AiThority Interview Series With Uri Kogan, Vice President of Product Marketing, Nuxeo

Tell us about your journey into Artificial Intelligence? What made you join an ECM and DAM Company?

I joined Nuxeo because I saw a company with the technology, attitude, talent, and resources to disrupt two relatively staid markets – ECM and DAM.

Nuxeo’s modern Content Services Platform appealed to me as the next generation in information management, offering the flexibility and scalability to address the complex content management challenges facing today’s companies.

I was initially drawn to artificial intelligence (AI) because I saw its potential to help make unstructured content easier to categorize and classify, surface relevant context for users, and also to enable content-related processes to be automated and streamlined – while ensuring a greater level of accuracy throughout the entire process.

Define your ‘Ideal Customer’ profile? What geographies are you targeting for Nuxeo in 2018-2022?

Our typical customer profile is a larger organization – usually within the Fortune 1000. Nuxeo’s primary sectors of focus include financial services, CPG, retail, media, tech, and government. Our geographical focus is global, with a targeted focus on organizations in North America, Europe, and Japan.

What are the core tenets of your business strategy focusing at the AI- tech industry?

Nuxeo looks to utilize AI within our Content Services Platform to help organizations automate manual content-centric processes and surface the most relevant content given a user’s context. We believe that AI offers tremendous potential to provide knowledge workers with the right information at the right time to do their jobs, and to deliver content in exactly the right context to what they’re working on.

In 2018-2020, what are the biggest challenges for Enterprise Content Management and DAM platforms? How could AI/ML technology solve these challenges?

ECM and DAM are part of a second generation of enterprise software. Whereas the first generation automated the transactional back office – accounting, supply chain, and so on – the second generation promised to transform the unstructured processes that form the bulk of knowledge workers’ time. While traditional ECM and DAM have helped accelerate some tasks, like bringing paper-based processes online, they haven’t fully delivered on their early promise.

Despite all the IT investments that have been made, most employees in big organizations still spend hours every day triangulating between files, contextual information, and communication across many different systems, instead of having content and context at their fingertips to quickly make the right decision or add their expertise to a process.

AI can help to relieve some of the drudgery of manual processes, enabling information workers to spend more time on high-value activities.

How do you build new methods of behavioral data science applied to your brand management models?

We are using existing building blocks, such as classification systems, to create new methods, such as automatic routing, auto metadata filling, and so on. This approach uses core algorithms that already exist with our own, brand-new applications.

One basic problem, for example, is how does the system surface the most relevant content? It’s about interpreting the context. You can do that either with business rules (e.g., other projects that you own, other files you’ve liked that have 50% or more of the same words in the title), but these rules are difficult to write well. Machine learning techniques can achieve results that are at least as good, at scale, much less expensively.

A second problem is surfacing processes. Traditionally, business processes were explicitly defined: step 1 leads to step 2; approval then leads to step 3; disapproval sends the process back to step 2. But reality is often more fluid. Instead of an initial approval or rejection, someone might ask for more information. But if that loop isn’t explicitly defined in the business process, the system breaks, and starts leaking value instead of creating it. Perhaps worse, the reporting it generates is misleading if people aren’t actually using it as intended. With AI, you can leave users more freedom, and let processes emerge from their behavior instead of defining them explicitly up front.

When work is happening in the system, AI should be able to interpret the patterns and determine when a document needs a decision. The system will infer, based on the data and previous actions, what to do now. When a new document arrives, for example, maybe it will pre-fill all fields, then leave it to a human to decide to whom the document should be routed for review. The actual routing is an open question, but we know that we need to route to someone.

How do you make AI deliver economic benefits as well as social goodwill?

The real question here is, “How we make sure AI doesn’t replace us?” I was originally trained as an economist, and one of the most important lessons of that field is that in the long-run, the only way to improve living standards is through productivity growth driven by technology. Moreover, technology disruptions are as old as the invention of the fire and the wheel. Trains replaced horses, electricity replaced candles (and much else), personal computers replaced rooms full of workers doing calculations. At the same time, all of those brought new jobs and an overall improvement to our productivity as a society. Trains, generators, and computers need to be designed and built, and they need to be operated. Now, we can reach places faster, carry heavy charges, and have factories running better, or manage complex systems and calculations. Seeing the benefits, would we resist electricity just because it would destroy the candle industry?

