Someone on your board mentions multi-touch attribution, and you spend an afternoon reading about first-touch, last-touch, linear, and time-decay models before realizing you have twelve customers and no data team to build any of them. The models assume a volume of touchpoints and a tracking setup that a three-person company doesn't have and doesn't need yet. The question you actually need answered is smaller than the one the models are built for.
Attribution at seed is a question about which sources you'll check.
Before you can debate models, you need a habit: someone looking, every week, at where new customers said they came from. That sounds too simple to be attribution, but it is the part that actually determines whether you learn anything. A company with a sophisticated model that nobody opens learns less than a company with last-touch numbers that a founder reads every Monday morning, and the second company usually makes better decisions faster.
The real failure mode at this stage is not picking the wrong model. It is building a dashboard and never returning to it, because building it felt like the work and reading it felt optional.
Last-touch is wrong, and it's good enough to start.
Last-touch attribution credits whichever channel a customer interacted with right before converting, which is obviously incomplete. It ignores the podcast they heard six weeks earlier, the friend who mentioned you at dinner, the three blog posts they read before ever filling out a form. Every marketer who has studied attribution knows this, and it is still the right place to start, because it requires no infrastructure beyond a form field asking how someone heard about you and a spreadsheet to tally the answers.
The goal at seed is not an accurate model. It is a directionally useful one that costs an afternoon to set up and a few minutes a week to maintain. You are trying to learn whether to spend more on a channel or less, not trying to publish a research paper on your own funnel.
A better model is a later problem, not a now problem.
The temptation to build something more rigorous, UTM parameters on every link, a proper analytics stack, a weighting scheme across touchpoints, is real, and it is worth resisting until the volume of customers makes the extra precision matter. With ten or twenty customers a month, the noise in any multi-touch model swamps whatever accuracy it adds over last-touch. You would be optimizing a measurement you do not yet have enough data to trust.
The right moment to upgrade is when the questions last-touch cannot answer start costing you real decisions, when two channels are close enough in last-touch numbers that you genuinely cannot tell which one to fund next quarter, and the answer would change what you do. Reading the same accountable number alongside your attribution habit each week is usually what surfaces that moment.
Pick two or three sources, and check them every week.
In practice this means a "how did you hear about us" field on the form, your ad platforms' own conversion numbers, and maybe one more source specific to your channels, a podcast host's promo code, a partner's referral link. Three sources, checked weekly, read together rather than in isolation, will tell a founder more than a model nobody has time to maintain.
Write down what you would expect to see if a channel were working, and check it against what you actually see. That comparison, done consistently, is the part of attribution that changes decisions. The model is the part that changes slide decks, and slide decks rarely change what a founder does on a Tuesday morning.
Nothing about this habit requires a tool purchase or a hire. It requires a recurring fifteen minutes on someone's calendar and the discipline to keep the source list short enough that checking it stays a habit instead of a chore. Startups that let the list grow to eight or ten sources usually stop checking it within a month, which defeats the entire point of picking sources you can actually see every week.
We set up exactly this, the smallest attribution habit that gets read every week, before we ever recommend anything more elaborate. Tell us what you're building.