Know if the concept works before you shoot it.
Coherence reads the brief, the script, the moodboard and the storyboard the way the platform's algorithm will read the finished asset. It returns one on-brief score, the exact points where your own materials drift from the brief, and the fixes worth making, all while changing things is still cheap.
Not a platform you buy. A capability we run with you.
There is no login to hand out, no seats to license and no tool for your team to learn. Coherence runs inside our stack, with our people reading the output next to your creative team. You get the gate without adding a system to your invoice or a project to your roadmap. It is a proprietary capability, the pre-production stage of the same engine that runs Wandrome.
Production is where the money goes, and where you learn nothing.
In the film projects we see, production and editing carry the large majority of the cost, typically between 70 and 90 per cent. The signal you actually need (click-through, conversion, reach) only arrives after all of it has been paid for.
The hero spot contaminates every derivative: if the master concept starts misaligned, every cutdown, every format and every market version inherits the same flaw. You are not losing one creative, you are losing a tree.
The logic is simple: the gate costs a fraction of the production it protects, and a run takes minutes. Moving the decision earlier, before the expensive part, is the cheapest lever on the whole creative budget. No impact figures on this page are claimed as measured results.
The algorithm is not a black box all the way down.
The signals the platform's algorithm rewards, Meta first, are identifiable, and most of them are already visible in pre-production. Every one can be inferred from a brief, a script, a storyboard and a set of frames, long before the camera turns on. For television and other non-algorithmic media the same read covers coherence and persona; algorithm adjacency applies to the digital cutdowns.
The grid runs on six axes (persona, hook, brand, message, motion, audio), and persona alone resolves eleven declared dimensions: age, gender, segment, lifestyle and the rest. Gating by artefact type means a written brief is never judged on motion.
Four steps, from artefact to decision.
Bring what you have
Brand brief, agency brief, script, moodboard, storyboard, frames, animatic, rough cut. Text or files, in any combination. You do not need the full set to start.
It anchors on the brief
The brand brief is the anchor, and everything else is measured against what the brand actually asked for, not against an average of the room. The score tells you how on brief the concept is, artefact by artefact.
You get a matrix
Every artefact against every applicable axis, plus an aggregate row for the project as a whole. Where the numbers drop, the disagreement is specific and locatable.
You decide
Every recommendation carries the lift it would produce if applied. Go to production, or iterate and run it again, with the cost of each fix in front of you.
One grid, and "on brief" stops being an opinion.
Rows are your artefacts. Columns are the axes the read cares about. Each cell scores how well that artefact agrees with the brief on that axis. The bottom row is the whole project, measured against the persona and the objective you declared.
| Artefact | Persona | Hook | Brand | Message | Motion | Audio |
|---|---|---|---|---|---|---|
| Brief | 0.82 | 0.70 | 0.60 | 0.85 | n.a. | n.a. |
| Script | 0.78 | 0.88 | 0.65 | 0.80 | 0.50 | 0.90 |
| Storyboard | 0.75 | 0.82 | 0.78 | 0.70 | 0.72 | n.a. |
| Frames | 0.70 | 0.85 | 0.80 | 0.65 | 0.85 | n.a. |
| Aggregate | 0.76 | 0.81 | 0.71 | 0.75 | 0.69 | 0.90 |
The aggregate row answers one question: is the concept on brief? Everything above it shows exactly where the script, the storyboard or the frames drift from what the brand asked for, while it is still cheap to pull them back.
Three scores, the divergences, and the fixes with their price.
On brief
How closely everything you have agrees with the brand brief.
Persona alignment
Does the work match the person you said you were talking to?
Algorithm adjacency key number
How close the concept sits to what the platform algorithms reward, Meta first.
Visual coherence complementary
A separate signal that compares the artefacts to each other by image rather than by description. Reported on its own and, by design, it never contaminates the main score.
It is not a report. It is a loop that closes when the concept is ready.
As new artefacts arrive the read is cumulative: the storyboard is scored against the brief and the script, the rough cut against everything before it. The concept converges instead of drifting, and by the time you commit the production budget the disagreements are already resolved.
Your data shapes the advice, never the score.
Feed your brand library into the scoring step and the read stops being a mirror of the algorithm and starts being a mirror of you, which is comfortable and useless. So we keep it out until after the score is set.
The artefacts, and nothing else
- Brief, script, moodboard, storyboard, frames, cuts
- The persona and objective you declared
- Fixed weights, no temperature in the model calls
Your world, in the recommendations
- Your brand library and visual codes
- Your category and competitive context
- Written into the fixes, not the number
The read is reproducible by design: fixed weights, temperature-zero calls, and no randomness of our own. Every run is stamped with its scoring version and the model behind it, so two reads can always be compared and explained.
There is a science behind the read.
You do not need this section to use Coherence. But if you want to know why the number is defensible in front of a technical buyer, here is what sits under it.
01 Agreement, not correlation
Each cell is an agreement-and-overlap measure between two artefacts on one axis, combined with cosine similarity over the features a language model extracts from each. It is not a Pearson correlation and it is not a vibe. Weights are fixed and the calls run at temperature zero, so a re-run reproduces the cell.
cell = f(agreement, overlap, cosine(features))
# features extracted by LLM, temperature = 0
02 Normalised per signal type
Different axes are different kinds of data, so each is normalised on its own terms: age as interval overlap, socioeconomic segment as an ordinal distance, brand cues as Jaccard set similarity. A number on the persona axis and a number on the brand axis are made comparable on purpose.
