In July we published a page of lead source statistics: attribution data from real customer websites, positioned as more honest than the usual vendor survey because it showed its sample size, its limitations, and its cohort bias up front. The single most quoted number on that page was that phone taps made up 52.7% of recorded leads, the majority lead type.
That's not the majority anymore, and part of why the original number wasn't representative is a mistake in our own pipeline, not just a small sample growing up. We're publishing this correction on its own rather than quietly updating the numbers, because a page whose whole premise is data honesty has to hold itself to that standard when the data turns out to have a problem.
What we got wrong
The July dataset was described as "9 Australian small-business websites, managed by one agency." The pull that produced it did not exclude our own internal sales-demo account, which seeds three fabricated business websites (a real-estate agency, a plumber, a dental practice) with synthetic, nightly-regenerated leads used for sales calls and product screenshots. Those three fake sites were very likely counted alongside the 9 real client sites in the original figures.
We didn't catch this while building the original page. We caught it while pulling numbers for this quarter's refresh, when the site and lead counts didn't line up the way they should have.
What's different now
Across the current, larger sample of 26 real contributing sites (up from 9), form submissions are now the majority lead type: 48.2% forms to 44.3% phone taps. Two things happened, and they're both true at once.
First, the pipeline fix above removed synthetic data that skewed the original figure. Second, the tracked cohort genuinely grew, from 9 trades and home-services sites managed by one agency to 26 sites that now include SaaS, agency, e-commerce, healthcare, and immigration-advisory businesses. Form-led businesses joined, and the average moved toward forms.
We checked which of those two actually explains the reversal, rather than assuming. Recomputing phone share on the exact same 9 original sites, extended to the current window, phone taps are still running above 50% there, slightly higher than in July, not lower. The finding didn't reverse for those businesses. The mix of businesses changed underneath the headline number.
The number that survives
Phone share by trade: 97.1% for plumbing (now across 2 sites and 623 leads, up from a single-site 98.1%), 70.7% for cleaning, 51.8% for locksmith and car keys, 36.5% for electrical, 18.6% for blinds and window furnishings, and 18.0% for painting. That range was always the more useful fact than any single headline average, and it's what the phone-versus-form section on the stats page leads with now, not the overall split.
Everything else that moved
Unpaid channels (organic search, direct, and referral combined) are now 67.9% of leads, up from 54% in July, as paid traffic's share fell from 46% to 32.1%. Direct traffic, leads with no referrer and no UTM data at all, rose from 19% to 24.1%. The lead journey shifted slightly longer: a median of 2 pages viewed before enquiry instead of 1, 1.8 minutes to enquiry instead of 1.6, and 83.6% converting on the first visit instead of 87%. Small movements individually, but they all point the same direction as the cohort broadens beyond fast-decision trade callouts.
Day-of-week and hour-of-day figures also needed a second look. Restricted to the Australian-majority sites in the cohort (mixing in real international traffic was distorting the "AEST" framing), Tuesday is now the busiest day, not Monday, though the weekday spread is much flatter than it was. Out-of-hours enquiries are 45.6% on that same Australian-only slice, close to the original 45% once you account for the two releases using slightly different definitions of business hours.
What we're changing going forward
The stats page methodology now publishes the exact exclusion rule we use: internal and demo accounts, any site on a sandbox domain, any account created for internal testing, and our own product, agency, and personal websites. That rule is checkable against every future refresh, which is the actual fix here, not just corrected numbers for one quarter.
We're also holding back a Google Ads search-term section rather than publish one prematurely: 9 accounts now have Google Ads connected, but only 5 have enough resolved search-term data to be worth showing, and most of that is one site. We'll build that section once it reflects more than one business's keywords.
Full numbers, the by-trade breakdown, and the complete methodology are on the lead source statistics page itself, refreshed 7 September 2026. The CSV export has been updated to match. The next refresh is due in December.
