Watch all the Transform 2020 sessions on-demand here.
Whether it’s self-driving cars, robotic hotel concierges, or Amazon’s intelligent delivery drones, it’s easy to see why business leaders are aflutter about the promise of artificial intelligence. Here’s what most don’t realize: As exciting as AI and its potential may be, it’s only as good as the data and training methods used to fine tune it, and that process currently requires humans.
In other words, all AI needs human input to function successfully. And that’s why there is no reason to fear that AI will replace people.
We are the humans that make AI successful
The AI behind something as simple as the auto-suggest on your queries in Google is getting assistance from a human — you! You are supplying the data through your responses or non-responses. Your decisions train the Google algorithms.
A more complex example is when Facebook asks you to tag the people in your photos. Facebook won’t automatically name them for you, but it will take a guess about who they are. When you say “yes,” you’re training the Facebook algorithm. And when you say “no,” you’re also training the algorithm. Guided by human input (yours), Facebook’s AI engines work their magic to build suggestions not only for the people you know but the people who know you.
June 5th: The AI Audit in NYC
Join us next week in NYC to engage with top executive leaders, delving into strategies for auditing AI models to ensure fairness, optimal performance, and ethical compliance across diverse organizations. Secure your attendance for this exclusive invite-only event.
AI in the enterprise
A recent Gartner report predicted that AI technology will be a part of most software products by 2020. Human input, also known as human-assisted AI, can also work to guide AI’s handling of the complexities of big enterprise data.
One of the challenges of B2B is that the large quantities of high-quality, properly tagged data necessary to pull off unsupervised AI are just not as available as they are in consumer settings. And there are many highly resource-intensive IT systems that enterprises use to track their data and tag it in different ways, which makes it difficult to combine the intelligent systems. Also, a lot of key enterprise data from unstructured sources is curated in spreadsheets and emails and is nearly impossible to leverage without a deep level of manual intervention.
Garbage in, garbage out
Companies’ data sources and documents — their most important commercial relationships — are all too often buried in unstructured documentation or across many fragmented IT and business systems, which are incomplete and infrequently updated. When humans step in, they need to reduce the confusion and complexity by organizing and standardizing the data to make AI work. Otherwise, AI applied to bad data only creates more garbage.
For example, in many of the enterprises we work with, such as NCR, CenturyLink, HPE, and Novelis, executives want to use data and analytics to reduce customer churn and maximize revenue yield. The companies need to know quickly when individual contracts are coming up for renewal to stem the exits. They also need to ensure that all of their pricing matches up with what they agreed to in contracts. Should be easy to pull up the expiration dates and pricing rules, right?
As it turns out, finding their customer contract expiration dates and active pricing is like finding a needle in a haystack. It needs to be derived from a complex sequence of data that often involves invoice dates, delivery dates, and signature dates, which exist in multiple documents or systems. To make matters worse, there is a possibility after all of this is done that the offering or the terms of the original customer contract have changed.
This is not an isolated example of the data mess facing most companies founded more than 20 years ago. Most negotiated B2B relationships don’t have something as simple as a visible, easily found contract expiration date — a variety of manual processes are used to try to accurately maintain that data. If you don’t have that information available, how can you drive the sequence of actions required to retain and grow revenue with that customer? You can’t.
So where does AI come in? AI approaches can be used as a method to acquire and organize the data, and can then be used as a method to leverage the organized data to drive results. In our human-assisted AI approach, AI tools are used to identify valuable information inside contract families and related document sets, and then an integrated team of experts curates the resulting data to ensure 99 percent accuracy.
Without an integrated team of experts, the AI methods would not produce data with the level of accuracy and actionability required to drive a business objective. Without leveraging AI, the team of experts would be too slow and cost-inefficient to get to the data in the right time frame and cost point. So both need to work together to get to a valuable outcome, and that’s the brilliance of human-assisted AI.
Less data can be more
In human-assisted AI thinking, what’s important is to identify the right subset of the torrent of data to work. Back to the Facebook photo tagging example — the ask from users is very specific. If the company asked too much of its users, the ability to tune the algorithm would be limited, and therefore so would its usefulness.
To drive real results for particular revenue growth or sales objectives, a company needs only a dozen or so data points, but the information must be complete and available with 99 percent accuracy. In fact, it’s often better to have 10 to 12 key data points at 99 percent accuracy than 30 data points with lower accuracy, since inaccurate information requires too many manual steps to be useful.
Working together with AI systems, human experts can identify and curate those critical key data points and translate them into actionable information so business teams can easily exploit every opportunity or revenue moment as it happens.
AI does have great potential in enterprise with the right techniques. Human-assisted AI can streamline all commercial relationship operations, making it easier and quicker for companies to understand and act upon existing customer, partner, and supplier relationships. AI alone, applied to big data, can miss what enterprises are looking for it to do in the first place: to find and harness critical, accurate, and useful data.
So while the robots won’t be sitting at our desks within the next five years, they may be helping us organize our content.
Praful Saklani is founder and chief executive officer of Pramata, a company that operationalizes the details of commercial relationships for maximizing revenue, reducing risk, and driving business efficiencies for enterprises.
