Keith Buenaventura

← Back to work

Idea to launch,
simplified.

This is the roadmap I actually run when a brand hands me a paid social brief: the nitty-gritty, minus the jargon, pulled straight from my own internal creative process doc.

01

Research

  • Audit competitors and top-trending content in the same niche
  • Scan the Meta Ads Library for what's currently running and for how long
  • Flag new formats, tools or platform features worth testing
02

Idea

  • Dig through the DAM for what already exists and what's fatigued
  • Decide: brief an in-house shoot, or source a UGC creator?
  • Pick a perspective: Friend, Them, or Unique/Experimental POV
03

Plan

  • Lock the offer: what's actually being sold?
  • Map it to the funnel: ToFu, MoFu, BoFu
  • Pick a template: Hook → Proof → Offer → CTA
04

Build

  • Shoot, source, or repurpose, then edit against a QA checklist
  • Build or update the landing page to match the hook
  • QA the checkout, since a great ad still dies at a broken cart
05

Launch

  • Ship format variations: Reels, Stories, static, carousel
  • Test hook quality and volume strategy against each other
  • Read CTR, CVR and hook-hold rate weekly
This isn't a straight line. It's a loop. Week 5's results become week 6's Research. Nothing here ends at Launch.

Why Research matters more since Andromeda

Meta rebuilt its ad-retrieval engine in 2024–2025. It changed which lever actually controls who sees your ad.

Before: Manual Targeting

You define the audience: interests, demographics, behaviors

A limited, rule-based pool of ads gets matched to that segment

Winning meant micro-optimizing targeting and bids

Now: Meta Andromeda

AI evaluates tens of millions of ad candidates per person, in real time

Broad targeting plus creative diversity wins, since the algorithm decides who sees which ad

Near-duplicate creative gets grouped and suppressed as "clutter"

The practical result: creative diversity is now the targeting lever. That's exactly why Research and Idea come first in this loop: Andromeda rewards a wide spread of genuinely different concepts more than a narrow set of micro-tweaks.

The funnel

Same product, different job depending on who's watching.

ToFu Cold audience Win the hook
MoFu Warm audience Show the proof
BoFu Ready to buy Push the offer

What that loop produces in a year

Volume is only useful if the process behind it holds up. Here's what a full cycle of Research to Launch adds up to.

300+ Creative deliverables shipped per year
40+ Variations tested off a single winning concept
10+ Weeks a top performer can be extended, not retired

Reading the data, then reacting to it

Three numbers tell most of the story. The job is knowing which one is broken.

CTR Click-Through Rate

Did the hook earn a click?

If it's low: the hook is the problem. Recut the first 3 seconds before touching anything else.

CVR Conversion Rate

Did the click become a customer?

If it's low: the ad is lying to the landing page. Align the offer and message before blaming the creative.

CPO Cost Per Order

What did this customer actually cost?

If it's climbing: the concept is fatiguing. Time to re-cut, not just re-target.

AOV Average Order Value

How much did they buy while they were there?

If it's flat: the creative sold one hero SKU. Brief a bundle or cross-sell angle next round.

Read the numbers Diagnose which stage broke Brief the next round Ship it back into testing

What this looks like on your account

A rough picture of where most accounts start, where this process has taken Kind Science, and the realistic goal in between.

Cost Per Order
Climbing −70%

The most common pattern I fix first: a fatiguing concept quietly driving cost up.

Creative Lifespan
1–2 weeks 10+ weeks

Every winner gets a second life through iteration, instead of getting retired.

Revenue Growth
Flat 2×–4×

Growth that compounds from the creative system, not just from spending more.

Figures reflect the Kind Science program discussed in the Ellen DeGeneres case study. "Climbing / flat / 1 to 2 weeks" describes a common starting pattern, not a specific competitor's data.