Demand Gen spends across four completely different places and hands you one number for all of them. These fifteen skills read that number before you act on it.
A Demand Gen campaign reports ninety conversions for the month. The shop backend records thirty-one orders in the same period. Nobody is lying and nothing is broken.
Join the waitlist and get notified when the new AdLume is ready. It catches wasted ad spend automatically and optimises campaigns around the clock.
One email, on the day access opens. Nothing else, and you can leave the list with one click.
Three things are happening at once, and the campaign shows you the sum of them.
The campaign counts three kinds of credit. A click-through conversion is somebody who clicked the ad and bought. An engaged-view conversion is somebody who watched roughly ten seconds, never clicked, and bought within a few days: on this campaign type that credit sits inside the headline conversion figure, which is why the count can look large while your traffic does not move. A view-through conversion is somebody who saw the ad, did not engage with it, and converted later, usually reported in its own column. Subtracting the wrong pair of columns produces a click-through figure that is not click-through at all.
The campaign spends across several channels at once. YouTube, Shorts, Discover, Gmail and Display behave nothing like each other, and by default the campaign reports them together. You can pull the channel-level report and you can turn channels off in the settings, so this is a solvable question rather than a permanent fog — but you have to go and look, because the default view will not show it to you.
One trap worth naming before you touch that setting: a campaign whose only assets are vertical videos will lean to short-form by construction. That looks like a channel problem and is a creative one, and switching the channel off shrinks delivery without fixing anything. Check what shapes you own before you change where the money can go.
The numbers keep arriving after the day ends. A result judged on the Monday after a week closes can be judged on less than half of what that week eventually recorded.
None of that is a reason to avoid the campaign type. It is a reason to read the report properly before moving budget on it.
| # | Skill | What it answers |
|---|---|---|
| 1 | Where Your Money Actually Went | Which surface took the budget, and whether it earned it |
| 2 | Are These Views Real Attention | Which view counter belongs in the report and which is decoration |
| 3 | Did They Buy Or Did They Just Watch | How much of the conversion count is exposure credit rather than action |
| 4 | What Your Creative Is Missing | Which absent asset ratio is quietly narrowing delivery |
| 5 | Which Video Actually Did The Work | Which asset carried the campaign, and what to brief next |
| 6 | Is This Bringing New Customers | Whether upper-funnel spend found people who had never bought |
| 7 | Who Google Thinks You Want | Whether your audience signal is steering delivery or decorative |
| 8 | Are You Paying Twice For The Same Person | Where campaign types are converting the same people |
| 9 | Is Your Feed Ready For Demand Gen | Whether the catalogue can actually serve |
| 10 | Why Your Frequency Is Climbing | Saturation, auction pressure or creative fatigue |
| 11 | Should This Bid To Clicks Or Conversions | Whether the bid strategy matches the volume you have |
| 12 | When The Numbers Stop Moving | How many days before a result is safe to judge |
| 13 | Would These Sales Happen Anyway | Whether the spend causes outcomes or reports them |
| 14 | Where The Click Lands | Whether the page fits traffic that never searched for you |
| 15 | The Weekly Demand Gen Readout | Facts, hypotheses and decisions, kept separate |
Every skill is a single Markdown file with the same seven parts: when to use it, the inputs it needs at minimum and the ones that help, what to check before analysing anything, the analysis workflow, the decision rules, the output format, and its guardrails.
Two things are deliberate and worth naming.
The full library of AdLume skills: paid ads, landing pages, B2B leads, GTM tracking, e-commerce.
The bundle goes straight to your inbox, then one email a week. Unsubscribe in one click. We never share your address.
Every threshold carries a unit and a label. Where a number appears, it says what it counts and that it is a starting point to recalibrate against your own account. Most of the figures people quote about this campaign type began life in somebody's blog post and are repeated until they sound like documentation. The volume you need before bidding to a target, for instance, is usually quoted per thirty days rather than per week, and the difference between those two readings decides whether your campaign is allowed to optimise at all.
Each skill tells you which report to open. On this campaign type, finding the right export is most of the work, and the columns that carry the answer are frequently not in any default view. A skill that assumes you already have the data is a skill you never run.
Nothing here logs into your account, changes a bid, pauses a campaign or edits a feed. Each skill ends with a decision handed back to you.
Four of them go further and refuse to answer under stated conditions. The lift test tells you when your account's week-to-week swing is larger than any effect a holdout could detect, instead of designing a test that returns a number nobody can trust. The feed check fails an entire feed on a single wrong price, because that defect damages trust with a customer rather than merely reducing reach. The lag reader refuses outright on a single export, because one pull cannot show you how a figure filled. The audience check stops dead when the account will not report performance per segment, rather than reading the segment list and calling that an answer.
That is the part worth stealing even if you never install any of this. A check that always returns something is not a check. Most reporting failures in this campaign type are not wrong numbers; they are real numbers answering a question nobody asked, and the only defence is a tool willing to say it cannot see.
A third does something related. When your account cannot report the surface split at all, the money skill neither stops nor guesses: it falls back to the read your account can support and tells you which of the two it ran.
The pack is on GitHub under an MIT licence. Clone it and copy the skill folders into your Claude skills directory, either inside a project or at user level.
git clone https://github.com/mardab96/demand-gen-youtube-skills.git
mkdir -p ~/.claude/skills
cp -r demand-gen-youtube-skills/*-demand-gen demand-gen-youtube-skills/scripts ~/.claude/skills/Start a new session in that project and the skills activate when what you are asking about matches one of them. If you use Claude somewhere other than Claude Code, the files are plain Markdown and you can paste the relevant one into the conversation.
Two skills ship a small Python helper for arithmetic that should not be estimated: the fill curve that produces your waiting rule, and the baseline spread that decides whether a holdout is readable at all. The install command above copies them.
Open Did They Buy Or Did They Just Watch first. It needs the least setup, the conversion columns and your own order count for a closed period, and it most often changes a decision that was about to be made. Its own escalation rule is the honest one: a high exposure share alongside flat revenue is not proof the campaign is worthless, it is the exact condition a lift test exists for.
AdLume watches your ad accounts and brings the next fix to your team's tools.
One email, on the day access opens. Nothing else, and you can leave the list with one click.
No. These skills read exports. Nothing in the pack requires account access, and nothing in it can change a campaign.
Parts of it will, but the pack is written for Demand Gen specifically, and several skills turn on reporting behaviour that differs by campaign type. Read the required inputs before assuming a skill transfers.
No, and the pack says so next to every one of them. They are starting points from real accounts, meant to be recalibrated. Where a figure is widely quoted but shaky, the skill says which part is uncertain.
Several skills handle exactly that case explicitly: one falls back to a narrower read and says so, two decline to produce a verdict at all. A skill that invents a number from data you do not have is worse than no skill.