Channel attribution is now adversarial to good strategy.

The last-touch report is lying to you. Not because the number is wrong, but because the unit is wrong. Buyers don't move through channels anymore. They move through a run of small engagements with your business, and with the category you sit in, across a stack of surfaces: an LLM prompt here, a Reddit thread there, a LinkedIn post, a G2 skim, a podcast in the background, then a paid search click at the end. Some of those surfaces you own. Some of them you don't. Some of them you didn't know were part of the picture.

Taken together, that stack of surfaces is the discovery space. Sharper definition further down. Here's the take first: channel attribution is now actively adversarial to good strategy. It hands the credit to whichever channel got the final click, and gives nothing to the surfaces that actually shaped the buyer's decision. If your reporting still treats "paid search" and "organic" and "LinkedIn" as clean, separable buckets, you are optimising the wrong unit, and the more you tune inside that unit, the further you drift from what's actually moving the number.

The moment the reporting stopped making sense

Here's the shape of it. A B2B mid-market business, somewhere in the $50 to $150M ARR band. Steady growth for years, a channel mix the team could recite in their sleep. Then something shifted. The data stopped telling the story per channel it used to. The top channel changed for no reason anyone could name. Deals took longer to close and then faster to close, sometimes in the same week. Paid search still "converted" the buyer, but the conversation on the demo call made it clear the buyer had already made up their mind before the click.

They'd been on Reddit. They'd asked ChatGPT for a shortlist. They'd read two competitor comparisons, watched a YouTube walkthrough, and skimmed a founder's LinkedIn post that mentioned the category. Then they searched the brand name and clicked the paid ad because it was at the top.

The paid team got the conversion in the report. They also got the budget for next quarter. The Reddit thread, the ChatGPT answer, the YouTube walkthrough, the founder's LinkedIn post: none of those got credit, because none of them fit the model.

That gap isn't hypothetical. 6sense's 2025 Buyer Experience Report found 84% of B2B deals are already decided before marketers even know a buyer exists, and the vendor ranked first on the buyer's pre-engagement shortlist wins around 80% of the time. Forrester's 2026 Buyers' Journey Survey named GenAI chatbots the single most influential source B2B buyers use to build those shortlists, ahead of review sites and analyst reports. The decision is being shaped somewhere your channel dashboard cannot see.

That's not a measurement problem you can fix by adding another touch to the model. It's a category error.

What a discovery space actually is

A discovery space is the set of surfaces a specific buyer triangulates across in a single research arc before they'll take a meeting. For a RevOps buyer that's probably an LLM prompt, a Reddit thread, two LinkedIn posts, a G2 skim, one competitor site, and yours. For a founder buying accounting software it's a completely different set. For a procurement lead evaluating a vendor it's a third.

The space isn't a channel. It isn't the sum of channels either. It's the specific combination of platforms, content, other people's opinions, and small actions the buyer runs through, in a rough order that varies per buyer, over a time window that used to be six weeks and is now anything from six minutes to six months.

The path is longer and shorter at the same time. Longer because there are more surfaces to check, and buying groups are bigger: 6sense's 2025 data has average B2B buying groups at 11 people. Shorter because AI compresses research. 6sense also clocked average deal cycles falling from 11.3 to 10.1 months between 2024 and 2025, with first vendor contact happening six to seven weeks earlier in the cycle than it used to. Both true. Both breaking the funnel model you've been reporting on.

Why the old model turned adversarial

Last-touch was always a bit of a fiction, but for a long time it was a useful one. When channels had distinct audiences and distinct intent signals, giving credit to the last one was rough justice, not fraud. It over-rewarded search a bit, under-rewarded brand a bit, and everyone got on with their job.

Multi-touch tried to fix it and mostly failed, because it kept the channel as the unit. It just spread the credit around inside the same broken frame. You ended up with attribution theatre: a dashboard that looked more sophisticated and answered the same wrong question with more decimal places.

