Want to Win at AI? Aim High, But Start Small
Brand leaders are starting to recognize that they are no longer competing on the basis of products or services alone. Increasingly, their prime battleground is consumer experience. In a recent survey, more than half of consumers said they would likely switch brands if a company didn’t personalize its messages to them. Consumers expect brand engagement that is seamless and customized, regardless of the channel they’re using.
The key to meeting these consumer demands is artificial intelligence. AI driven tools that turn data on consumer behavior into actionable predictions are the only way to achieve personalization and unified engagement at scale. Many leaders realize this, but they don’t know where to start. Eighty percent of marketing executives say they’re confident that AI will revolutionize their industry in the near future, but only 10 percent are using the technology today.
Much of this disconnect stems from a sense of overwhelm. Executives assume they need to invest huge sums, collect every possible piece of data and hire an army of data scientists to get started with AI. This all-or-nothing approach leads to delays in implementation, as competitors speed by. Instead, I advocate the “crawl, walk, run” approach: start small, make an impact quickly and build up to tackling more complex problems. Here’s a roadmap for the pragmatic executive seeking results, not just grand aspirations:
Do Your Homework
There’s no need to reinvent the wheel. Before you get started, do some research about how peers or similar companies have already successfully used AI to tackle business challenges. But don’t stop at reading the splashy media coverage. That image recognition app may have created buzz, but did it really create value for the business? Dig beyond the headlines to discern whether an AI project was a one-off pilot that fizzled or a sustainable solution to a tangible business problem.
Pick One Problem To Solve
Once you’ve done your due diligence, focus on just one use case. Ideally, this is a problem or opportunity in your industry that has benefited from an AI solution (as uncovered by your research) and one that can be solved in increments, not one requiring a massive investment of time and effort up front. Every industry has such problems; keep looking until you find one most relevant to your business.
Get Your Data Into Shape
Armed with a clear idea of the problem you want to solve, the next step is to go out and get data. But your AI solution is only as good as the data you feed it: As the saying goes, “Garbage in, garbage out” As you collect data, think about the following factors to make it as useful and effective as possible. But remember: Don’t wait for all of your data to be in place before you start on the problem. Focus on the minimum data you need for your initial use case first. There’s no point in generating loads of pristine data if you don’t do anything with it.
- Ensure your data is clean. Data that is machine-generated, such as tracked visits to a website, is usually in good shape. But when you’re relying on a human to enter information, the data will inevitably be riddled with errors. For example, if you ask for a birth date from a drop-down menu, you’ll find an inordinate number of people born on January 1. If you don’t clean up your data, those errors will persist and produce a nonsensical model.
- Watch out for biases. AI systems perpetuate the biases inherent in the data used to build them. For example, if lenders historically favored people of a certain race or socioeconomic background, that data can create bias in algorithm-based loan approvals. Beware of skewed data so you can correct for it.
- Prioritize security. We’ve all watched massive data breaches damage brand reputations, and consumers value trust more than ever before. Build security and transparency into the process of collecting, storing and using data from the beginning to avoid problems down the line.
- Governance. When it comes to data, both agility and security require good governance. A data analyst who wants to make decisions quickly needs to understand where data is housed, who owns it, which permissions they need, and whether it is the latest version. Clearly defined roles and procedures will prevent AI projects from get bogged down. Likewise, solid governance ensures that only those who need to access to a certain dataset can get to it.
Apply AI and Take Action
With data at hand, put a data scientist and a business expert on the job to build a minimum viable approach together that is both technically sound and integrated with the current business processes and workflows.. The focus should be on building “just enough” so that you can deploy the solution in practice and see how it survives contact with the real-world. This will give the team invaluable feedback on how to improve the approach. Iterate in this manner until you have a solution that’s working well and demonstrating value. Only then should you complete the build out of the solution and dot the “I’s” and cross the “T’s.”
Once you’ve demonstrated success with the first use-case, widen your scope and repeat the cycle. Enlarge the team, identify multiple problems and iterate on them concurrently. Starting small and iterating allows you to take a high-velocity approach to AI that pays off immediately and allows you to improve over time. In parallel, look to adopt products and tools that already have AI capabilities baked in. This allows you to benefit from the technology and remain focused on driving success in your company without having to be an expert or reallocate resources.
The “crawl, walk, run” approach can help executives get over the roadblocks to adopting AI to start reaping the benefits. By choosing an important but manageable problem, collecting and cleaning relevant data and applying actionable AI insights, you can quickly gain an edge on competitors. Meanwhile, you’ll be building your company’s AI muscle to take advantage of transformative opportunities in the near future.
About the author:Rama Ramakrishnan is Senior Vice President of Data Science at Salesforce Commerce Cloud, formerly Demandware. Rama founded CQuotient in July 2010 and had served as its Chief Executive Officer until it was acquired by Demandware. Prior to CQuotient, Rama taught analytics in the MBA program at MIT Sloan School of Management. Prior, Rama served as Chief Scientist and Vice President of Product Development for Analytics Products in the Retail Global Business Unit of Oracle Corporation. He holds a bachelor’s degree from the Indian Institute of Technology, Madras, an M.S. and a Ph.D. from the Massachusetts Institute of Technology.