I see this more like the introduction of electricity in society – not as an industrial revolution. One assumption is that this new evolution will arrive faster, and destroy more jobs, than any previous disruption. This might be true for the scale, but job replacement by new technologies is not new. Still, this would be totally overshadowed by the broad societal benefit.

I see AI as a tool to improve workers’ performance, not to replace them, the same way that computers allow me to do calculations much faster, leaving me with time to focus on the strategic and creative parts of my job.

What are your predictions for the AI market in 2018?

We are now seeing the integration into platforms of low-hanging fruit: easy-to-reach results on known and tested problems. These are text and image classification, or recommendation systems. As companies start to collect more data, and learn internally what can be achieved with AI, we will start to see new experimentations with automation of process and intelligent interaction.

On the academic front, things are moving very fast, but at the DAM level I see more intelligent use of already existing technologies, and not so much integration of completely new algorithms. For ECM and DAM, it’s about going beyond auto classification to other areas.

What is your vision in making AI technology more accessible to local marketing communities and agencies? Do you provide any teaching or learning programs for Content Managers and B2B marketing communities?

Our open architecture enables us to be open and flexible to continue to incorporate AI technologies and community input.

What technologies within AI and computing are you interested in?

A toddler, upon pointing at a chair and asking what it is, will generalize this new information and thereafter recognize other chairs (more or less). In AI this is called “one-shot learning” – being able to classify images and texts from only a few examples, rather than from hundreds or thousands. Toddlers can do this because through observation they have created a model of the world – what is called “unsupervised learning” – that lets them infer a model from only a few examples. To us, this development offers the most potential benefit, and is proceeding in two directions: the unsupervised model, which is complicated; and the transfer learning model. In transfer learning, we learn a model for a particular task where we have ample data, and afterwards we adapt it to our particular model, using much less data. The main fields for this are images and text; images are much more advanced at this stage.

In general, we’re also interested in Natural Language Processing (NLP) and concept disambiguation / representation. Currently words are just a feature, or a vectoral representation. We do not have a connection with concepts that humans understand. This work might produce new knowledge structure in the processing pipelines, or simply better representations of words. Managing this will allow us a more generic system, and the possibility of truly conversational agents for questions and answers. In particular, connecting images with text – images that explain what kind data you are looking for, be it text or images, instead of giving words that you believe to be in a document or a tag.

As an AI leader, what industries you think would be fastest to adopting AI/ML with smooth efficiency? What are the new emerging markets for AI technology markets?

I see the media segment as being right on the leading edge in effectively adopting AI technologies. Retail is also growing as a leading sector for new AI technologies.

What’s your smartest work-related shortcut or productivity hack?

I use fragments of functionality from different pieces of software together to deliver what I need. Creating a table with merged cells is tedious in PowerPoint, for example, so I do it in Excel and then paste it to PowerPoint. Another example involves converting a list of bullets in an email into rows in a spreadsheet. Instead of copying/pasting leads to bullets in each cell, which is inefficient, I paste them into Word, which converts text bullets into “smart” bullets. Then I clear the formatting in Word to get straight text. Then I paste into text editor to make sure there is no funky Word formatting left. Finally, I paste that into the spreadsheet.

The lesson? Even at the individual level, we use a multiplicity of tools to convert information of one sort into another all the time. Systems that connect these dots more intelligently will become the business operating systems of the future.

Tag the one person in the industry whose answers to these questions you would love to read:

I think some of the consumer-facing companies that have massive customer bases (and thus more data to learn from) are probably doing the most interesting work here, so I will tag a friend, Deepak Tiwari, Head of Data Products, Product Management at Lyft.

Thank you Uri! That was fun and hope to see you back on AIthority soon.