We ran a session on the channels your attribution can't see: the podcasts, influencers, paid social, offline and word-of-mouth work that mostly lands in your reports as "direct" or gets quietly handed to search. This is a recap of what we covered.

We started by asking people what share of their conversions shows up as direct. The median guess was 30%, one in three put it at 50% or more, and the highest anyone reported was 80%. We're in the middle of a project for a client using one of the well-known enterprise attribution tools, six figures a year, and they were seeing 74% attributed to direct.

![Article image 1](/images/blog/how-to-defend-the-channels-attribution-can-t-see-workshop-recap/image-001.png)

#### A high direct share is a channel-mix problem, not a tracking problem

The first instinct, when direct is high, is to blame tracking. Fix the UTMs, argue about the consent banner, tune the user-stitching tool. Tracking usually plays some part, but it's rarely the real cause.

Direct just means we don't know. Nobody dreams about your company, wakes up, types your URL and converts. A direct session means they heard about you somewhere you couldn't see, and you can't stitch that back to the source. If a lot of your spend goes to channels whose journeys don't end on your website, no naming convention will close that gap.

#### You can't treat the part you can see as a sample of the whole

The dangerous reading is: attribution is never perfect, I can attribute 26%, so I'll work with that 26%. That treats the tracked share as a representative sample and assumes the rest looks the same.

It doesn't. The share hidden in direct is disproportionately the channels whose journeys break click-based tracking: paid social, podcasts, offline, word of mouth. So you can't take your 50% direct, double the tracked half and call it done. Measurability wins the budget argument by default, and that's a budget problem, not a reporting one.

> Direct is not a channel. You can't make decisions with the 26% alone.

#### The hardest channels to defend are the ones that only _look_ measurable

Nobody opens GA4 expecting to see their conference booth or their out-of-home. Those get judged on strategy, not analytics, and everyone's fine with that.

The trap is the channel that feels trackable but isn't. Paid social is the clearest example. Someone clicks an ad, the page opens in the in-app browser, then they reopen it in Safari to get out of Instagram, and that's a new user in a new browser. The click journey breaks. Finance sees a paid-social CPA five times paid search and asks why you're spending there at all. AI search is heading into the same category: ChatGPT has been appending something to the URL, but you only catch the click, never the impression that built the trust earlier.

![Article image 2](/images/blog/how-to-defend-the-channels-attribution-can-t-see-workshop-recap/image-002.png)

#### The five methods we use to recover the signal

None of these is a silver bullet, and most teams need a few of them together.

1. **Control the links you own.** UTM discipline, scheduling tools and leak monitoring. Worth doing, but it proves nothing on its own and won't move your direct share by itself. Most clients still tag by hand; automate it with platform and email macros and it holds up far better. Everything below depends on it. Here's [how to automate your UTMs with dynamic parameters](https://www.021newsletter.com/p/dynamic-url-utm-parameters-what-are).

1. **Survey data.** The easiest big lever. A post-purchase or lead-form "how did you hear about us?", used as a fallback when we can't map a touchpoint, and placement matters more than wording. One client went from 50% direct to 20% this way. If you can't touch onboarding yet, test it with a simple email to new users first, that's how I once found influencers driving 20% of signups on a dating app while the MMP showed almost nothing.

1. **Landing pages, vanity URLs and QR codes.** A dedicated destination per placement (`meta.` for Meta, a partner page per influencer), countable by design, so you don't depend on UTMs. For one client, vanity URLs and promo codes were the only signal podcast and direct mail had. What surprised us is how much direct still lands on these pages, which itself becomes a clean signal you can reclassify.

1. **Identity stitching.** Capture the full user agent to detect in-app browsers (Instagram, Messenger), then probabilistically match those against the direct sessions arriving in the same window. It recovers the conversions Meta and Instagram were never credited with, typically 10-20% you're confident about. Consent handling can break stitching before it starts, and we saw exactly that in an app audit.

1. **Incrementality and timing uplift.** When a TV or radio agency shares where and when impressions ran, look for the direct and brand-search sessions in that city and window. This is calibration, not reclassification: it answers whether the conversions would have happened anyway. Good enough to report the channel and defend it in a cut decision, not solid enough to attribute to an individual account, so we flag journeys as "potentially influenced" rather than treating it as a hard signal.

![Article image 3](/images/blog/how-to-defend-the-channels-attribution-can-t-see-workshop-recap/image-003.png)

#### Not everything gets recovered, and that's the honest part

No method reclassifies all of direct, and anything promising that is a black box. Some channels stay under-reported even with all of this. A podcast might lean 60-70% on the survey and the rest on landing pages, and there's still a dark spot of listeners you'll never see. That's fine. Some data and an honest caveat makes the budget conversation far easier than walking in with three tracked podcast conversions for the month. Our rule of thumb: direct plus referral above 10% of conversions means there's recovery work to do.

#### Not every marketing channel is an ad platform

One client runs an open-source product. People read the docs or go straight to GitHub without registering, so most of that never touched attribution. By reclassifying visits to their open-source pages, we could see open source appeared in around 20% of journeys, work that was completely invisible before, and hard to defend without a number on it.

#### Why this cannot live in your analytics tool

Channel grouping in GA4 is fixed and direct is the fallback by design, so you can't join surveys, user agents and offline data to sessions inside the tool. Recovering these signals means modelling all of that into one customer view in a warehouse, where marketing can actually read and edit the rules. A warehouse only earns its place once you have a genuinely hard-to-measure mix or a complex, delayed revenue model. If you're earlier than that, run one-off incrementality tests to check whether a channel works, and keep your data collection consistent so you can build on it later. The moment you add a new channel to a search-heavy mix is the right time to rethink measurement, before a paid-social test "loses" to search on a model that was only ever built for search.

#### What it looks like finished

We took one client, Zeffy, from 51% to 8% direct and referral share, recovered about 2% of traffic as mobile social through stitching, and lifted survey coverage from 80% to 90%, enough that they can now test programmatic display and evaluate channels beyond search with confidence. The [Zeffy case study](https://partnerwithpropel.com/case-studies/zeffy) has the full numbers. Timo also wrote up [what actually creates direct traffic](https://timo.space/blog/direct-traffic-engineering/), the decision tree behind it, and why the work belongs in the warehouse.

If you want to talk through what this would take for your own channel mix, book a call with us.

Propel builds marketing data foundations and warehouse-based attribution for teams whose spend runs across channels that click-based tracking can't see, combining survey, incrementality and journey data to make hard-to-measure channels defensible.