Onboarding Findings Report
Cumulative pool: ClickUp, Notion, PostHog, Attio, Linear, HubSpot
One new onboarding walkthrough was processed this cycle — HubSpot — and is analyzed in full below, screen by screen, with embedded evidence frames. The source material was two screen recordings (~23 min 54 sec combined, spanning two separate signup sessions under two different Google-linked HubSpot trial accounts) plus four still frames captured directly by the person relaying the files. ClickUp, Notion, PostHog, Attio, and Linear were evaluated in prior sessions; their footage is not present in this session's working files, so they are carried forward at their existing ratings with a short summary rather than re-analyzed from scratch.
Every friction point below was checked against five possible causes — copy, hierarchy, placement, load time, motivation — with the ones that don't apply explicitly ruled out and a reason given, not just assumed. One finding (H3) involves a checkout screen that was still on a loading spinner when the source recording ended; that finding says so plainly rather than asserting the checkout was permanently stuck.
Rated in a previous run; source frames aren't available in this sandbox, so no new screens are analyzed here.
Rated in a previous run; source frames aren't available in this sandbox, so no new screens are analyzed here.
Full analysis in the prior report run: a placeholder-styling bug, two stacked decision screens, and a developer-only path to real data.
Full analysis in the prior report run: a live business email rejected at signup with no graceful recovery path.
Full analysis in the prior report run: sub-30-second signup to a working, populated workspace.
HubSpot
~23 min 54 sec across two recordings · ~15 distinct product states · signup via Google OAuth · two separate Sales Hub Professional trial accounts
Both trial signups tested here complete without a hard block, and the core CRM — Contacts, Deals, Marketing Emails, Segments, Blog — is real and navigable in both accounts. The friction is cumulative rather than fatal: proactive chat interruptions before the tester has done anything, a guided setup step whose label doesn't match its own content, a purchase button positioned ahead of any onboarding guidance that leads into an unresolved checkout, and the same context-free loading spinner recurring across unrelated parts of the product. Four findings below, grouped by theme rather than strict chronology since the evidence spans two independent sessions.
H1 — Proactive chat prompts interrupt twice in one session, unprompted by any tester action
Neither prompt was triggered by a click on a help icon or any visible hesitation — both appear to be timed or scroll-triggered proactive messages. The second one surfaces directly beside a “Buy now” button (see H3), stacking a sales-chat nudge on top of a sales CTA in the same viewport.
Copy ruled out — both messages are polite, on-topic, and not misleading about what the bot can do.
Hierarchy ruled out — the widget is a small corner element, not blocking primary content or CTAs.
Placement ruled out — bottom-right is the standard convention; nothing is hidden underneath it.
Load time ruled out — both bubbles render promptly; this isn't a performance artifact.
Motivation is the likely driver — the prompts appear to fire on a timer or scroll trigger regardless of whether the visitor shows any actual sign of being stuck, which reads as interruption rather than assistance to someone who hasn't asked for help.
H2 — The setup wizard's first step is labeled “Business challenges” but shows an email-connection prompt instead
This was reached from the Home screen's “Open your setup plan” link, immediately after the “Connect Google” CTA was already visible once on Home itself — making this the second email-connection ask before the wizard has asked a single question that matches its own step label.
Copy ruled out — the visible panel text (“Connect your email and we'll enrich your contacts for a complete customer view”) is itself clear and accurate about what it's asking for.
Hierarchy is the actual issue — the step architecture doesn't match its own content; a step called “Business challenges” contains zero business-challenge content.
Placement ruled out — both the step list and the content panel are clearly positioned and visible.
Load time ruled out — no delay reaching or reading this screen.
Motivation ruled out as primary — the tester had just clicked into the wizard directly from Home; the mismatch is structural, not a dip in willingness to proceed.
H3 — The first prominent CTA on a brand-new trial's Home page is a purchase button, and clicking it leads to a checkout that hadn't resolved when the recording ended
The click on “Buy now” was decisive, not hesitant, and immediately produced a legitimate checkout context (the “Exit Checkout” escape hatch only appears inside real checkout flows). What happened next isn't fully known from this footage.
Copy ruled out — “Buy now” and “Talk to Sales” are honest, unambiguous labels.
