A builder’s case for intelligence you can trust
How to build marketing AI systems that actually hold up in the real world
Table of contents
Most AI conversations in marketing start with capability and end with confusion. More tools, more outputs, more dashboards, and somewhere in the middle, the decision still doesn’t get made.
This piece is for the people who feel that gap most acutely: the ones accountable for decisions that can’t be defended by a dashboard alone. Analytics leaders, brand marketers, insights teams, or anyone who has had to walk into a room with competing narratives and leave with a direction.
What follows isn’t a technology critique or a vendor argument. It’s a structural diagnosis, and a case for building differently. The goal here is simple: offer a clearer way to think about how AI should support decisions when they actually matter.
– Drew Kutcharian, Co-Founder & CTO at DISQO
Introduction
I have been building systems for most of my life. Not the cosmetic kind that exist to ship features or impress in a demo, but the kind that reveal their truth only after they have been stressed by reality.
Systems where you wake up thinking about failure modes because you know they will surface eventually. Systems where you cannot unsee the seams, where every abstraction leaks under pressure, and where the real measure of success is whether the logic still holds when the data is messy, the timeline is unforgiving, and a decision has to be defended in a room full of smart skeptics. That builder mindset is how I evaluate AI in advertising today.
We are in an extraordinary period of acceleration. There is more data than ever, more compute, more automation, and more sophisticated tooling layered across the marketing stack. At the same time, there is more confusion, more dashboards, more synthetic insight, and more operational drag. Teams are surrounded by information yet starved for conviction. Meetings multiply, narratives compete, and decisions slow down precisely when speed matters most. While the industry is moving quickly, the ground beneath it is still soft.
This tension is not caused by a lack of tools. It’s caused by the way our systems are designed. Modern analytical and advertising systems excel at producing outputs. They summarize performance, surface correlations, and generate interpretations at scale. What they fail to do consistently is preserve the logic that connects evidence to action. They tell us what happened without carrying forward why it matters, what assumptions were made, what tradeoffs were accepted, and what would invalidate the conclusion. As a result, confidence fails to build and instead erodes over time.
Brand marketers, analytics leaders, and insights teams experience this every day. Budgets move faster than certainty. Creative decisions are made before understanding settles. Risk is managed through delay rather than clarity. AI accelerates the production of artifacts, but it does not resolve the underlying decision tension. In many cases, it intensifies it. This is not a call for less ambition or slower innovation. It’s a call for better systems.
This is a builder’s case for intelligence you can trust. Not as a feature, a compliance checkbox, or an afterthought, but as an architectural discipline. It is what allows scale without drift, preserves meaning through automation, and enables marketing organizations to turn speed into advantage rather than risk.
The next real advantage in AI is not more output. It's trusted direction.
For the last several years, we’ve relentlessly optimized for output. More copy, more variants, more assets, more summaries, more charts, more analysis layered on top of analysis. That focus was deliberate and effective, stripping friction out of production. It collapsed cycle times for execution and made teams faster at creating things that used to be slow and expensive to produce.
But it didn’t make organizations better at deciding. In many cases, it did the opposite. As output velocity increased, decision velocity collapsed, as options multiplied, narratives crowded the field, and artifacts piled up, demanding interpretation. What looked like progress at the surface introduced hesitation underneath, because the system could produce answers faster than it could establish belief.
Speed without clarity does not create advantage, it creates risk.
This is where the industry has been imprecise about what intelligence actually is. Insights tell you what happened. Intelligence tells you what to do next. Insights are descriptive. Intelligence is directive. Insights expand awareness. Intelligence compresses uncertainty into action.
Most marketing stacks are built to generate insights. They were never designed to generate trustworthy direction. They aggregate signals, visualize outcomes, and leave the hardest part to humans: reconciling conflicting evidence, assessing confidence, and choosing a path forward that can be defended when it matters.
AI is going to expose that gap, not close it. As models get better at generating analysis, summaries, and recommendations, the absence of decision structure becomes more visible. When every answer sounds plausible, conviction becomes scarce. When reasoning is opaque, confidence becomes performative. When systems cannot explain why one path is better than another, speed turns into hesitation disguised as rigor.
