by Rashed Haq
If the industry is going to achieve the predicted increase in value of artificial intelligence (AI)–$1.2 trillion this year and $3.9 trillion in 2022—companies will need to find ways to scale AI across the business and use it to do more than improve customer experience and increase operational efficiencies. While these are important, the greater potential of AI lies in its ability to help organizations create new revenue streams through new products, services and channels.
What does it mean to use AI at scale? This can mean many things, but in this context, scaling refers to two key elements:
Increasing AI adoption across its user base: AI use in many firms started with a few independent groups testing the waters with proofs of concept and pilot projects, but are now expanding its use across the enterprise. The goal? Having the entire company using AI systematically across all business units and functions, with tightly integrated metadata, governance and algorithms driving strategies and decision-making.
Growing more sophisticated in how data and AI is used: Companies that have scaled the use of AI are leveraging data and AI to become a more predictive enterprise—ensuring they know what’s likely to happen and predetermine how best to respond to it.
Through our work with companies across industries and around the world, we’ve identified seven key ways firms can enhance their data maturity and increase adoption to ultimately scale AI across the enterprise. These include:
1. Use data to drive insight
To get the right insights, you need both good data quality and maturity, as well as advanced analytical capabilities such as AI. It’s important to note that an organization cannot be mature in terms of analytics without data maturity.
Furthermore, it cannot sustain good data quality and maturity unless it actively uses that data through analytics. It is cost-prohibitive to sustain good data quality without actually using it.
Another key element is cultural change and ensuring that the business encourages and enables bottom-up, data-driven decisions. Without it, no amount of AI can help and these investments will be underutilized.
2. Ensure high-quality data
Data that is used for AI must be aggregated and cleansed. Because machines will be creating logic based on available data, it is important that the data be as accurate and timely as possible.
3. Add external data into the mix
Most companies have mountains of internal data that is integral to their data strategy. But companies should look for ways to instrument additional first-party data, such as user behavioral data from the website, as well as integrate third-party data, like retail foot traffic, satellite imagery, social listening, and more, to improve the precision of models and to create entirely new ones.
Related – Meet The Innovators: Publicis.Sapient
4. Make data a business asset
Data lakes were created by IT to house data. Since IT is the primary user, these were structured in a way that limits its usage and accessibility by the business. Since data is such a key element of AI at scale, companies need to transform their data lakes into enterprise data marketplaces, where businesspeople can become active – if not primary – users. This helps create a data and analytics-driven decision culture and improve overall data literacy – helping business users feel more comfortable asking for, and using, AI models.
5. Build and experiment with models as much as possible
The Wright brothers were the first in flight because they developed a way to increase the number of experiments they could run. Their competitors would spend nine months building a plane and flying it off a cliff where it eventually crashed. They would then pick up the pieces and analyze what went wrong. It was the right idea, but it took too long to make progress. Conversely, the Wright brothers built a wind tunnel that allowed them to experiment, collect, and analyze data daily – instead of once every nine months.
Organizations can do the same with AI. By building an AI platform to automate the full lifecycle of projects, they can free AI scientists to conduct more frequent experiments and realize business value more quickly.
6. Deploy, deploy, deploy
It’s simple. To get the most – or any – value from a model, you need to use it. An organization willing to deploy models and learn from them will be able to scale faster. But to be successful, deployment should include comprehensive development, testing, and staging.
AI DevOps and the right architecture are critical for this. Production monitoring, for example, is key to ensure the model is still performing at the same accuracy levels with new datastreams as when it was origianlly tested. Governance should also be automated as part of this testing and deployment phase.
7. Implement model governance across the lifecycle
AI models are very literal, short-sighted, and greedy. Also, the logic that works behind the scenes in AI models is not always easy to understand, since there are limited ways to query it to pinpoint exactly how it made a specific decision. That’s why model governance is so important. Before modeling, companies should use an appropriate or diverse set of objectives and establish a diverse set of goals to ensure you get quality results. After modeling, organizations should use pragmatic interpretability and fairness or model compliance testing. One common approach that has been successful is to think of AI interpretability as enabling the user to meaningfully combine the model’s information with her own knowledge. This approach to interpretation is very useful even though it may not shed light on a model’s inner workings.
To scale AI across the enterprise, organizations must have a platform and process on which they can build thousands of AI applications and uses. The ideas above can provide guideposts as you build these out, which will make it easier to find the next use case, build it, and deploy it. The more of this you can do, the more successful you will become. As Jeff Bezos said, “if you double the number of experiments you do per year, you’re going to double your investments.”
Over the last 20 years, Rashed Haq has helped companies transform and create sustained competitive advantage, through innovative applications of artificial intelligence, dynamic optimization, advanced analytics and data engineering. Today, he is Global Lead for Artificial Intelligence, Robotics, and Data at Publicis.Sapient.