Decision Marketplace℠
Crowdsourced
decision making
With each addition, the Decision Marketplace gets smarter.
Unlocking better decisions
Can you remove a human expert and still make great decisions for something as complex as ad selection and pricing?
We've been exploring this question for years.
Monolithic AI models sort of worked
Our prior approach to automated decision-making relied on monolithic, AI-driven models that were developed over months by a highly specialized team at a significant cost.
But still required human calibration
In production, these models still required continuous calibration from human experts to ensure that their ad selection and pricing decisions hit business goals for advertisers, publishers, and Decide.
This new era requires doing better
Human-driven ad targeting and pricing is almost dead. Each decision must adapt to novel conditions with expert-level precision—billions of times per day.
Solving this problem required going back to the drawing board.
In the real world, human experts still do a lot of work
Smoke and mirrors abound
Terms like machine learning (ML) and artificial intelligence (AI) are commonly abused and rarely understood.
Must be better than a human
Machine-driven predictions are only as good as their ability to enable real-world decisions in a manner that is faster and cheaper than human experts while also delivering a superior output.
This last mile is hard
In most fields, including digital advertising, replacing human experts and achieving this last mile is rather difficult.
Solving the last mile
Decision models need to be first-class citizens. Solving this last mile requires decoupling decision models from prediction models.
Data
DEFINITION
1. Snapshots of historical and current conditions
Predictions
DEFINITION
2. State of knowledge about future conditions
Decisions
DEFINITION
3. Calibration, assessment, and action
Capturing decision making in a product
A decision platform is a set of building blocks for decision making.
Data recipes
Predictions recipes
Decision recipes
When combined, these parts form a decision node.
A decision node is a collection of data, prediction models, and decision models, capable of making decisions without human involvement. There are different types of decision nodes—such as problem nodes and task nodes—that have different roles within a decision platform.
Diversity in models drives better outcomes
No single model is right for all real-world decisions. Diversity enables a system to adapt to novel conditions and improve outcomes. With each addition, the Decision Marketplace gets smarter.
Testing new ideas needs to be fast and cheap
Making models work in the real world requires getting real-world feedback.
Traditional R&D cycles are slow and expensive
For many real-world tasks, taking machine learning models from ideation to production often requires multiple teams and takes months to years.
Decision Marketplace R&D cycles are fast and cheap
Using the Decision Marketplace, a single person can go from ideation to production in minutes to days.
We test ideas in the real world
Rather than debate theory, Decide’s analysts test their ideas in the real world. A typical day might include:
FOR ADVERTISERS AND PUBLISHERS
Better outcomes.
Less work.
Gain access to an ever-growing marketplace of intelligent models. With each addition, the Decision Marketplace gets smarter.
A future-proof solution for a cookieless world
Stop clamoring to PII data.
Gain access to models that respect people’s privacy while hitting performance goals at scale.
Better ad selection for advertisers
Stop using humans to eyeball ad targeting.
Gain access to a diverse set of state-of-the-art models that compete against one another to increase ROI.
Better yield for publishers
Stop undervaluing inventory by having just a few demand-side models assess its value.
Gain access to a diverse set of state-of-the-art models that compete against one another to maximize yield.
Less work for everyone
For real-world tasks like ad selection and pricing, intelligent software systems are becoming capable of solving problems faster and cheaper than human experts while also delivering superior outcomes. This enables all of us to work on new and more interesting problems.
Get in touch