The organizations that win in the next phase will not be the ones that generate the most; rather, they will be the ones whose systems are designed to point, consistently and transparently, toward the next best move. Direction, not volume, is what compounds.
A builder’s perspective when AI systems meet reality.
Every technology wave separates narration from construction. Narration chases momentum, amplifies surface signals, and rewards immediacy. Construction accepts uncertainty and commits to the slower work of making systems that endure. It’s governed by constraints, refined through failure, and validated only when the system continues to function under pressure rather than ideal conditions.
Builders engage new capabilities with discipline. They examine the assumptions a system depends on because assumptions are where reality pushes back. They trace how meaning changes as data moves through layers, because abstraction always extracts a cost. They watch where trust is earned and where it quietly erodes, because confidence is cumulative and fragile. Their attention stays fixed on the smallest unit of work that alters outcomes in the real world, not on artifacts that perform well in isolation.
The excitement around AI is warranted; its capabilities are advancing quickly and delivering real leverage, but capability alone does not constitute completeness. Most systems today are very good at generating outputs, yet they struggle to retain context, intent, and the decision logic needed for confident action inside complex organizations. Larger context windows help, but LLMs still fall short when it comes to consistently following directions within those contexts.
When something looks impressive yet feels structurally unstable, builders recognize the signal early. Experience teaches that these weaknesses do not self-correct with scale. They intensify.
Data was never the constraint.
Decisions were.
The quiet truth is that brand measurement is hard for reasons that go beyond real-world complexity. It’s hard because the ecosystem is structurally fragmented. Measurement data lives inside walled gardens, and exposure is rarely connected cleanly to downstream consumer behavior. The expertise required to translate measurement into action sits at the last mile, concentrated in a small number of people who are forced to interpret incomplete signals under time pressure. This is a practical diagnosis. It reflects the operational reality we design against every day.
That structure explains why the industry continues to over-optimize on proxy metrics like impressions, views, reach, likes, and clicks, even as hundreds of billions of dollars move through the system each year. Not because marketers lack sophistication or intent, but because the system does not reliably connect spend to outcomes in a way that supports confident decision-making. When cause and effect remain uncertain, optimization defaults to what is easiest to observe rather than what matters most.
Teams adapt rationally to that environment, adding layers of reporting to compensate for missing connections. They hire more analysts to reconcile conflicting signals. They introduce more processes to manage ambiguity, and decisions slow. Not because people are cautious by nature, but because caution becomes the safest response when accountability outpaces clarity.
Then AI arrives and offers speed. Faster summaries. Faster charts. Faster interpretations. That is useful, but it misses the point. The constraint was never the speed of synthesis. The constraint was the ability to choose with confidence. What the system needed was not another explanation of what happened, but a clearer path toward what to do next.
Generative AI made everything faster, except the decision.
Most of the early AI gains in marketing have landed in production, and those gains are real. Teams can now draft briefs in minutes, generate creative variants at scale, summarize dense reports, write SQL without friction, and assemble decks almost instantly. Throughput has increased across the board.
But production was never the binding constraint for serious investment in brand advertising. The real constraint shows up in a tighter set of unresolved questions:
- What do we actually believe is true?
- Why do we believe it?
- What action follows from that belief?
- What tradeoffs are we accepting?
- How confident are we?
- What evidence would change our mind?
Most insight products don’t answer these questions. They decorate around them, adding artifacts without collapsing uncertainty. When a system responds to ambiguity by generating more charts, it does not create clarity. It increases cognitive load.
Speed without decision clarity increases risk rather than advantage. That is the thesis this work is built around.
Trust is an architectural choice.
Trust is often treated like an interface problem. Add explainability. Add citations. Add a panel that shows why a recommendation appeared. Add a more polished summary. Those things help, but they do not solve the core issue. They explain an output after it is generated. They do not preserve the reasoning that made the output worth trusting in the first place.
