Creative Gets Considered Before It Gets Ranked

I am a Decision Systems Advisor for founder-led organizations. That is the plainest way I have found to say what I actually do: I help founders build decision systems, so their teams get operational clarity and make better decisions without waiting on one person.

I did not arrive there through marketing. After 20+ years in healthcare leadership, I kept watching intelligent, committed people lose hours — sometimes weeks — not because they lacked skill or care, but because it was unclear who decided, who owned the next step, and where the real information lived. Creative strategy, AI adoption, workflow design, and grant work are not four identities I collect. They are four places where the same problem shows up: people cannot act well when the decision system is weak.

Advertising retrieval turns out to be a clean illustration of that, which is why this piece starts there.

Most conversations about advertising performance skip a step. They start at ranking — which ad won the auction, which creative got the impression — and ignore the stage that happens first.

Meta Engineering describes Andromeda as a retrieval layer: a machine-learning system that narrows an enormous pool of eligible ads down to a much smaller set of candidates. Later ranking models then choose the final ad from that set. Retrieval is a filter on possibility, not a verdict on quality.

Why the order matters

If your creative never enters the candidate set for a person and a moment, no amount of ranking strength can rescue it. The practical question shifts from “which ad performs best” to “how many genuinely different, relevant candidates does this system have from us in the first place?”

That is an argument for strategic variety, not volume. Retrieval rewards creative that is meaningfully distinct — different audiences, problems, objections, formats, and moments of use. It does not reward twenty near-identical variants with a shuffled color palette, which collapse into the same signal.

• •Vary the problem you name, not just the headline you write.
• •Vary the person you are speaking to, and say who that is.
• •Vary the form — demonstration, comparison, objection, story — so each asset teaches the system something new.
• •Retire assets that duplicate an existing angle rather than adding one.

The same shape appears in answer engines

Search with AI features works on a comparable sequence. Systems assemble a set of sources they consider relevant and trustworthy, then compose an answer from that set. Google Search Central's guidance for AI features is unglamorous on this point: there is no separate trick for AI surfaces. The same fundamentals — useful, people-first content, clear structure, and crawlable pages — determine whether you are eligible to be drawn on.

Being citable is mostly a clarity problem. A model can only summarize what it can parse without guessing.

• •State positioning plainly: who you help, what changes, and what you do not do.
• •Publish evidence — process, artifacts, and outcomes — rather than adjectives.
• •Keep facts consistent across every page and profile, so nothing has to be reconciled.
• •Write buyer-focused content that answers the question directly in the first paragraph.

What this does not mean

It does not mean creative volume is a strategy, and it does not mean the platforms have published a formula. Retrieval and ranking systems change, and any specific number attached to them ages badly. What holds is the sequence: consideration precedes selection.

"You cannot be chosen from a set you never entered."

So the work is upstream of optimization. Give the system distinct, honest, well-labeled candidates — in the feed and on your own pages — and let selection do its job.

The work behind the title

What are decision systems? They are the conditions that make a decision possible: who decides, what counts as enough information, when it happens, who owns the next step, and how the outcome gets recorded so nobody relitigates it. Most organizations have one — it is just undocumented and living in someone's head.

Why do founder-led organizations struggle with them? Because the founder is usually the system. Early on, that is an advantage: fast calls, full context, no overhead. Later it becomes the bottleneck. Every ambiguous decision routes back to the one person with the whole picture, and the team learns to wait instead of act.

Why am I qualified to work on this? Two decades of healthcare leadership, where unclear ownership has consequences you cannot spin, plus years building AI tools, creative systems, and funded programs. Different rooms, same failure pattern. I have had to fix it under real constraints, not on a whiteboard.

What changes after working together? Fewer decisions bottleneck with the founder. The recurring ones get named, owned, and documented. Teams stop guessing at intent, and the work that used to stall moves without a meeting.

I want to be honest about the limits of this. It is not about having all the answers or arriving with a system to impose. Most of what I do is help leaders see what has become invisible to them — the standing decisions they no longer notice making — and then build clearer conditions for other people to act.

That is the same argument as retrieval, applied to a company instead of an ad account. Selection is downstream. If you want better decisions, widen and clarify what your organization is able to consider — and if you are a founder looking for operational clarity, that is the work I do.

Sources

• •Meta Engineering — Andromeda: https://engineering.fb.com/2025/03/12/production-engineering/how-meta-is-advancing-ai-ad-recommendations-with-andromeda/
• •Google Search Central — Top ways to ensure your content performs well in AI experiences: https://developers.google.com/search/docs/appearance/ai-features