Attribution Models Explained for Operators, Not Analysts
A founder emailed me last month with a spreadsheet full of UTM data and one question: "Which attribution model should I use?" She was spending $18K a month across Meta, Google, and a podcast sponsorship, and nobody could tell her what was actually working. This is the most common question I get from operators running their own budgets, and it deserves a straight answer, not a framework diagram.
Here's the answer up front. For a marketing attribution model for small business use, start with last-click plus one honest gut check: a post-purchase survey asking "how did you hear about us." That combination will outperform any multi-touch model you build with under $500K in annual spend and no dedicated analyst. I'll explain why, and I'll tell you when to graduate off it.
What the problem actually looks like
You look at Google Analytics and Meta Ads Manager and they both claim credit for the same sale. Meta says it drove 340 conversions last month. Google Ads says 290. Your total order count was 410. The math doesn't work because every platform is grading its own homework.
Then someone suggests a multi-touch attribution tool. You sign up, connect it, and get a dashboard with linear, time-decay, and U-shaped models sitting side by side, each giving different numbers. Now you have three wrong answers instead of two.
Meanwhile the actual decision you need to make is boring: do I put next month's $5K into Meta or into the podcast? The dashboard doesn't answer that. It just makes the uncertainty look more sophisticated.
Why it happens
Every ad platform uses a self-serving attribution window. Meta will claim a sale if someone saw an ad and bought within 7 days, even if they never clicked. Google Ads does something similar with view-through conversions. Both platforms are incentivized to inflate their own contribution because that's how you justify more budget.
Multi-touch attribution models try to fix this by spreading credit across every touchpoint in a customer's journey. That sounds rigorous. In practice, it requires clean tracking across every channel, a data warehouse to unify it, and enough volume that the model has statistical signal. A business doing $2M in revenue with 4,000 orders a year doesn't have that volume. The model ends up interpolating noise and presenting it as insight.
I saw this exact failure mode on a $2.2B infrastructure project I marketed. We had six agencies each measuring their own channel's performance with their own model. PR claimed credit for stakeholder sentiment shifts. Digital claimed credit for the same shift because of a coordinated push. Nobody's number was wrong exactly, but adding them together overstated total impact by roughly 40%. Small businesses hit the same trap at a smaller scale, they just don't have six agencies to notice the overlap.
What we do about it
Here's the actual procedure I use with small business clients and used on my own DTC brand when it scaled from $100K to $3M.
- Put a "how did you hear about us" field on the post-purchase or signup flow. Make it required, make it a dropdown with 6-8 options, and always include "other" with a text box. This is the single highest-leverage attribution tool available to a business under $5M in revenue. It costs nothing and takes one afternoon to implement.
- Use last-click in your ad platforms for tactical, day-to-day decisions. Which ad creative is working this week, which keyword to pause. Last-click is directionally fine for short-cycle optimization even though it's wrong for total credit.
- Reconcile monthly against your survey data. If Meta says it drove 40% of conversions but your survey says 15% of customers cited social media, trust the survey more. It's asking the human directly instead of inferring from a pixel.
- Run one incrementality test per quarter. Turn off a channel for two weeks in a low-risk period and watch total revenue. If shutting off Google Ads for two weeks drops revenue by 3%, but Google Ads was "claiming" 25% of conversions, you've just found your inflation factor.
- Only build multi-touch modeling once you clear roughly 10,000 transactions a year across at least three consistent channels. Below that, the model has too little data to be more accurate than the survey-plus-last-click combination.
The real cost here is time, not tooling. The survey field is free. The incrementality test costs you two weeks of a channel's spend, which for a $10K/month budget might mean $2,500 of "wasted" test spend. That's cheap tuition for knowing whether a channel actually works.
What it costs to ignore
I've watched founders keep a channel alive for a year based on platform-reported ROAS that was pure fiction. One DTC account I advised kept a $6K/month influencer retainer running because Instagram attribution showed it "driving" 22% of sales. When we finally ran the survey question for 90 days, only 4% of customers mentioned that influencer by name. That's roughly $60K a year spent on a channel that was working at a fraction of its claimed rate.
The bigger cost is decision paralysis. Owners who don't trust any of their numbers stop making budget calls with conviction. They split spend evenly across channels "to be safe," which is the same as making no decision at all. Evenly split budgets rarely outperform concentrated bets on what's actually working.
Most common mistake
The most common mistake is chasing a more complex model before fixing bad tracking. Businesses buy a $500/month multi-touch attribution tool while their UTM tagging is inconsistent and half their email links have no parameters at all. A sophisticated model built on dirty inputs is worse than a simple model built on clean ones. Fix your tracking hygiene first. Model complexity second.
FAQ
What's the best attribution model for a business under $1M in revenue?
Last-click for platform optimization, paired with a mandatory "how did you hear about us" field at checkout or signup. Don't pay for multi-touch software at this stage. It won't have enough data to be reliable.
When should I upgrade to multi-touch attribution?
When you're running at least three paid channels consistently, generating 10,000+ transactions a year, and you have someone who can own the data pipeline. Below that threshold, the added complexity produces false precision, not better decisions.
The practical takeaway: pick the simplest attribution setup that changes what you actually do next month. If a dashboard doesn't move a budget decision, it's decoration. Start with the survey field this week. It's the cheapest, fastest, most honest data you'll ever collect.