That distinction matters. In real decision systems, trust is not created at the presentation layer. It is created upstream, in how data is structured, how meaning is carried forward, how assumptions are captured, and how decisions remain traceable from input to action. Once that chain breaks, no amount of polish can restore it.
Builders learn this early. Context lost upstream does not magically reappear downstream. Logic that was never encoded cannot be audited later. Workflow intent that was never captured turns recommendations into generic advice. Semantics that drift across systems create automation that looks right on the surface while quietly introducing error underneath.
This is why trust cannot be bolted on. It has to be designed into the system from the start.
In advertising, that means more than making models explainable. It means building systems where consumer signals, exposure data, performance outcomes, and workflow context remain connected. It means preserving the decision path, not just the result. A system that only knows what happened can summarize the past. A system that preserves how and why a conclusion was formed can help teams decide what to do next.
That is the real architectural line between output and intelligence.
Trustworthy intelligence is not just fluent. It is grounded. It carries context forward and makes assumptions visible. It shows where confidence comes from and what would change it. It gives teams direction they can interrogate, defend, and act on under pressure.
That is not a UX feature. It is a system property. And in the next generation of marketing AI, it will be one of the few that actually matters.
The missing link in AI is context
This is where generic AI tools stall. They can ingest customer data, summarize reports, and generate fluent responses, but marketing decisions demand more than synthesis. They require context that persists across time and binds together consumer reality, media exposure, campaign performance, prior decisions, and the actual workflows teams run week after week.
That long-running context is what allows an AI system to stitch things together. It is what creates the steel threads between signals, decisions, outcomes, and changing market conditions. Without it, each prompt starts too close to zero. The system may sound intelligent in the moment, but it is still reconstructing meaning instead of carrying it forward.
Today’s models are still fundamentally prediction systems. They are powerful at pattern recognition and language generation, but they do not naturally hold the durable operating memory that human teams build over time. They do not accumulate understanding on their own unless the system around them is designed to preserve it.
Context is what turns analysis into intelligence. It is the difference between observing what happened and forming a judgment about what should happen next. And it does not appear by accident. It is the result of architectural choices about what the system remembers, how meaning is maintained, and how prior reasoning stays connected to future action.
Generic insights systems take customer data in and produce generalized conclusions. Advertising intelligence combines consumer data, exposure data, and performance data, grounds them in an operating context that endures across decisions, and produces direction that can be acted on with confidence. That distinction defines whether a system merely informs or actually guides. 
The system we need is a platform
No marketer wants more tools. They want momentum, defensible decisions, and the ability to move at market speed without accruing hidden risk.
The problem is that most “systems” in marketing today are not actually systems. They are collections of tools stitched together by workflow, interpretation, and human effort. Each campaign restarts the process without preserving the logic behind it. Context is rebuilt. Assumptions are re-argued. Decisions do not compound because the underlying structure does not persist. It supports moments of analysis, not continuity. And without continuity, intelligence resets.
That is the gap. A platform resolves that.
A platform is not just a more integrated set of tools. It is a system where data, context, and decisions persist over time and compound with use. That is why platform thinking matters. It creates the conditions for progress to build. In our product vision, intelligence emerges from a layered architecture that reflects how decisions actually form.
An operational data platform ensures signal integrity. A meaning (ontology) layer gives raw data consistent semantics. A context layer applies real use cases and workloads. An application layer surfaces that intelligence through the products teams actually use. AI agents automate work across workflows rather than creating isolated moments of efficiency.
This is not abstraction for abstraction’s sake. It is a map of where intelligence is actually formed. Systems built only at the application layer produce dashboards that age quickly. Systems built with meaning and context can interpret signals consistently, connect cross-channel effects, preserve decision logic, and earn trust because they hold up across repeated decisions, not just one-off analyses. That shift turns reporting into prediction.
Collaboration requires shared reasoning
AI collaboration is often reduced to interface design. A group chat with your AI Assistant app is not collaboration. Collaboration requires shared reasoning across time.
