The World Summit AI put on a fantastic conference in Amsterdam this week. The 2,400 attendees were a diverse mix of C-suite executives, engineers, data scientists, professors, startup founders and students, all interested in the cutting edge of Artificial Intelligence (AI) and what it can do for businesses and their customers.
I was impressed with the breadth and depth of content: from deeply technical talks by machine learning professors to candid case studies by bigco execs. Examples of AI ranged from relatively mainstream automation to futuristic deep learning.
I was invited to speak on a panel alongside Jan Morgenthal, Chief Product Owner for AI at Deutsche Telekom, and Jop Ekering, an AI Consultant focused on financial services clients.
A theme that stood front-and-center on our panel and throughout many other talks, was around best practices for those looking to adopt AI in the enterprise. Companies are aware that we’re in the midst of an AI revolution, and that they have to adopt new technologies and methods quickly if they want to keep up with their competition. But adopting technology with a short track record is also risky, so how should leaders think about the risk-reward?
Here are four takeaways I noted:
1. The uses of artificial intelligence should start small & specific
The age of generalized artificial intelligence, like robots that can solve any problem a human can, is still pretty far away. Companies should start by solving specific problems with a tailored solution. One solution to automate document processing, another to highlight fraud risks, etc. Starting small makes it easier to debug issues when they arise, and to improve algorithms / solutions over time.
RJ Pittman and Japjit Tulsi, eBay’s SVP / Chief Product Officer and VP Engineering, respectively, explained for example how they built and launched their first version of an AI-powered bot in a matter of months. The bot’s only functionality was to ask its user questions that helped narrow down a search using natural language, like "what price range are you looking for", and "what size”? Over time, the bot took on more complexity. Today, you can take a picture of any object, a striped shirt for example, and the bot will instantly search all of eBay to find something that looks similar.
Jan Morgenthal of Deutsche Telekom shared an example where bots were introduced to handle the first line of customer service inquiries, and route customers to actual help desk people depending on the issue. Early on, there was suddenly a huge influx of cases relating to billing questions, which are actually the most expensive help desk reps. The engineering team was able to debug the issue and learn that most of these cases were actually people who had forgotten their password, and were logging in for the first time in many months, to learn about a surprise change in their bill. To the customer, they were calling about a billing issue, but the first issue to fix was actually just to reset their password, so they’d find details of their bill online and spare the time of the support reps. It would have taken much longer to fix this issue is DT had been using a bot with many more applications than to route customer inquiries.
2. Successful companies bet on small AI vendors
In a space with such rapid advancements in technology, companies have to find ways to bet on startups if they want to stay at the cutting edge. One large company talked about a startup whose technology it evaluated once and found not to be accurate enough in interpreting human questions. Eighteen months later, the startup came back with a new method that was found to be groundbreaking in impact and accuracy. This openness to testing startup solutions turned into a very fruitful relationship.
On my panel, I shared my observation that some of the most forward-thinking adopters of new technology have found clever ways to “have a say” in their vendor's’ product roadmaps. Two examples that come to mind are JPMorgan and Goldman Sachs. Both investments banks have entire teams dedicated to testing technology and “doubling-down” in certain areas by (a) investing in their vendors and (b) becoming development partners. Goldman invites vendors onto their trading floors to work on their product alongside the traders or other intended users of the product.
3. Garbage in, garbage out
The importance of underlying data came up again and again. Without clean, structured data, there are severe limitations to the impact of any AI applied to it. Some companies have in-house data teams dedicated to producing training data and identifying false positives and negatives.
I noted that positive trend of CRMs such as Salesforce being used in companies big and small. Work that used to be done in spreadsheets and therefore sat completely disconnected from a company’s database now often sits in cloud-based CRMs, so partners like DataFox can help unearth insights based on a company’s past activity, wins, losses, etc.
4. Develop a base-level understanding of the technology behind AI solutions
In Q&A at the end of sessions, students in the audience often asked for advice for those seeking to ride the AI revolution. Speakers and panelists suggested they get an engineering degree or at least a deep understanding of the business applications of machine learning, NLP, and other aspects of AI
One speaker highlighted the insight that Goldman Sachs’ cash equities trading group, at its height in 2000, had 600 traders in New York, where that group has since dwindled to just two traders. In the meantime, the firm has grown to have over 10,000 engineers, more than Facebook or Salesforce. The Investment Bank is a technology company through and through. Soon, any company vying for market leadership will have to be a technology company through and through.