Hierarchy is partly responsible — the purchase CTA is the first bold interactive element reached while scrolling Home, outweighing any onboarding-guidance element in visual priority.
Placement ruled out — the CTA sits in the natural reading flow, not hidden or mislabeled.
Load time is the open question, not ruled out — a spinner with no progress indicator was the last thing visible; whether it was a genuine hang or a normal transition caught mid-render can't be confirmed from a single frame at the end of the recording.
Motivation ruled out — nothing in the tester's behavior suggests reluctance; the click was direct.
This finding is intentionally reported as uncertain rather than as a confirmed outage — the evidence supports “a purchase flow was entered and a spinner was the last thing seen,” not “checkout is broken.”
H4 — The same generic, unlabeled loading spinner recurs across at least five unrelated contexts
The identical unlabeled circle — no percentage, no skeleton UI, no estimated wait, no timeout message — appears at a full-page app boot, mid-navigation inside a loaded page, on a feature's live-preview panel, at the multi-account switcher, and at the checkout in H3. It is the product's only loading affordance anywhere it was observed.
Copy ruled out — there is no copy in any instance; the absence of status text is itself the finding, not a wording problem.
Hierarchy ruled out — the spinner is always centered and immediately visible, never buried.
Placement ruled out — consistent center-of-viewport placement across every instance.
Load time is not the root cause — some of these resolved quickly and some (the AEO preview) did not resolve at all within the session; the shared problem is the total absence of progress indication regardless of actual duration.
Motivation ruled out — occurs on routine navigation and at checkout alike, unconnected to any specific dip in the tester's intent.
Outside signal
Independent signal on HubSpot's onboarding is genuinely mixed, and honesty requires reporting that rather than only the corroborating half. On the positive side, G2-derived data puts “ease of setup” at roughly 85% and notes most reviewers go live within weeks without dedicated IT support — “plug-and-play setup rather than a complex onboarding process” is a fair characterization of the core CRM for most users. On the friction side, independent write-ups describe the platform as capable of feeling “confusing if you're new to CRMs” given “the sheer number of dashboards and features,” and separately flag that paid onboarding is itself a real cost line ($1,500–$3,500 depending on tier) — consistent with H2's observation that HubSpot invests deliberately in a guided setup flow, which makes a labeling mismatch inside that same flow more notable, not less. Nothing in outside signal directly corroborates the specific chat-interruption or spinner-recurrence findings (H1, H4); those are being reported on direct observation alone and should be weighted accordingly.
Neither trial account tested here hit a hard block — both signups completed and the core product (Contacts, Deals, Marketing Emails, Segments, Blog) is real and usable. But a new user absorbs two unprompted chat interruptions, a guided-setup step that doesn't match its own label, a purchase CTA that outranks onboarding guidance in visual priority, and a loading spinner that offers zero status information anywhere it appears — including, unresolved in this footage, at the one moment (checkout) where clarity matters most. None of these are fatal individually. Stacked together across two independent sessions, they're a real, repeated tax on first impression rather than a one-off.
Cross-cutting observations, this cycle
- Loading feedback is the single weakest point observed across this entire cycle. HubSpot's unlabeled spinner (H4) recurred more times, in more unrelated contexts, than any single finding in the prior five-tool pool — a strong candidate for a cross-tool pattern to watch for explicitly in future evaluations.
- HubSpot is the only tool evaluated so far where the two accounts tested (Sales Hub vs. Marketing-Hub-navigation trial) surfaced meaningfully different Home-page content for what should be a comparable first-run state — one led with onboarding guidance, the other led with a paid upsell. Whether that's account-type-driven personalization or session variance isn't confirmed by this evidence and would need a controlled re-test to separate.
- This is the first tool in the pool where a finding (H3) is reported as evidentially incomplete rather than resolved either way. That's a deliberate choice, consistent with the standing rule against manufacturing certainty the footage doesn't support.
Pool is now 6 tools (ClickUp, Notion, PostHog, Attio, Linear, HubSpot) — the 6–8 threshold for the 3-finalist case-study section has been reached. That synthesis hasn't been produced in this run; it needs an explicit decision on selection criteria before three of the six get promoted to full case studies, which wasn't part of this cycle's brief.