How Adobe moves AI, machine learning research to the product pipeline

Adobe will outline a series of features that will go from research and development to its analytics platform. Here's how Adobe approaches the product pipeline.
Written by Larry Dignan, Contributor

Video: AI boom is creating new business value

Adobe outlined a series of new artificial intelligence and machine learning technologies and features from its labs that it plans to roll into the company's analytics tools. But the more interesting item may be how Adobe thinks about developing new AI-driven technologies and collaborates across functions.

The tools break down like this:

  • Intelligent Forecasting, which will aim to allow customers to address business metric shortfalls from real-time data. Adobe's intelligent forecasting effort aims to crunch billions of data points to optimize operations for online retailers, companies looking to improve conversion rates, and manage seasonal slowdowns.
  • Predictive Pathing, which will allow customers to spot things like app installs and emerging problem areas. Adobe will analyze the paths a customer takes across screens.
  • Automated segmentation to manage audiences and customer bases across ages and demographics. AI and machine learning will automatically segment audiences.
  • Analysis Workspace Assistant, an AI tool that takes a question and runs them across historical queries so work isn't replicated. The assistant is designed to improve over time.

Adobe has more than 200 PhDs in machine learning, economics, physics, and computer science in its analytics unit. This brain power sits within Adobe's R&D unit as well as within various departments and units.

"We provide a Hubble telescope into data to see much deeper and further," said John Bates, group product manager for Adobe Analytics. We spoke with Bates and executives overseeing design, customer experience and software engineering to get a feel for how Adobe moves AI projects from theory to product.

Read also: Artificial intelligence will be worth $1.2 trillion to the enterprise in 2018 | IT pros are not confident their organizations have the skills to take advantage of AI | Deep Learning: The interest is more than latent | What is AI? Everything you need to know about Artificial Intelligence

Here the lens Adobe uses to view research:

Maintain focus on customer needs. Bates said the collaboration between research, product, design, and software development, and, ultimately, what gets produced revolves around one key question: Does this product bring value to customers? "Our customers come use us for two things from an analytics perspective -- content and data," explained Bates. "Our customers are swimming in data and only using 1 to 3 percent of what's collected. Anywhere from 97 percent to 99 percent of the data is below the covers and customers don't see it on some initial view."

For instance, Intelligence Forecasting is a more conceptual feature, but the value is in allowing users to set a target and then see whether the forecast is falling within range. Jordan Walker, design manager for Adobe Analytics, says when a forecast falls out of range there will be drill-downs to analyze further via a package of Adobe algorithms.


Aim to surface unknown unknowns. From a research perspective, Adobe is focused on finding ways to identify and surface "unknown unknowns," or that data that is collected but not seen, said Bates. This unknown category of data is both unstructured and structured. An example would be anything from click streams on products for an e-commerce company. HP has millions of products it sells and an array of different types. "To look at one million plus products is impossible. To scale, you need machines. Machines to the heavy lifting to surface data that wouldn't be seen with the naked eye and then humans can optimize," said Bates.

Enable citizen data scientists. Bates explained that companies typically don't have large data science teams. The goal of Adobe -- as well as companies like Tableau -- is to "take the everyday user of analytics and allow them to do the work of data scientists," said Bates.


Focus on common problems. Bates said Adobe approaches research by spending time on site with customers and understanding the business challenges they are facing. This effort includes research, product managers, and designers. "We are looking for common problems where we can provide scale for a solution," said Bates. Possible solutions can come from the product team or research.

Look for "white space opportunities." Bates said beyond solving more immediate areas, Adobe's research team is looking three years to five years out at white space opportunities that apply to multiple units in Adobe and various customer bases. Emotion detection was an area of white space research where there was an opportunity to gauge sentiment and couple it with other data points.


UX has to present research and data in a usable way. Research collaborating with designers and product teams represents the art of R&D. Engineers have to present to a non-engineering persona. Designers have to take the algorithms and build narratives for customers. Once some research is identified as broadly valuable, the real collaboration begins, said Bates.

Alexandra Hadley, experience designer for Adobe Analytics, noted that algorithms in a feature such as Predictive Pathing are a great start, but UX is needed to make it all relevant. "Algorithms are produced and plugged into a UI, but the customer has to understand that the data provided is relevant," she said. "We have to move customers through the journey with better insights."

Previous and related coverage

Adobe acquires Sayspring in voice technology, AI push

The startup's voice technologies may be used for voice commands in digital creative products.

Adobe patches critical vulnerabilities in Flash, InDesign

Workstation users should treat the latest Adobe security bulletin seriously.

Editorial standards