03 Brief-anchored comparison
The comparison mode is brief_anchored: the brand brief is the fixed reference and every other artefact is measured against it. Consensus between artefacts is only used as a fallback when no brief exists, so the read never quietly drifts toward the average of the room.
04 Native video, logged inference
Video is analysed from the file itself, not reconstructed from sampled stills, up to the size the model reads inline. Every inference is logged with model, tokens, cost, human oversight and risk level, so the trail exists before anyone asks for it. Fixed weights, reproducible formulas, auditable end to end.
Reproducibility is a deliberate choice, not a limitation. A read you can version, compare and explain is a read you can argue with, and a read you can argue with is one a serious team will actually use.
Recent, in use, and built to be checked.
We are candid about maturity. Coherence is a recent capability, already running with clients, and engineered so a technical buyer can verify the claims rather than take them on trust.
Used with brands and distribution partners on live projects. The engine underneath is already commercial. Client names stay private until they choose otherwise.
Audited at code level by two independent models, with every finding adjudicated against the live source before it was accepted. The improvements it surfaced are in the current version.
Fixed weights and temperature-zero calls, every run stamped with its scoring version, so two reads can be compared and explained.
A security and compliance overview, written for the people your procurement team will send it to.
See a live read on one of your own concepts in about 45 minutes. A real matrix on your own brief beats any number we could put on this page, and you keep the read whether or not we go further.
One ruler, applied to every room.
Coherence pays off wherever concepts get produced at scale and quality drifts between teams, agencies and markets. Three profiles get the most from it.
In-house or dedicated-agency
Put a gate before the production budget goes out, so a concept that will not distribute is caught while it is still a document.
Creative & production
Replace the subjective argument about whether the execution meets the brief with a shared, specific reference, artefact by artefact.
Multi-market
Apply the same ruler across agencies, markets and every derivative of a spot, so quality stops depending on which room made it.
Category-agnostic and medium-agnostic: if your team touches the concept before it is produced, Coherence has a place in your process, from a single hero film to a wave of localised cutdowns. Where it does not fit, we say so in the first call, not the third.
See a concept read live, before you commit a budget.
Around 45 minutes. One brief, and whatever exists around it.
The earlier the gate, the cheaper the mistake.
Coherence is stage one of a three-stage chain. Each stage catches a different cost, and real performance feeds back into both reads.
Coherence
Validates the concept before there is an asset. A clear go or iterate, while changing is cheap.
before you produceWandrome
Scores the finished creative the way the algorithm will read it, before you put media behind it.
before you investReal performance
Live results feed both earlier stages and recalibrate the read for your market.
after it runsCoherence only pays off if you have a hand on the production cycle. If all your creative is commissioned externally and arrives finished, there is no gate to place and the value sits in stage two, not here. It is not a revenue forecast and it is not a guarantee of distribution: it is a structured, reproducible read of how close a concept sits to what the platform algorithms reward, delivered while the concept is still cheap to change. The read is calibrated on the markets we already have live data for, and it recalibrates as your results come in; in a market we have not calibrated yet, treat the first cycle as directional. The go or no-go stays a human decision. See Wandrome for the finished-asset read →
Built to be audited, not just used.
Isolated by client
Every project, artefact and run is anchored to your organisation. Files live in private storage and are only ever served through short-lived signed links.
No customer data
Coherence reads creative material, not your audiences. It needs no access to your ad account and carries no end-customer personal data.
Every inference logged
Model, tokens, cost, human oversight and risk level are recorded for each analysis, so the trail exists before anyone asks for it.
Human in the loop
The output is a recommendation. The decision to produce is always yours. Built for EU AI Act readiness, with our assurance pack available on request.
Start with a concept you are about to produce.
Bring one brief and whatever exists around it. You leave the first session with the read, whether or not we work together.
A live read
Around 45 minutes. We run one concept in front of you and walk through the matrix, the divergences and the fixes.
Format: a single working session, no commitmentOne project, end to end
We put Coherence on a single campaign from brief to rough cut, with the loop running at every step, and you see what it changed before production.
Format: a fixed-scope pilot on one campaignThe gate on the next wave
Coherence sits as a standing checkpoint between concept approval and production sign off, applied to every piece and every derivative.
Format: an ongoing retainer across the pipelineCoherence is a Wise Pirates proprietary capability, the pre-production stage of the same engine that runs Wandrome. Run for clients and partners.
The questions we hear most.
Last updated: August 2026. Illustrative figures on this page are examples, not guaranteed results.
What exactly is Coherence?
How is it different from Wandrome?
What do we need to have ready to start?
Does it replace our creative team or our agency?
Will the score be biased by our brand guidelines?
Does it work for video and for TV, or only for static?
What happens to our materials, and is any of this a compliance problem?
Do we have to buy licences and learn another tool?
Validate the concept before it exists.
Bring one brief. In about 45 minutes you will see the matrix, where your own materials disagree, and the fixes worth making, while changing them is still cheap.
Start a concept read →
