Generating Value from AI: A Five-Step Program for Media Companies
With some up-front thinking, tight alignment with business objectives, strong data hygiene and careful governance, content organizations can move AI from the sideline to the business core and deliver on the technology’s lofty expectations.
Media industry leaders have heard the hype around artificial intelligence (AI) — now they want to see results. They recognize the need to reduce costs, create engaging consumer experiences and drive higher revenues. AI’s potential to help achieve these goals (even amid the COVID-19 crisis) has been widely touted across the media value chain.
But while some media organizations have made great strides with AI — most famously Netflix, with some estimating that its AI engine is delivering value in the billions — progress is far from uniform. Most media businesses that are struggling to accelerate productive adoption of AI face a number of barriers. We’ve tapped our experience working with major businesses in the industry to offer recommendations for overcoming these barriers.
Identify cases that can deliver value quickly and that you can build on incrementally
Return on investment (ROI) is important, but speed is critical. One common mistake that organizations make in selecting the problem to solve with AI is to pick the most complex one — after all, the reasoning goes, AI is a tool for tackling complexities that the human mind can’t get a handle on.
But that doesn’t necessarily maximize the payoff. There may be a simpler problem which, if solved by bringing in AI-enabled automation or by leveraging an AI-enabled technique, would generate more value sooner. So it’s wise to choose problems based on envisioned value rather than to focus on complexity.
ROI must also play a role in decision-making. The cost of the solution shouldn’t exceed the value of the benefits, so a precise grasp of expected ROI is helpful. One way to select the right projects is to factor in a higher cost of capital in ROI models, as this will emphasize those that deliver value earlier while also reflecting the higher-level failure risk of AI projects. With AI, business validation is needed to decide whether a given insight will be helpful or not; some projects are scrapped after a technically successful pilot if business leaders cannot leverage the outcome in the way they want.
Always involve the business
Staying connected with the business throughout the lifecycle of an initiative is essential. Trade-offs and decisions will be needed along the way, so it’s critical to understand what the business needs to achieve; how important the problem is to the business; and how quickly the solution is needed.
Understanding all this and agreeing to the way forward will help secure the necessary investment — and even increased funding if early results exceed anticipations. Managing expectations is crucial too; AI cannot guarantee 100% automation or 100% accuracy, so it’s important to define a tolerable margin of error and how the business plans to resolve errors or exceptions. Such an understanding will clarify where, how and to what extent AI can help the business and where leaders might need a fallback plan.
Have the right data, at the right time, ready for analysis
Any savings that an organization can gain using AI can be eroded through costs incurred to feed data into an AI engine. That data must be of the right quality and trustworthy, and its inevitable biases must be well understood and mitigated so they don’t undermine the organization.
The right data should be available at the time it is needed (stale data can generate an insight, but that insight may be of no value). And it should be prepared, categorized and classified in a way that makes it readily analyzable.
While media businesses today are flooded with data, this data is often unorganized, unclassified, of poor quality, or out of date. So investment in modern data platforms and data management capabilities is critical. An enterprise strategy must combine the data (via a data lake or the equivalent) and allow all areas of the business to access it freely. For example, a customer’s history on a direct-to-consumer platform is useful for product managers to understand what features are preferred; for customer service to know what the customer has done on the platform; and for marketing to create personalized campaigns.
Start small, fail fast, be nimble
The phrase “first time right” does not necessarily apply to AI, especially regarding predictions and forecasts. Achieving acceptable accuracy might take a number of iterations and continuous course corrections. Failure, then, must happen fast so the business can learn what to correct.
Since the stakes are high and there is always risk of failure, it is also important to start with a smaller problem (or a subsection of a large problem). This reduces risks associated with failure. There’s no shame in dropping an idea and rethinking the approach. In fact, that willingness to rethink is vital. If the viability of a solution is in doubt, persisting with it — and by doing so wasting time and money — is never the right way to go.
Rather, we advise businesses to course-correct or, in some cases, drop an idea altogether and pick up with a new one. AI is not a magic wand; it cannot solve every problem. Once a smaller problem is solved and the business sees its value — and associated ROI — the solution can be scaled up.
AI projects are different from IT projects — so measure outcomes, not outputs
IT and AI projects are inherently different. IT projects start with a clear idea and a set target. AI, in contrast, is mostly used in the quest to understand the unknown. It’s therefore impossible to know what the output will be ahead of time.
AI must be tuned, monitored and modified over time. Success may require more iterations than originally planned, and a solution may not deliver the accuracy or automation initially imagined. Success should therefore be judged by the degree of impact it creates and how much value that yields.
For example, a media organization spent eight months exploring multichannel advertisement impression data from 112 drama series to identify three impression pattern types. When advertisement impressions on 55 comedy series were explored, the same three impression pattern types came to light, proving that the segmentation criteria were able to scale and were generally applicable across different genres.