Here's why it's now adversarial and not just imprecise. The surfaces doing the heaviest lifting in the discovery space are the ones your reporting can't see, or can barely see. LLM answers typically leave no reliable UTM. Workshop Digital's analysis of 181 million GA4 sessions found roughly 22% of ChatGPT sessions and 32% of Perplexity sessions landing in "(not set)". ChatGPT only started appending its own UTM on desktop in mid-2025, and mobile app traffic still breaks the referrer. Reddit threads don't send referral traffic if the buyer copies the URL, which Semrush's 2025 citation research suggests is happening a lot: Reddit accounts for around 40% of all sources LLMs pull from when answering commercial questions. A founder's LinkedIn post that shaped the buyer's category understanding gets zero credit if the click came later from Google.

So the team that owns those surfaces looks like it's losing, and the team that owns the click at the end of the space looks like it's winning. Budget follows the report. Budget moves toward the click. The surfaces doing the corroboration get starved. The space thins out. Conversion rates on the winning channel start to drop, because the corroboration that used to feed it is no longer there. And nobody can work out why.

That's the loop. It's not neutral. It's actively pulling budget and headcount away from the work that's moving the number.

How to map your own space

You can do this without buying anything. It's uncomfortable, not expensive.

Pick one recent closed deal. Not a whale, not an anomaly, a normal one. Get on a call with the buyer. Ask them, in order, every place they encountered your name, your category, or the problem you solve, in the four weeks before they took the meeting. Don't interrupt. Don't try to map it to your channels while they're talking.

Do it three more times. Different buyer profiles, same question. You'll start to see the shape of your space. It won't match your channel mix. That's the point.

Scale it with email. Phone calls give you depth from a handful of buyers. Email gets you breadth from the ones who won't get on a call but will answer two questions. Send a structured, short survey to the last thirty or so closed-won and closed-lost accounts. Two questions is enough: "Before you spoke to us, where did you first come across us or the category?" and "What else were you reading or watching in the weeks before you decided?" Free text, no dropdowns. Dropdowns pre-decide the answer for them and you lose the surface you didn't know existed.

Write down the surfaces that came up more than once. Those are the load-bearing surfaces of your discovery space. Some of them you own. Some of them you don't. Some of them you didn't know existed.

Overlay your reporting. For each surface, ask: does anything in our current stack tell us whether this is working or not? For most of the surfaces on the list, the honest answer is no.

That gap is the strategy work. Not filling every gap with a tracking pixel, which is where most teams go next and where most teams then get stuck. The work is deciding which surfaces matter enough to invest in, and building the observability that lets you tell whether the investment is doing anything.

What to actually do

Once you can see the space, four things follow.

Build combined views. Stop reporting on channels in isolation. Start reporting on the space. That means stitching data sets that don't naturally join: GA4, CRM, LinkedIn engagement, LLM referral logs where you can get them, brand search volume, community mentions. Ugly, slow, worth it.

Monitor trends, not conversions. In a discovery-space model, week-on-week channel conversion is noise. The signal is in the shape of the space over months: which surfaces are growing, which are drying up, which correlate with pipeline moves six weeks later. That's the number worth watching.

Make correlation bets and then prove or kill them. If podcast mentions in month one correlate with paid search conversions in month three, that's a hypothesis, not a fact. Treat it as a working bet. Design the smallest test that would break it. Run the test. Keep the ones that survive. Kill the ones that don't. This is where most teams stall out, because "correlation isn't causation" gets used as an excuse to not decide. Decide anyway. Then test.

Create, monitor, test, iterate. Ship into the space, watch the shape change, learn something, ship again. The cadence is faster than the old channel-plan cycle. Fortnightly reviews at minimum. Quarterly plans are too slow for how buyers are researching now.

Closing thought

The uncomfortable version of all this: if a buyer triangulates three sources before they reach you, and those three sources are telling three different stories about who you are and what you do, you have a strategy problem, not a marketing problem. Channel-level ownership makes that incoherence almost guaranteed. One owner per channel, no owner for the space, and a reporting model that pays the wrong person for the wrong reason.

Find the signal. Move the number. Nothing else about the job has changed.