Building a strategic AI roadmap for your business
Karthik Ramakrishnan Karthik Ramakrishnan
July 30 7 min

Building a strategic AI roadmap for your business

In an era when many businesses are experimenting with AI — and a significant number of AI projects are failing — it’s never been more important to have a plan. Building an AI roadmap means so much more than simply trying and failing. It’s about discovering, analyzing and prioritizing your AI investments to deliver business value.

An AI roadmap is a portfolio of vetted AI opportunities prioritized to achieve strategic business goals over the short and long term. It’s the first step to having a solid strategy for AI, but it’s not a comprehensive plan to transform your whole enterprise. Instead, it kicks off the sequence of work laid out in the four pillars of intelligent AI adoption: Strategy, Data and Technology, People and Organization, and Governance.

There’s something to be said for the experimental approach to adopting AI. It’s impossible to see every AI opportunity ahead of time, even for teams who are skilled at judging good opportunities from bad. By just jumping in, you learn fast by failing fast. But failing fast doesn’t set you up for long-term success.

Learning by experiment is a critical supporting tactic for adopting AI, not a strategy. An AI roadmap is that strategy, allowing you to plan and choose the best tactics for intelligent AI adoption at your organization.

Through our work with clients in multiple industries over the last few years, we’ve refined a repeatable process to build a roadmap quickly over three activity phases:

  1. Discover AI use cases.
  2. Analyze AI use cases and capabilities for impact, effort, and risk.
  3. Prioritize AI use cases and capabilities, given dependencies and complements in the business plan.
    AI roadmap process.

    Discover AI use cases

    In the Discover phase, the outcome is to build a portfolio of AI use cases to evaluate in the subsequent phases.

    Discovery is about looking beyond the mechanics of AI, the “how-to” questions, and instead looking at the business opportunities. What can your business do better by using AI? What can it do that’s new? What value should you create next, and why?

    To begin, the ideal place is to scope the work at the level of a business line — not the whole business, nor a single process or task. The former is too big to create meaningful plans, while the latter is too granular. At financial institutions, for example, we would look at use cases in wealth management or credit lending rather than banking as a whole.

    After choosing your focus, educate change leaders on how to recognize a good AI use case. A use case should have data that describes both the input and expected output for a business task — like the detailed product descriptions available in multiple languages that enabled eBay to specialize performance of its machine translation AI using back-and-forth examples.

    Finally, work across teams to design new possibilities, matching AI capabilities (such as natural language processing) to use cases (such as translation of product listings). Ideas do need to be practical, for example: what prediction will be made, using what data, and how will the predictions be applied to create value? However, imagination and ambition pay off, too — among early adopters of AI, more than 60% reported discovering a new business model according to an IDC survey.

    To generate the best ideas, balance top-down and bottom-up insights. Interviews with employees and customers, for example, can help create early buy-in as well as a deeper understanding of on-the-ground business operations.

    At the conclusion of the Discovery phase, your team should have a set of use cases estimated to be of high impact that still require detailed validation.

    Analyze AI use cases and capabilities

    Analyze AI use cases.

    After assembling a set of potentially high-impact use cases, the Analysis step means determining the impact, effort, and risk of each one — paying special attention to AI capabilities that can be reused across multiple scenarios.

    When estimating the impact of an AI use case, look at more than just incremental improvements to accuracy or efficiency. It’s about working smarter. Let’s look at two examples from an investment banking context to help clarify.

    For a portfolio manager who reads corporate reports and analysts notes, an AI capability like natural language processing could provide them with summaries and key information more quickly than combing through everything themselves. The initial metric for this use case could simply be time saved. Another use case might be optimizing a trading strategy using a deep learning model. For this use case, the metric could be an improvement in what traders call the alpha, the active return on investment above the market benchmark.

    These two use cases translate into cost savings for their initial metric, but if a portfolio manager has two extra hours in their day to do other high-value work, then cost savings alone is hardly the only impact. The portfolio manager might be able to manage a larger portfolio, spend more time building client relationships, or other tasks as determined by the distinct business line strategy. Similarly, an improvement in ROI might unlock new opportunities for the business at a certain tipping point.

    Each use case and capability then requires development, integration, and change management. To estimate this effort, consider the use case in the context of Strategy, Data, Technology, People, and Governance. What data and technology infrastructure might you need for this use case to work? In the banking example, do you have the trading data to drive optimization, in the right state for it to be used at the right speed to make an impact?

    Finally, AI systems must be safe, reliable, trustworthy, and accountable. To estimate risks in each of these areas, a system-level perspective is critical. For each use case, what is the potential physical, economic, or psychological harm for employees, customers, and other stakeholders? Are the model and process sufficiently explainable?

    At the conclusion of the Analysis phase, use cases and capabilities have the data necessary for decision-makers to come together around a common vision and plan.

    Prioritize AI projects

    Prioritize AI projects.

    In the Prioritize phase, align your team to sequence AI investments for maximum impact, balancing three goals:

    1. Setting up immediate next steps with valuable, actionable projects.
    2. Setting up bigger wins downstream with a deliberate plan for building and scaling capabilities over time.
    3. Unlocking support and budget by aligning executives and change leaders around a common vision.

    The first two goals are about balancing short- and long-term value. The third is about unblocking progress by securing buy-in from leadership about the realistic costs and benefits of AI investments.

    The work of the initial phases should provide the details for you to use in decision-making, but in this final phase, be sure to build in time for evaluating any remaining critical dependencies and uncertainties. In particular, AI capabilities are reusable in ways that previous technologies are not, expanding the possibilities for ROI for a given project — as well as possibilities for risk.

    Conclusion

    Across four articles, we’ve discussed how to create business value from artificial intelligence by adapting AI to your business and adapting your business to AI. We call this process intelligent AI adoption, outlining the four pillars required to build and scale enterprise AI. Because preparations are time-consuming, because AI solutions need to be fine-tuned, and because there’s limited time to win, you should get started now.

    In this final post, we discussed how to build a roadmap for your AI journey. Taking the time to discover, analyze, and prioritize AI investments can make the difference between getting ahead or wasting time.

    If you’d like help roadmapping your own AI ambitions—and taking them all the way to real applications—click the button below to contact our team!