Attribution Models Explained: Knowing What Actually Drives Revenue

First-touch attribution gives 100% credit to the first channel a customer engaged with. Last-touch gives 100% to the final one before conversion. Linear splits credit evenly across all touches in the journey. Time-decay weights recent touches more heavily than earlier ones. Position-based (U-shape) weights the first and last touches most heavily. Data-driven attribution uses machine learning to assign fractional credit based on observed conversion patterns.

Each model tells a different story about the same customer journey. Understanding which story matters for which decision is the entire skill of attribution.

Last-touch attribution systematically over-credits Google Search (because it is often the final query before a call) and under-credits every awareness channel — Meta, YouTube, direct mail, referrals, local sponsorships. Businesses that optimize purely to last-touch data eventually starve the top of their funnel and watch overall pipeline collapse.

The classic failure pattern: an operator moves budget out of Meta into Google Ads because 'Meta doesn't convert.' Six months later, Google Ads volume drops because the awareness pipeline Meta was feeding has dried up. By the time the operator realizes what happened, they have lost a full quarter of pipeline.

First-touch has the opposite problem. It over-credits the awareness channels and under-credits the conversion channels. An operator optimizing purely to first-touch would over-invest in top-of-funnel Meta at the expense of the Google Ads or LSA campaigns that actually closed the sale.

Neither extreme works alone. The customer journey is genuinely multi-touch, and any model that assigns 100% credit to a single touch is definitionally wrong.

Most local businesses do not have the data volume for statistically-valid data-driven attribution — you need thousands of monthly conversions for the ML to learn meaningful patterns. A workable hybrid for local service businesses: track first-touch AND last-touch for every lead, weight them 40/60 by default (adjustable), and cross-reference with self-reported 'how did you hear about us?' survey data captured at intake.

This is imperfect but far better than either extreme alone. The intake survey is the tiebreaker for ambiguous cases — a customer who touched Meta, then Google Search, then LSA, will often self-report 'my neighbor recommended you' or 'I saw your truck at a jobsite,' revealing offline touches that no digital attribution model captures.

GA4's default is data-driven attribution, which is a substantial improvement over legacy last-click Universal Analytics. For businesses with 300+ monthly conversions, GA4's data-driven model produces meaningfully more accurate channel credit than any single-touch alternative.

For businesses under 300 monthly conversions, GA4 falls back to position-based attribution — which is a fine starting point but should still be cross-referenced with intake survey data.

The hardest part of local service attribution is offline touches: yard signs, truck signage, direct mail, door hangers, community sponsorships, word-of-mouth referrals. These produce meaningful pipeline but rarely appear in any digital attribution model.

The only reliable way to capture offline attribution is the intake survey — a single question at first customer contact: 'Just so we know what's working, how did you hear about us?' Ask it every time. Log the answer in your CRM. Report it monthly alongside digital attribution.

The three attribution-informed decisions that matter most for local businesses: (1) which channels to increase spend on next month, (2) which channels to cut or renegotiate, (3) which new channels to test with reallocated budget. All three decisions should be informed by 90-day trailing attribution data (not week-to-week noise) combined with LTV data (channels producing higher-LTV customers deserve budget priority).

The five mistakes we see most often: making budget decisions on 30-day attribution instead of 90-day; ignoring offline touches entirely; treating GA4 as truth without cross-referencing intake surveys; over-crediting brand search (brand search converts because other channels produced the awareness); and rebuilding attribution setups every time a new agency or consultant recommends a different model.