[ Teardown 01 · measured 2026-08-13 ]

I pointed a robot at seven B2B signups. It found almost nothing.

That is the finding. The signup form — the thing everyone means when they say “our onboarding needs work” — is the one part of activation that is already solved almost everywhere. The losses start one field past where any robot has to stop.

I built a small instrument that walks a product's public path the way a first-time visitor would: land on the homepage, find whatever invites you in, follow it, and read the form at the end. It counts what is asked of you and how much stands in the way. It creates no account and submits nothing.

I ran it against seven well-known B2B products on the same afternoon. I expected a catalogue of friction. Here is what came back.

[ What the instrument counted ]
ProductHow signup startsClicksInputsRequiredSSOPrices publicCard for trial
Monday.comDedicated page110GoogleYesNot required
IntercomDedicated page120GoogleYesNot required
AmplitudeDedicated page110GoogleYesnot stated
AsanaEmail box on the homepage010none yetYesNot required
PipedriveEmail box on the homepage031none yet403 to a robotNot required
ZendeskBot challenge before the formYes
HubSpotNot reached past the consent wallYes

Inputs = fields visible at the point signup begins; site search boxes excluded. “none yet” = no SSO offered at that step, which is not the same as none at all. Blank cells are unknowns, not zeros.

[ Four things worth saying ]

SOLVED

The form is not where anyone is losing people.

The median product here asks for one field, one click from the homepage, offers Google sign-in, wants no credit card, and publishes its prices. There is no fat left to trim. Any consultant who arrives, counts your form fields and leaves is selling you a finding from 2014.

STRUCTURAL

Two of seven make you commit before you can see the queue.

Asana and Pipedrive both begin with a single email box on the marketing page. It reads as generous — one field! — but handing over an email buys you no information about what comes next: how many more questions, whether a card appears, whether a colleague has to approve something. People who cannot see the length of a queue assume the worst. The ones who do continue arrive braced for a sales call.

NEW

Two of seven cannot be entered by an automated visitor at all.

Zendesk puts a bot challenge in front of its registration page. HubSpot’s consent dialog covered the page and my walker found no way through to a signup control behind it. Both are deliberate, both are defensible, and the second is as much a limit of my instrument as a finding about the product. But it is worth knowing precisely where that wall sits — because an increasing share of evaluation now happens with an assistant driving, and everything without a normal browser fingerprint stops exactly there.

HYGIENE

The page that should be lightest is often the heaviest.

Asana’s entry step pulled roughly 7.7 MB and 51 third-party scripts; Monday 23 scripts, Amplitude 15, Intercom 8. Marketing tags accumulate on precisely the screen that needs to be fastest, and each one is another way for the form to break for somebody, somewhere, on a worse connection than yours.

[ Where the instrument has to stop ]

The measurable part
is the part that is fine.

Lay the funnel out end to end and mark the point past which no automated audit can go.

Landingcounted
Invitationcounted
The formcounted
STOP
Empty workspaceunmeasured
Getting real data inunmeasured
Inviting the teamunmeasured
First job actually doneunmeasured
Day twounmeasured

Solid = observable without an account. Dashed = where B2B activation is actually won and lost.

Everything to the left of that marker is cheap to measure, which is why it has been measured to death and why all seven products score well on it. Everything to the right needs an account, a colleague, a real dataset and a second day — and it is where a trial quietly becomes a customer or quietly becomes nothing.

So when a tool offers to audit your onboarding automatically, notice what it is promising: a very precise reading of the five percent that is already fine. I built one of those tools. This is me telling you what it cannot do.

[ The questions a machine cannot ask ]

These are what I go looking for once I am inside a product, and none of them are countable.

  • What is the first moment this product is unmistakably worth the trouble — and how many minutes and decisions away from signup does it sit?
  • What does the workspace look like with nothing in it, and does that emptiness read as an invitation or as homework?
  • Who else has to be present before the product does its job, and at what point are they asked for?
  • What is the product asking someone to decide before it has earned an opinion from them — configuration, naming, structure, plan choice?
  • What brings them back on day two, and was it set up on day one?
[ Method, and its limits ]
  • Measured on 2026-08-13 with an ordinary desktop Chromium at 1440×900, US English, from a single location, on the public path only.
  • No account was created and no form was submitted on any product here.
  • No bot protection was bypassed. Where a challenge appeared the run stopped and that was recorded as the result — which is why two rows are mostly empty.
  • Counts are facts; the thresholds that turn a count into a flag are stated heuristics, not laws.
  • One afternoon, one run each. Marketing pages get tested and rotated constantly, so treat any single cell as a snapshot rather than a verdict.
I will point the same instrument at your activation path and send back the measurements — and then, unlike the instrument, tell you which single thing I would change first and why it is that one rather than the other thirty-nine.