Strategy decks talk about brand. This series measures it.
Four artifacts measuring Czech advertising from three sides — picture, sound, language — each one a different AI model reading a different signal of brand identity. Plus a multimodal tool that combines all three on any spot URL. Built so a strategist can run the analysis on the client's next pitch on Monday morning.
Data & cloud engineer at Alma Career. Built across a few weekends to test what brand competitive analysis looks like when it starts from public data instead of pitch decks. All four projects open-source; every finding committed publicly before verification (see PREDICTIONS.md).
A worked example. Pilsner Urquell's image spot is the only spot in the 47-corpus that commits to a world simultaneously by picture and by sound. Here is what that looks like across all four measurement frames.
Color · brand-vs-world gap
102°
Distance on the hue wheel between Pilsner's brand green and the dominant on-screen color. Anything above ~90° reads as a different color. Pilsner sits in the world-color camp.
Color · brand-color presence
0.2%
Share of the spot's dominant on-screen colors that sit within ±15° of Pilsner's brand green on the hue wheel. The brand color is essentially absent from the picture.
Sound · music fraction
86%
The single most music-led spot in the corpus. The next-most-music spot is 77%. The other 45 spots are voice. Pilsner is the lone music-led outlier.
Sound · tempo and key
81 bpm
F# major. Slow and warm, well below the 127 bpm corpus median. Music-led spots are slow; tempo and voice-vs-music move together.
The synthesis. When Pilsner committed, they committed consistently. The brand color steps back across the picture (102° gap, 0.2% presence). The brand voice steps back from the audio mix (86% music, 14% speech). Tempo slows. Both modalities make the same creative call independently. That coherence is the finding. Most spots in the corpus commit on one axis and hedge on the other, producing the muddle. Pilsner's Alfons is the textbook example of "decide what camp you're in, then commit by every available signal."
What a strategist does with this. Run Spot Scorer on a draft before greenlight. If the brief says "image-led, emotional, world-building" then the scorecard should show: gap above 90°, anchor share below 5%, music fraction above 50%, tempo below 100 bpm. If the dailies are scoring 50° gap and 70% voice, the spot is drifting into the muddle. Catch it in pre-mix, not after picture lock.
03
And one that almost lands
A counter-example. Albert's spring 2026 spots commit to a world visually but to a voiceover audially. Each commitment is extreme on its own axis. The pair doesn't reinforce: the world-color picture promises atmosphere, the voiceover delivers product copy. This is the muddle's stealthiest form — looks decisive metric-by-metric, reads as conflicted in motion.
Color · brand-vs-world gap
174°
Far end of the world-color camp — the dominant on-screen color (warm bistro tones, wooden boards, ambient food light) is nearly opposite to the Albert brand blue on the hue wheel.
Color · brand-color presence
2%
Albert blue appears on essentially no part of the picture. The end-card logo carries the brand; the spot's world doesn't.
Sound · voice fraction
88%
Top quartile voice-led across the corpus. Voiceover narrates product, brand, value-prop with music as a thin bed.
Topography · neighborhood
Mass-market
Sits in the same homepage neighborhood as Lidl, Rohlík, McDonald's, KFC — brands whose homepages share a promotional template shape. Same neighborhood as everyone they're trying to differentiate from.
What's incongruous. World-color spots usually pair with music-led (cinematic mode — picture and music build a world together). Voice-led spots usually pair with brand-color flood (promotional mode — picture and voiceover sell product together). Albert combines world-color with voice-led — picture promises atmosphere, voiceover delivers product copy. The two modalities pull in different directions; the spot reads as either an image campaign with too much talking, or a promo with too little brand color, depending on which signal the viewer notices first.
The brief Albert almost shipped. Either commit to both world signals (let the bistro footage carry it; drop the voiceover, raise the music to 50%+, slow tempo) — and you get a Pilsner-shaped image spot. Or commit to both brand signals (raise Albert blue to 40%+ of the picture, keep the voiceover) — and you get a Lidl-shaped promo spot. The current Albert spot picks neither half all the way. The single-line brief: "Pick a coherent world or a coherent promo. The muddle is invisible."
04
What you can hire me to do
Three concrete services derived from the methodology above. Each ships a single artifact, on a fixed turnaround. Pick one as a pilot; the next conversation tells us whether to scale.
01
Competitive scan for an active pitch
5–8 spots from the client's category + 3 competitors, run through color + sound + topography. One slide deck with: where the category lives on the map, where the client currently sits, where the client could land to be distinctive, and the numerical brief for getting there.
Turnaround: 4 working days · Output: 8–10 slide deck + raw data · Best for: the pitch presentation slide you wish you had a number behind
02
Spot scoring + 1-pager for any draft URL
Single video URL in → 1-page brief out, in the same shape as the McDonald's sample. Three numerical findings, color reading, sound reading, position reading, single strategic synthesis with a specific next-spot recommendation.
Turnaround: 24 hours · Output: 1 PDF / Notion page · Best for: a dailies-review checkpoint, or feedback on a finished cut before client sign-off
03
Pre-shoot brief audit
Before greenlight, score the client's current homepage + their last three published spots + their three nearest competitors. Identify which "homepage template" the brand is trapped in. Generate the numerical creative brief that would move them out. Catch the drift in the brief stage, before the shoot.
Turnaround: 3 working days · Output: 1 numerical brief + 1 strategic synthesis page · Best for: the upstream conversation with the creative director, before storyboarding
30-minute walkthrough, no pitch attached.
Show me one of your active client briefs or a recent shipped spot. I'll run it through the toolkit live and walk you through the output. No deck, no pre-work from you. Most useful when you have a real ongoing pitch I can demo against, not a hypothetical.
One operator, not a team. Everything on this page was built solo. Three concurrent client engagements is realistic; if you want this scaled into an agency-side team product, that's a different conversation worth having.
First three clients get pilot pricing. This toolkit hasn't been commercially engaged before. The AI pipelines have been stress-tested on 47 public Czech spots and 24 public homepages; not on confidential client data yet.
The reports describe what the AI sees, not what you should do. The strategist (you, or me on the call) does the judgment layer. The brief generator is a structured first draft, not a directive — the Pilsner Urquell brief in the Strategic Topography report shows what an "obviously wrong" mechanical recommendation looks like, and that's where the human edit lives.
Everything is open and reproducible. Code is MIT, datasets CC-BY-4.0, findings committed publicly before verification (PREDICTIONS.md). If you don't like a number I show you, the code that produced it is one click away.
The tools have failure modes I name in the open. Spot Scorer can fail on YouTube videos that hit a bot challenge; the topography brief generator points Pilsner toward Lidl when nobody sensible would. Both reports flag those failures where they appear, so you see what you're buying.