TL;DR
Seven attribution models compared by cost to run, not theory. Why first-touch plus last-touch beats data-driven attribution until you clear 500 conversions a
→ See how this applies to your business (free 30-min call)Most service businesses should run two attribution models side by side: first-touch and last-touch — and stop there until they're spending north of $25,000/month on media. First-touch tells you which channel *created* demand; last-touch tells you which channel *closed* it. The gap between them is where your budget decisions actually live. Linear, time-decay, position-based (U-shaped), and W-shaped models are refinements that matter mostly above ~500 conversions/month, and full data-driven attribution (Google's DDA, or a Markov-chain model on your own data) needs roughly 600 conversions and 15,000 ad interactions in 30 days per campaign before Google will even build the model. Below that volume, a "better" model produces worse decisions than a spreadsheet.
The Seven Models, Ranked by What They Actually Cost You to Run
Here's the honest ordering for a US home-services, legal, medical, or B2B services company doing $500K–$20M in revenue.
The model you choose changes reported channel ROI by 20–60% without a single dollar of spend changing. That's not a rounding error — it's the entire basis for next quarter's budget.
The Comparison Table Nobody Publishes
Run the same 90 days of data through different models and the same channel looks like a hero or a liability. A realistic pattern for a $40K/month multi-location HVAC advertiser:
| Channel | Last-touch CPL | First-touch CPL | Position-based CPL |
|---|---|---|---|
| Branded search | $18 | $71 | $34 |
| Non-brand search | $96 | $88 | $91 |
| Paid social | $210 | $109 | $143 |
| Local Services Ads | $61 | $74 | $66 |
| Organic/GBP | $0 | $0 | $0 |
Paid social goes from "kill it" ($210 CPL) to "keep it" ($109) depending purely on the model. Branded search goes from your best channel to your fourth. Neither number is wrong. Both are incomplete. The correct move is to run a geo holdout or spend-down test on paid social for 4 weeks and see what actually happens to total lead volume — attribution modeling is a hypothesis generator, not a verdict.
The Section That Costs Us the Sale: When Attribution Modeling Is a Waste of Money
We sell attribution work. We also turn it down regularly, because for a large share of service businesses it's a negative-ROI project. Read this part twice.
Don't buy multi-touch attribution if:
Failure modes even when you do buy it:
The honest replacement for most sub-$25K/month advertisers: incrementality testing. Turn a channel off in 3 matched DMAs for 4 weeks, keep it on in 3 others, compare total revenue. Costs nothing but nerve, and answers the only question that matters — "does this channel produce sales I wouldn't have gotten anyway?" Our ROI calculator will show you the spend threshold where modeling starts to pay.
What to Do at Each Spend Tier
The Thing Nobody Tells You: Attribution and Bidding Are the Same Decision
Here's the point most attribution articles miss entirely. Your attribution model doesn't just change reports — it changes what your ad platforms optimize toward, because Smart Bidding trains on the conversions you feed it. If you import last-touch-credited conversions into Google Ads, the algorithm learns to buy last-touch-ish inventory: branded terms, high-intent remarketing, cheap clicks that were already going to convert. You get a beautiful ROAS number and a shrinking business.
Two practical consequences:
If you want to see how this plays out for specific verticals, our case studies show the before/after on offline conversion imports, and our services page covers where attribution fits in a full build.
A 30-Day Starting Plan
Days 1–5: Audit tracking. Confirm GA4 events fire, DNI is live on every page, and every campaign uses one UTM convention. Expect to find 2–5 broken things.
Days 6–10: Add a required "how did you hear about us?" field to intake and to your call script. Self-reported attribution is noisy and biased, but it catches the untracked 30% that no model sees. Compare it monthly against modeled data.
Days 11–20: Export 90 days of path data. Build last-touch and first-touch views side by side. List every channel where the two disagree by more than 30%.
Days 21–30: Pick the single biggest disagreement and design one test — a geo holdout, a 50% spend-down, or a 2-week pause. Run it for at least 4 weeks. Whatever the test says beats whatever the model says.
The best attribution model is the cheapest one that changes a decision you were going to make anyway. For most US service businesses under $25K/month in media, that's first-touch versus last-touch plus one honest holdout test per quarter — total cost under $500/month, and it beats a $30,000 algorithmic build that nobody acts on.
Frequently Asked Questions
What is the best attribution model for a small business?
Run first-touch and last-touch side by side. First-touch shows which channel created demand, last-touch shows which closed it, and the gap between them drives budget decisions. More complex models only pay off above roughly 500 conversions per month or $25,000 in monthly media spend.
How many conversions do you need for data-driven attribution?
Google's data-driven attribution requires approximately 600 conversions and 15,000 ad interactions within a 30-day window per campaign before the model is built. Below that threshold, the algorithm lacks enough path data to assign credit reliably, and its output is less trustworthy than a simple spreadsheet comparison.
What is the difference between first-touch and last-touch attribution?
First-touch assigns 100% of conversion credit to the channel that started the customer journey, revealing demand creation. Last-touch assigns 100% to the final interaction before conversion, revealing what closed the sale. Running both exposes channels that generate awareness but rarely appear in last-click reports.
Is multi-touch attribution worth it for service businesses?
Usually not at low volume. Linear, time-decay, position-based, and W-shaped models distribute credit across touchpoints, but they need enough conversion paths to produce stable patterns — roughly 500 conversions monthly. Below that, small sample noise makes a sophisticated model produce worse budget decisions than two simple ones.
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.