Skip to content

The argument · Compounding

The tool should know your brand better in month twelve than in month one

A new agency takes three months to sound like you. Most software never gets there at all, because nothing it learns in January is available to it in June.

20 August 2026 · 6 minute read

Almost all marketing software is stateless. Each brief, each campaign and each report starts from an empty context, and whatever was learned last quarter exists only in the heads of whoever was in the room. That is why the fourth campaign takes as long to brief as the first, and why a tool that produces creative still needs the same corrections every time. A system that keeps what it learns behaves differently: the voice specification gets more precise each time someone overrules it, the segments carry the history of what worked on them, and the competitor set has a memory of how each rival has moved. This compounds slowly and unglamorously, and it is the main reason to run the work inside one platform rather than assembling it per campaign.

What stateless actually costs

It is not obvious, because nothing is broken. The tool works exactly as well in month twelve as it did in month one, which sounds like reliability and is actually the problem.

Every time you start somethingIn a stateless toolIn one that keeps context
Brief a piece of creativeRe-explain the brand, correct the same thingsThe corrections from last time are already applied
Choose an audienceRebuild the segment, hope it matches last quarterThe segment is the same one, with its history attached
Respond to a competitorStart from what they did this weekStart from how they have moved over eighteen months
Judge a resultCompare to a benchmark from somewhere elseCompare to what this brand did last time, on this segment

Read the middle column as a description of a new hire who is replaced every month. Competent, fast, and starting from nothing each time.

What is actually worth accumulating

  1. The voice, as a specification rather than a document. A brand book is written once and interpreted forever. A specification gets sharper every time somebody rejects an output, because the rejection is the correction.
  2. What worked, attached to who it worked on. A result that is not joined to a segment teaches nothing transferable. Joined to one, it becomes a prior for the next decision about that group.
  3. How your competitors move. A single observation is a data point. Eighteen months of them is the difference between noticing a competitor changed their positioning and knowing they do it every October.
  4. The language your customers use. Not the category vocabulary, and not last year's. What accumulates here is a record of how the words shifted, which is usually the earliest signal that a market has moved.

The uncomfortable part

Accumulated context is only as good as what went into it. A brand that uploads inconsistent creative and never corrects an output ends up with a system confidently reproducing its own worst habits. This is not a flaw specific to us; it is what accumulation means. It is worth knowing before you assume the compounding happens on its own.

Why this is an argument for running the work here, not just the research

This is the honest reason we encourage brands to bring their own creative in and run campaigns from the platform, rather than using it only for research. It is an encouragement rather than a requirement: the modules are chosen individually and research-only is a legitimate way to buy. But the loop only closes if the result comes back, and that is easier when the execution is in the same place.

Research that leaves the building never learns whether it was right. A campaign run somewhere else produces a number in someone else's dashboard, and the segment that was targeted never finds out how it performed. Everything above depends on the outcome returning to the same place the decision was made, and that only happens if the execution is there too.

What it does not do

It does not make the first month better. A system with no history is a system with no history, and anyone claiming their platform arrives pre-trained on your brand is describing a generic model with your logo on it. The first quarter is ordinary. The argument is about the second year.

A tool that works exactly as well in month twelve as in month one is not stable. It is forgetful.

If you are an agency

For an agency this cuts both ways, and it is worth being clear-eyed. Accumulated brand context makes you faster and reduces the cost of onboarding a new person onto an account. It also makes the context portable in a way a folder of decks is not, which a client may eventually notice. Agencies that treat it as a shared asset rather than a private one tend to have the easier conversation.

How we do this ourselves

This is what Brand DNA is for, and why it sits underneath the rest of the platform rather than beside it. Everything Signal learns and everything Studio produces writes back to the same brand context, which is what makes the second year different from the first.

How the brand context is built

Common questions

Does this mean you train models on our data?

Your brand context is yours and is not used to improve anything for another customer. The accumulation described here happens for your brand alone.

What if our brand deliberately changes?

Then you overrule it, and the overrule becomes part of the specification. A repositioning is a large correction rather than a reason to start again.

How long before this is noticeable?

Realistically a quarter before the creative corrections stop repeating, and closer to a year before the competitor and segment history is worth much. Anyone promising faster is describing the demo, not the product.

Can we take it with us if we leave?

Yes. The brand specification, the segments and the underlying data export. We would rather lose a customer cleanly than hold them by making the exit expensive.

Know First, Act Faster

We do the research most platforms skip, then run the campaigns it points to.

Book a demo
Book a Demo