Section 1
The five challenges at a glance
Attribution for a small service firm fails in five compounding ways. First, the journey is mostly invisible: Gartner's buying-journey research finds B2B buyers spend only about 17% of their total buying time meeting with potential suppliers, and when comparing multiple vendors, any one firm may get 5-6%, with the rest spent in independent digital research (Gartner, 2024). Second, last-click reporting is structurally biased: Ron Berman's formal analysis in Marketing Science shows last-touch attribution rewards publishers and channels positioned late or prominently in the funnel and often decreases ad allocation efficiency (Berman, 2018). Third, correlation masquerades as causation: when eBay ran large-scale experiments pausing paid search, brand keyword ads showed no measurable short-term benefit and average non-brand returns were negative, because buyers simply clicked organic listings instead (Blake, Nosko & Tadelis, 2015). Fourth, the tools narrowed: Google sunset first-click, linear, time-decay, and position-based models in 2023, citing under-3% adoption, leaving advertisers only last click and opaque data-driven attribution (Search Engine Land, 2023). Fifth, out-of-market influence goes unmeasured: with ~95% of buyers not in market in a given quarter, brand-building touches pay off quarters later, far outside any report's window (Ehrenberg-Bass, 2021). The table summarizes the five; the sections that follow unpack the evidence and then assemble a measurement stack sized for a small firm.
Section 2
The invisible journey: what buyers do before they ever contact you
The first body of evidence concerns how buying actually happens. Gartner's research on the B2B buying journey finds buyers spend only about 17% of their total buying time meeting with potential suppliers; because that time is split across all vendors under consideration, a single firm's sales conversations may occupy just 5-6% of the journey (Gartner, 2024). The remaining time goes to independent problem identification, solution exploration, and requirements building across digital channels, a pattern Gartner formalized while also projecting that 80% of B2B sales interactions would occur in digital channels by 2025 (Gartner, 2020). For a service firm, the implication is direct: by the time a prospect submits your form, they have typically read your articles, checked reviews, asked peers, visited your site several times from different devices, and compared alternatives, none of which your last-click report records. The report instead logs the final step, often a branded Google search for your firm's name, and credits 'paid search' or 'organic search' with the entire client. The demand-side research compounds the problem. The Ehrenberg-Bass Institute's 95:5 heuristic holds that roughly 95% of category buyers are not in market in any given quarter, because firms change service providers around every five years (Ehrenberg-Bass, 2021). Marketing that builds memory with the 95%, content, community, consistent visibility, generates clients quarters later, when no tracking window connects cause to effect. Measured naively, your best long-term marketing will always look like your worst.
Section 3
The formal evidence: last-click bias and the causation trap
Two strands of academic research quantify the error. The first is Ron Berman's analysis in Marketing Science, which modeled how attribution methods shape advertiser and publisher behavior. The finding: under last-touch attribution, popular publishers and those appearing early in the funnel extract disproportionate benefit, advertisers misprice their buying, and the equilibrium often results in decreased ad allocation efficiency, while adopting more sophisticated attribution actually lowers the prices advertisers pay (Berman, 2018). Translated for an operator: the measurement rule itself distorts the market you buy ads in, and last click is among the most distorting rules available. The second strand is experimental. Researchers Blake, Nosko, and Tadelis ran large-scale field experiments at eBay, halting paid search in selected markets while continuing elsewhere. Published in Econometrica, the results showed brand keyword ads produced no measurable short-term benefit, users simply clicked the organic listing below, and returns on non-brand search were on average negative, because the ads largely reached frequent users who would have purchased anyway (Blake, Nosko & Tadelis, 2015). Click-based reports had implied healthy returns; the experiment revealed substitution, not causation. The honest summary for a small firm: attributed revenue is an upper bound on caused revenue, and the gap is largest exactly where reports look best, branded search, retargeting, and any channel that intercepts buyers who were already coming. This is why practitioners like Avinash Kaushik argue that nearly any move beyond last click improves decisions (Kaushik, 2013).
Section 4
The tooling reality: fewer models, more opacity, smaller data
Even firms that want better attribution face a narrowed toolbox. In 2023 Google removed four rule-based attribution models, first click, linear, time decay, and position-based, from Google Ads and Google Analytics 4, citing adoption below 3% of conversions, leaving only last click and its machine-learned data-driven attribution (Search Engine Land, 2023; Search Engine Journal, 2023). Existing conversions were migrated to data-driven attribution automatically (Ruler Analytics, 2023). For large advertisers, data-driven attribution may model credit reasonably; for a small service firm winning ten or twenty clients a month, it presents two problems. First, opacity: the model's credit assignments cannot be audited, so a founder cannot interrogate why a channel gained or lost attributed value. Second, data sparsity: machine-learned attribution needs conversion volume a small firm rarely generates, degrading gracefully toward, in practice, something close to last click. Layer on the measurement environment itself: cross-device journeys, cookie restrictions, and the offline steps that dominate service buying (phone calls, referrals, word of mouth) all fall outside web analytics entirely. A referral client who Googles your name before calling appears as 'organic search.' A podcast listener who types your URL directly appears as 'direct.' None of this means measurement is hopeless; it means the default stack measures a minority slice of a mostly offline, mostly invisible journey (Gartner, 2024) with a biased rule (Berman, 2018). The pragmatic response is to triangulate with data sources the defaults ignore, which is exactly what the solution framework below assembles.
