pilotads

Results / Teardown · HR-tech SaaS · Utrecht · 10 weeks

Demo CPA from €212 to €96: a teardown

By Ines Beltrán, Senior Google Ads Strategist · Published 30 June 2026 · Figures are representative and anonymized

An HR-tech SaaS company in Utrecht came to us paying €212 per demo request on roughly €18,000 of monthly spend. Ten weeks later the same budget bought demos at €96. Nothing clever happened: two structural failures got fixed in the right order. Here is the whole mechanism.

€212 → €96

cost per demo request

31%

of spend on unreviewed terms

412

negative keywords added

What was broken

The audit found two failures feeding each other. First, search-term hygiene: 31% of spend sat on queries nobody had reviewed in six months. The account sold HR software for mid-sized employers, yet paid for clicks on salary questions, job searches, and a competitor's help-desk queries. Broad match had drifted far beyond the seed keywords, and the negative lists had not been touched since launch.

Second, measurement. The consent banner blocked Google tags outright for non-consenting visitors instead of running Consent Mode v2, so no modeled conversions existed for EEA traffic. Reconciling against the CRM showed Google Ads undercounting demo requests by roughly a quarter. Smart Bidding had been optimizing a €212 CPA that was partly a counting error, and reacting to phantom cost swings week after week.

The fixes, in order

  1. Weeks 1–2: Consent Mode v2 wired through the CMP, GA4 demo events rebuilt and imported as primary conversions. Measurement first, because every later judgment depended on real numbers.
  2. Weeks 2–5: search-term purge. 412 negatives added across shared lists in three themed groups: careers, informational salary queries, and competitor support terms. Weekly review cadence started, per our standard guardrails.
  3. Weeks 5–8: restructure. Two bloated campaigns became four intent tiers with exact-match cores; tCPA was reset from actual post-repair data instead of the old phantom baseline.
  4. Weeks 8–10: budget re-allocation toward the two tiers producing demos below target, and a capped Performance Max test that was kept only after it stopped cannibalizing brand queries.

The CPA line, week by week

WindowCost per demo requestWhat was running
Intake baseline€212 (partly phantom)blended structure, broken counting
Weeks 1–2€168 measuredconsent repair alone; the drop is counting, not skill
Weeks 3–5€139negative purge underway, budget unchanged
Weeks 6–8€121restructure live, learning phases in progress
Weeks 9–10€96reset tCPA, budget re-weighted to converting tiers

The first drop deserves its honest label: €212 to €168 was a measurement correction, revenue the client had been earning all along without seeing. We print the split so the mechanical work and the real optimization each get their own credit, and so the next client can recognize which part of their own gap is phantom.

What we left alone

No landing-page rebuild, no new creative platform, no bid to expand into Display. The demo pages converted adequately once the right queries reached them, and every euro of attention went where the audit priced the leak. One proposal from the client's side, pausing brand campaigns entirely to "save money," was declined with the counter-math in the change log: brand clicks cost cents and defended a funnel competitors were bidding on. Restraint documented is trust earned; the log entry took four sentences and settled the question permanently.

The lift, with context

Cost per demo request ended at €96 against €212 at the start, measured on the corrected conversion data both times. For context, the 2026 WordStream benchmarks put median business-services cost per lead at $93.69, so €96 for a scheduled product demo in a competitive B2B niche is good, not miraculous. Impression share on the exact-match core terms rose nine points. Ines Beltrán ran the weekly reviews on this account and still flags it internally as the cleanest example of sequencing: measurement, then hygiene, then structure, never the reverse.

What this teardown generalizes to

If your account has not had a search-term review in months, the same leak is almost certainly present; the median across our audits is 21% of budget. The order of operations matters more than any single tactic, which is why the €1,900 audit sequences its fix plan by recovered euros per hour of work, and why tracking repairs always come first in it.

Suspect the same leaks?

The audit that started this engagement is fixed-price and takes ten business days.

Get a proposal