THINXSTER
Blog/AI Marketing
AI Marketing7 min readAugust 23, 2026

Best Attribution Model for Marketing: How to Choose

Data-driven attribution plus a "how did you hear about us?" field beats every alternative for most service businesses. Here's when to add more, and when not t

RK
Ryan Korsz
Founder & CEO, Thinxster

TL;DR

Data-driven attribution plus a "how did you hear about us?" field beats every alternative for most service businesses. Here's when to add more, and when not t

→ See how this applies to your business (free 30-min call)

Most service businesses should use data-driven attribution in GA4 plus a self-reported "How did you hear about us?" field on every form and phone call — and treat last-click as a diagnostic, not a decision tool. If you spend under roughly $10,000/month on ads and close fewer than 100 deals a quarter, stop there; anything more sophisticated will cost more in setup time than it returns in better decisions. Above about $50,000/month in blended spend, add a media mix model or holdout tests. The honest answer is that no attribution model is "correct." Each one is a different opinion about who deserves credit for a $4,200 roof repair that started with a Facebook video in March and ended with a branded search in June.

The Six Models, Ranked by How Often They Mislead You

  • Last-click — 100% of credit to the final touch. Systematically overpays branded search and retargeting. In a typical home services account, branded search will show a 12:1 ROAS under last-click and 3:1 under any incrementality test, because those people were already coming.
  • First-click — 100% to the first touch. Overpays top-of-funnel and awareness. Useful as a *bookend* against last-click: if a channel looks strong in both, it's real.
  • Linear — equal credit across all touches. Simple, transparent, and wrong in a predictable direction: it inflates cheap, high-frequency touchpoints like display and email.
  • Time-decay — credit weighted toward recency, usually a 7-day half-life. Reasonable default for sales cycles under 30 days. Most HVAC and plumbing emergencies close in under 48 hours, so time-decay and last-click produce nearly identical numbers there.
  • Position-based (U-shaped) — 40% first, 40% last, 20% split among the middle. Defensible for 60–120 day cycles like roofing, solar, or commercial cleaning.
  • Data-driven (DDA) — Google's algorithmic model, comparing converting and non-converting paths. It's the GA4 default and the best free option, but it needs roughly 600 conversions and 400 non-converting paths per 30 days in a given model type before Google will produce stable weights. Below that threshold, GA4 quietly falls back toward last-click and you won't get a warning banner.
  • That threshold is the single most-ignored fact in attribution advice. A plumbing company doing 90 leads a month has a data-driven attribution setting turned on and is functionally reading last-click numbers with extra steps.

    Match the Model to Your Sales Cycle, Not Your Ambition

    Pull your average days-to-close and average touches-to-close from your CRM before choosing anything. Rough guidance:

  • Under 7 days, 1–2 touches (emergency plumbing, locksmith, towing): last-click is fine. Attribution modeling will change your budget by less than 5%.
  • 7–45 days, 3–6 touches (HVAC replacement, pest control contracts, dental implants): time-decay or DDA.
  • 45–180 days, 6–20 touches (roofing, solar, commercial services, legal): position-based or DDA, plus self-reported attribution.
  • 180+ days, committee buying (B2B facility services, MSPs): no click model will work. Use self-reported attribution and quarterly geo holdouts.
  • The Model Nobody Puts in the Comparison Chart: Self-Reported Attribution

    Add one required field — "How did you hear about us?" — as an open text box, not a dropdown. Dropdowns bias answers toward the first option. Open text catches the answers that no tracking pixel will ever see: the neighbor's recommendation, the truck wrap, the Nextdoor thread, the ChatGPT answer that named you.

    In practice, self-reported and click-based data disagree by 30–50% on the top channel for local service businesses, and the gap has widened as more discovery moves into AI assistants and zero-click search. When a client's GA4 says 62% of leads are paid search but 41% of new customers write "someone recommended you," the referral engine is being underfunded — and no attribution model built on cookies will ever tell you that.

    Cost to implement: about two hours of form and CRM work. It is the highest-ROI attribution project available to a business under $2M in revenue, and it costs nothing per month.

    The most accurate attribution model most service businesses will ever run is a text field on a form and someone who actually reads the answers every Friday.

    When This Isn't Worth It — and Who Should Skip It Entirely

    This section costs us business, so read it carefully.

    Skip attribution modeling entirely if:

  • You spend under $3,000/month on marketing. Model choice will reallocate a few hundred dollars. You'd get more from fixing your speed-to-lead — answering inbound calls in under 5 minutes instead of 45 typically moves close rates 8–15 points, which dwarfs anything attribution will find.
  • You run a single channel. If 95% of spend is Google Ads, there is nothing to attribute *between*. Buy better keywords instead.
  • Your close rate isn't tracked in a CRM. Attribution on form fills instead of revenue optimizes for cheap leads. We've watched a $180 cost-per-lead channel beat a $310 channel on every dashboard while producing one-third the revenue, because nobody joined leads to closed jobs.
  • You're under 40 conversions/month. Statistical noise exceeds the effect you're measuring. A 20% swing in a 35-lead month is seven leads — well inside random variation.
  • Failure modes of the sophisticated options:

  • iOS and browser privacy have gutted click data. Roughly 30–40% of conversion paths are now partially invisible. Every model you run sits on incomplete input, and modeled conversions in ad platforms are estimates presented with false precision to the second decimal.
  • Multi-touch attribution correlates, it doesn't cause. MTA tells you which touchpoints appeared on winning paths, not which ones changed the outcome. Only a holdout or geo test answers that. Expect a $5,000–$15,000 budget and 4–8 weeks for a credible geo test — and expect a real revenue dip in the holdout region.
  • Media mix modeling needs 2–3 years of weekly data and typically $25,000–$100,000 to build. Below roughly $2M in annual media spend, the confidence intervals are wider than the decisions you'd make with them.
  • Attribution changes rarely change budgets. Switching from last-click to DDA moves channel credit by 10–25% in most accounts. If that wouldn't change what you fund, you've bought a report, not a decision.
  • The model you pick becomes political. Whoever runs the channel that DDA favors becomes its loudest defender. Pick the model *before* you look at the results, and write down the decision rule.
  • Agencies that sell "advanced attribution" as a $2,500/month line item on a $12,000/month ad budget are selling measurement you can't act on. Ask any vendor — including us — what specific budget decision the model will change and what it costs. If they can't answer in one sentence, don't buy it. Our pricing is public for that reason, and our ROI calculator will tell you in about 90 seconds whether your spend level justifies any of this.

    The Setup That Actually Works for a $10k–$75k/Month Advertiser

    1.

    Set GA4 to data-driven attribution, and set the lookback window to match your actual sales cycle — 30 days for emergency services, 90 days for considered purchases. GA4's default 90-day acquisition window is wrong for over half of service businesses.

    2.

    Enable enhanced conversions and server-side tagging. This typically recovers 10–20% of otherwise-lost conversions and costs $50–$200/month for a server-side container.

    3.

    Push closed-won revenue back into Google Ads and Meta via offline conversion import. This is the step that separates real measurement from lead-count theater, and it's a one-time 4–8 hour CRM integration.

    4.

    Add call tracking with dynamic number insertion. Roughly 40–60% of service business leads arrive by phone; if calls aren't tracked, you're modeling the minority of your pipeline.

    5.

    Add the self-reported field and review it monthly against your click data.

    6.

    Run one geo holdout per year on your largest channel — usually branded search — for 4 weeks. It's the only way to price the difference between attributed and incremental revenue.

    Steps 1, 4, and 5 cover about 80% of the value. Steps 2, 3, and 6 are where the remaining 20% lives, and they're worth it above roughly $30,000/month in spend.

    What Changed in 2025–2026 That Most Guides Haven't Caught Up To

    AI assistants now sit between the search and the click. When someone asks ChatGPT or Google's AI Mode for the best commercial HVAC contractor in Tampa and then types your company name into a browser, every attribution model on earth records that as direct traffic or branded search. That traffic is growing, and it's structurally unattributable by click-path logic.

    Two practical responses: track branded search volume as a *leading indicator* of upper-funnel work rather than a channel to optimize, and keep an llms.txt file plus clean, factual service pages so AI systems can cite you accurately. The self-reported field is currently the only instrument that catches "ChatGPT recommended you" — and it shows up in about 3–8% of open-text answers for service businesses now, up from near zero two years ago.

    The One-Sentence Answer

    Use GA4's data-driven attribution if you clear 600 monthly conversions, time-decay if you don't and your cycle is short, position-based if your cycle runs 45+ days — and in every case, weight self-reported attribution and one annual holdout test above whatever the dashboard says. The model matters less than joining your marketing data to closed revenue, which most businesses still haven't done.

    Frequently Asked Questions

    What is the most accurate attribution model?

    No attribution model is accurate in a strict sense; each is an assumption about credit. Data-driven attribution in GA4 comes closest for most advertisers because it uses observed conversion paths rather than a fixed rule. Pair it with self-reported attribution to catch offline and word-of-mouth sources analytics cannot see.

    Is last-click attribution still useful?

    Yes, as a diagnostic rather than a budget tool. Last-click reliably shows which touchpoint closed the session, which helps debug landing pages and conversion paths. It systematically overcredits branded search and retargeting, so using it to allocate spend pushes budget toward channels that harvest demand instead of creating it.

    When should I use a media mix model instead of attribution?

    Consider media mix modeling once blended ad spend passes roughly $50,000 per month. MMM uses aggregate spend and outcome data rather than user-level tracking, so it survives cookie loss and covers offline channels. Below that spend level, the modeling cost and data requirements exceed the decision quality it returns.

    How do I add self-reported attribution to my forms?

    Add an open-text or short dropdown "How did you hear about us?" field to every form, and script the same question for phone intake. Keep it optional to protect conversion rates, store the answer on the lead record in your CRM, and compare it monthly against GA4 channel data to spot gaps.

    Free Weekly Briefing

    One AI Marketing Tactic.
    Every Tuesday. Free.

    What's actually working across our client accounts right now — ROAS moves, follow-up sequences, creative angles. The stuff that isn't in any blog post yet.

    No spam. Unsubscribe anytime. 1,200+ business owners already in.

    Ready to Deploy

    SEE THIS IN
    YOUR BUSINESS.

    30 minutes. We scope the exact systems that apply to your situation and give you a plan.

    ★★★★★ Trusted by 47+ local service businesses

    BOOK A STRATEGY CALL →