Section 5
Innovative solutions
The research points toward triangulation rather than a single perfect model. First, self-reported attribution: adding a simple 'How did you hear about us?' field to every form and intake call captures the influence channels analytics cannot see, podcasts, referrals, communities, long-ago content. Practitioners pair it with click data precisely because the two disagree in informative ways; the gap maps the invisible 83% of the journey Gartner documents (Gartner, 2024). Second, CRM-source discipline: every lead gets a source, a self-reported source, and a closed-revenue outcome, so reporting runs on clients and dollars rather than clicks and sessions. Third, blended metrics as the executive layer: blended CAC (total marketing spend over total new clients) and channel-level payback sidestep per-touch credit disputes entirely, an approach consistent with Berman's finding that naive per-channel credit misprices spending (Berman, 2018). Fourth, simple incrementality tests: the eBay study's logic scales down, pause one channel for four to six weeks, or run it in one service area and not another, and watch total lead flow rather than attributed conversions (Blake, Nosko & Tadelis, 2015). Brand-keyword spend is the classic first test. Fifth, longer evaluation windows for brand-building activity: since ~95% of buyers convert in later periods (Ehrenberg-Bass, 2021), content and nurture should be judged on quarters-long cohorts, not monthly attributed revenue. Kaushik's framing is the operating principle: anything beyond last click is an improvement; the question is how much better you can practically get (Kaushik, 2013).
Section 6
Solution framework
A pragmatic measurement stack for a 5-7 figure service firm has four components. Core functionality: capture both digital and self-reported signals for every lead, tie them to closed revenue in one CRM, report on blended economics monthly, and test incrementality on the largest line items annually. Component one, dual-source capture: UTM-tagged campaigns and call tracking for the digital trail, plus a mandatory 'How did you hear about us?' on every intake, covering the ~83% of the journey that happens away from suppliers (Gartner, 2024). Component two, revenue-tied CRM: source fields persist from first touch to invoice, so the firm analyzes cost per client and payback by channel, not clicks, avoiding the mispricing dynamics formal modeling associates with last-touch credit (Berman, 2018). Component three, the decision dashboard: blended CAC, channel-level cost per client (both click-attributed and self-reported), and a rolling 12-month cohort view that gives slow-burn channels their real window (Ehrenberg-Bass, 2021). Component four, the experiment cadence: one incrementality test per quarter, starting with branded search, the channel the eBay experiments showed can collapse to zero incremental value (Blake, Nosko & Tadelis, 2015). The value proposition: firms reallocate budget based on caused revenue rather than claimed revenue, typically discovering both an overfunded interceptor channel and an underfunded creator channel. This measurement layer ships as the reporting core of LeadOS within LeverageOS. Implementation requirements: a CRM with custom source fields, call tracking, UTM conventions, a monthly one-page report, and the founder's commitment to ask every new client one question.
Section 7
Evidence-based action plan
Week one: add self-reported attribution everywhere, one required field on web forms, one scripted question on intake calls, logged verbatim in the CRM. This single change illuminates the referral, audio, and community influence the click trail misses, which Gartner's journey research says is most of the journey (Gartner, 2024). Week two: impose source hygiene. Standardize UTM tags, install call tracking, and create CRM source fields that persist through to closed revenue. Week three: build the one-page monthly report, blended CAC, new clients by click-attributed source, new clients by self-reported source, and revenue by channel. Where the two source columns disagree, investigate before reallocating; the disagreement is data, not error. Month two: run your first incrementality test on branded search, pausing it for four weeks while watching total lead volume, the eBay experiments found organic listings absorbed the clicks at zero cost (Blake, Nosko & Tadelis, 2015), and a small firm can verify the same in a month. Month three: re-baseline content and nurture on a 12-month cohort window, consistent with the 95:5 reality that most buyers convert in later periods (Ehrenberg-Bass, 2021). Ongoing: distrust precision, trust direction. The research consensus, from formal modeling (Berman, 2018) to field experiments (Blake, Nosko & Tadelis, 2015) to tool-vendor behavior (Search Engine Land, 2023), is that perfect attribution is unavailable at any price, but directionally honest measurement is available to any firm willing to triangulate. For adjacent evidence in this series, see [Referral Systems That Actually Work: What the Research Says](/blog/referral-systems-that-actually-work-research-playbook) and [The Compounding Economics of Content: A Research Deep Dive](/blog/compounding-economics-of-content-marketing-research).