fine. Different jobs. A contract product must not skip citations. A PDF compressor must not skip the file picker.

A checklist if you are building a utility

1. Name the job in the user's words, not your category ("merge PDFs," not "document OS")
2. Make the first successful output possible with zero account, zero upload if trust is the blocker
3. Instrument request-per-visit, return visits, and share events, not just uniques
4. Charge for a second job, not for unlocking the first
5. Keep viral spikes off the path that has a GPU bill

Where this applies on client work

When a founder wants an "AI PDF product," we ask which of the two jobs they mean: in-browser tools with no custody of files, or server-side extraction with an audit trail. Mixing them is how you get a slow, expensive, untrusted hybrid. Deskzy is the first job. LexVault is the second. We run both. We do not pretend they are the same stack.

If you are scoping a public utility or a private document system, see how we work or send a brief. The Deskzy product page is the live constraint.