For AI to function as a real collaborator inside a marketing organization, it must hold context beyond a single interaction, including campaign history, measurement strategy, brand constraints, and prior decisions. It must show traceable reasoning with clear chains of evidence rather than plausible language. It must speak in the language of decisions by mapping guidance to levers teams can actually pull. It must invite human judgment, recognizing that intent, taste, and risk tolerance remain human responsibilities.
Interpretation that leads to aligned action is the product. Dashboards, models, and reports are means to that end. Alignment is the outcome that matters.
Speed and precision only matter
if they are real
Speed and precision are a competitive advantage when they reduce time to confident action. They are a liability when they accelerate ambiguity.
Marketing operates in compressed cycles where budgets shift, creative saturates, competitors react, channels splinter, and sentiment turns. When measurement systems cannot keep up, teams fall back to the simplest available proxies. That behavior reflects rational adaptation to systems that fail to enable timely decision-making.
The objective is not to generate more artifacts. The objective is to reduce cycle time from signal to action without sacrificing trust. Workflow automation matters because workflows are where time disappears. Time loss is where advantage erodes.
The best systems preserve meaning end-to-end
This is not a prediction. It’s pattern recognition shaped by building systems that endure. The systems that last prioritize decision guidance over raw output. They reduce uncertainty rather than just labor. They preserve meaning through the pipeline. They make trust measurable. They embed directly into workflows instead of demanding process change.
Intelligence you can trust scales action because teams do not have to re-litigate every conclusion. AI does not need to replace humans. It needs to make the next move clearer. AI is only useful if its recommendations can be trusted when decisions actually matter.
What teams should expect
from AI measurement systems
Serious teams should expect a clear shift in what their systems deliver.
- They should expect forward guidance rather than retrospective reporting, with outputs that translate measurement into next-actions and expected outcomes.
- They should expect coherent semantics rather than fragmented metrics, so teams reason consistently instead of acting as translators.
- They should expect trustworthy confidence rather than opaque assurance, with assumptions and uncertainty made visible.
- They should expect workflow automation rather than tool switching, removing repetition without removing accountability.
- They should expect compounding intelligence that carries memory forward instead of resetting every campaign.
None of this is abstract. It shows up in shorter cycle times, fewer meetings, faster reallocations, and stronger conviction in creative and partner strategy.
How we are building at DISQO
At DISQO, we build toward a simple idea: advertising intelligence should be trusted, easy to adopt, built for scale, and intelligent by default.
Making that real requires more than model output alone. It comes from connecting consumer sentiment, verified ad exposure, and behavioral outcomes within a shared context, while preserving how each signal contributes to a recommendation.
Our proprietary consumer data is central to that system. Because we have direct consumer relationships, we can ground intelligence in a richer understanding of what people saw, how they responded, and what they did next. That gives us more than isolated signals. It gives us the ability to carry meaning across the full decision chain.
We are also deliberate about where AI plays a role. We use AI for what it does best: synthesis, reasoning across complex inputs, and translating evidence into clear next steps. But we do not ask AI to generate the math. Calculations, measurement logic, and core numerical outputs are handled through deterministic analytics tooling and specialized ML models, where methods are explainable, and results can be validated.
That separation is not incidental. It is architectural. It ensures that recommendations are informed by AI, while the underlying numbers remain grounded in transparent and auditable mechanisms. The result is intelligence teams can follow, interrogate, and trust.
A builder’s definition of progress
Technology waves mature along a familiar path, from output to insight to decision guidance. AI is entering the decision guidance phase now.
The winners will not be defined by the loudest demonstrations. They will be defined by systems that preserve meaning, show their reasoning, earn trust under pressure, and help humans decide together.
Decisions are a scarce resource. Teams that can decide clearly, quickly, and defensibly will outperform teams that cannot, even with access to the same channels, platforms, and models.
Your brand is your most valuable asset. Treat it accordingly. Invest in systems that create shared understanding, trustworthy direction, and decisions that hold up in the real world. The future will reward clarity of action, not volume of dashboards.