Building the SME Automation Assessment

Every owner of a small business I talk to about AI asks a version of the same question: where should we even start?

This summer I launched Olaia Labs, the B2B brand for my training, automation, and advisory work. The site is deliberately lean. One page, one offer, one primary action. That action was “Request a diagnostic”, backed by a contact form. It worked, but it asked a lot of a visitor who had known the brand for ninety seconds. I wanted something that answered their question first and asked for their details second.

The result is the SME Automation Assessment, which went live in July 2026.

The problem with “book a call”

A diagnostic call is the right first step in an advisory relationship. It is the wrong first step on a website. Nobody books a call with a company they found ten minutes ago, and a form that says “tell us about your problem” puts all the effort on the visitor.

What I had been doing in those first calls was fairly repeatable. Ask about the company, ask how structured the processes are, ask about data and tools, ask who would own an automation once it exists, and then ask where the hours are actually going. From those answers I could usually say which workflow to automate first and roughly what it was worth. If the questions were that stable, they could be a questionnaire, and the first part of the answer could be free.

What the assessment does

The public tool is a five-step wizard:

1. Company profile. Industry, headcount, revenue range, and the primary business objective.
2. Process maturity. A handful of statements rated from one to five.
3. Technology and data. Data quality, integration, and how far digital tools have been adopted.
4. People and governance. Whether anyone would own an automation, and whether the organisation could sustain it.
5. Current pain. Hours per week on repetitive administration, hourly cost, manual handoffs, monthly errors, and the single most painful process.

Clicking “Generate assessment” produces an automation readiness score out of one hundred, with a short status and summary. That part is free and immediate.

The rest of the report sits behind a work email: the estimated annual value and recoverable capacity in hours, an operational profile across the four readiness dimensions, a recommended first workflow, an opportunity map of automation candidates, a 90-day implementation roadmap, and the risk and governance controls to put in place. Once unlocked, the whole report renders on the page and can be exported to PDF. The estimates are indicative, and the page says so.

Decisions that shaped it

Value before the gate, not after it. The score is the hook. It costs nothing, it is personal to the answers given, and it creates the obvious question of what is behind it. The email gate sits exactly at the point where the visitor already knows the tool has understood their business.

No pricing on the public page. There is an internal version of the same tool, used only in sales conversations, that adds the commercial model: implementation fee, retainer, running costs, payback, and return on investment. That logic does not exist in the public build at all. It is not hidden with a flag, it is absent from the shipped code, because anything that reaches a browser can be read. The internal tool can load a visitor’s exact answers from a code included in the lead notification, so the follow-up conversation starts from their numbers, not from scratch.

A static site, no backend. The whole assessment runs in the browser. Answers stay on the visitor’s machine until they ask for the full report, at which point the submission goes to Formspree and lands in my inbox with readable fields. Two Plausible events, one for a completed assessment and one for a captured lead, give me the funnel: how many visitors finish the five steps, and how many of those trade an email for the report. A large gap between those two numbers is the signal to rework the gate copy before touching anything else.

Portuguese first-class. The core market speaks Portuguese, so the assessment exists at /pt/assessment/ with the same scoring and a dictionary of display strings on top. The internal tool stays in English and still loads Portuguese submissions, because the process options are stored under English keys in both locales.

The fake email problem

The first test with a real user found the gap within minutes: typing asdf@asdf.com unlocked the full report. For a lead tool that is fatal, so the gate got harder. It now checks the address syntax strictly, refuses a list of disposable email providers, and performs a live DNS lookup over DNS-over-HTTPS to confirm the domain can actually receive mail. A non-existent domain gets a polite inline error, and the rejection fires its own analytics event so I can see how often it happens.

One rule mattered more than the checks themselves: they fail open. If the lookup times out or the network hiccups, the report unlocks anyway. Losing a genuine lead to my own validation code would be worse than letting a few fakes through. The remaining weakness is honest and documented: an invented mailbox at a real domain still passes, and the only true fix is an email round-trip, which a static site cannot do on its own.

What I took from it

Answer the visitor’s question before you ask yours. The old form asked for effort up front. The assessment gives something useful first, and the email request lands when it feels like a fair trade.

Anything that ships to the browser is public. Commercial models, rates, and margins do not belong in a client-side bundle, however well hidden. The right place for them is a separate tool that only you open.

Instrument the funnel on day one. Two events were enough to turn an opinion about the gate into a number. Like SegmentAI Analyst, this was built with Claude Code, and the time saved on implementation went into the parts that only a human can decide: what to give away, what to protect, and what to measure.

If you run a small or medium-sized business and want to know where automation would pay off first, take the assessment. The score takes a few minutes and costs nothing.