Keys to Succeeding with AI

Share it:

While hiring smart people and providing a sandbox environment seems an obvious place to begin, real-world AI initiatives gain little from sandbox innovation. An effective AI initiative needs top-down vision, strategy, and tempo. Decide what elements of AI define the future of your industry and link those to strategic capabilities your company will need to vertically integrate. Finally, activate the vision with priorities and tempo.

Master AI Engineering

AI is a vast and volatile landscape that is changing at exponential speed. The rate of change makes it difficult to keep pace with research discoveries and most companies under $1B in revenue should avoid strategies that invest in fundamental AI research. For the majority, it is enough to master AI engineering.

AI engineers apply first principles to untouched domains and make use of available tools and techniques. Successful companies in this class must acquire rigor and sophistication to sustain rapid innovation and to minimize engineering blunders.

With AI, little is gained by dabbling. Proof-of-concept projects can be inspiring but most are ultimately pointless, as they do not lead a company to master AI engineering and industry leadership. Even successful proofs-of-concept that showcase high-value capabilities are meaningless in the absence of professional industrialization or productization. It is easy to develop an impressive feature-level demo, it is difficult to accumulate strategic capabilities.

A company must hone its readiness to transform the entire industry with AI.

Pick One Big Thing

While AI is vast, you can’t boil the ocean with it. Decide a single way in which the industry will inevitably be transformed by AI. Focus your company on delivering this one singular promise before any competitor, in a way that is difficult for followers to imitate.

Begin the process of selecting the one big thing by answering these questions:

  • How will your industry be transformed by AI?
  • Will the transformation be defined by autonomy (self-directing decision agents)?
  • Will it be led by sensory perception (summarizing or encoding real-world environments)?
  • Will it be led by generative intelligence (producing realistic dreams, designs, media, and art)?
  • Which sensory domain is most important, i.e., vision, audio, language, or something more exotic?

These questions matter because you will use the answers to set strategy. Select a single vision that will redefine your industry. Once you have this strong center vision to define your AI initiative, all investments in AI must be aligned and prioritized to support it.

Do not delegate this level top level of strategy. The C-level has vision for the industry. AI strategy must come from the top, not from technical or management ranks.

After a clear vision is defined, priorities and tempo should also be set from the top.

Vision, Priority, Tempo

Allow no room for the vision to grow or branch. An AI vision must be kept simple and targeted. Internal priorities must map to this vision and be kept minimal, otherwise execution will become a hydra of decentralized AI projects. Stress the importance of speed. In matters of strategy, the adversary gets a vote, therefore you cannot be slow or late.

Step 1: Formulate a single targeted vision and keep it simple.

Step 2: Establish priorities that map directly to the vision. Any priority that surfaces which does not directly implement the vision gets dropped from the AI initiative.

Step 3: Focus on tempo. To be a leader and define the industry, speed is important. Hold teams accountable to an uncomfortably rapid tempo.

Expect Big Things

The technical mechanisms of AI are common, and every large company has access to at least one good team. As a rule, if your team accomplishes something in a few weeks, your competitors can easily do the same. Avoid the temptation to be overly impressed by AI progress demos. AI performs astonishing tricks out of the box. A student intern armed with Python and some pre-trained models almost makes it look easy, but it is not easy. Real-world AI, the type that defines industries, is difficult. Experienced AI engineers, not interns, are needed.

Drive your teams to the big picture vision by holding them accountable to your priorities and tempo. At every step, verify that the AI produced is real-world, not a toy or prototype.

Accept Your Data Limitations

This applies equally to AI and machine learning (ML) projects. AI/ML does not care how special you believe your data is. Companies that seriously commit to developing AI/ML solutions quickly discover problems with their treasured data. Whether the trouble is quality, tonnage, consistency, shape, coverage, or governance – there is always a data problem. Data needs to be cleaned, reshaped, and augmented with other sources. Investments in data cleaning, augmentation, and synthesis are not optional and will often dominate budget and schedule allocations.

This is strategically relevant because well-established firms with vast stockpiles of data are tempted to assume an inherent advantage over startups doing AI/ML. However, a startup that accepts from day-1 that they have no stockpile of data to leverage might surpass an established firm because the startup is forced to be creative about data curation and conditioning strategies. Meanwhile, the established firm spends months or years discovering its AI/ML initiative has been crippled by data governance policies or chronic data quality concerns.

Existing tonnage is not always an asset, it can be a liability. Approach your AI/ML initiative as if your data does not exist. This forces you to explore opportunities from the perspective of a startup. To the extent your existing tonnage is valuable, it will be exploited along the way.

Go Fast

To establish an industry-defining leadership position by harnessing AI, you will need to move uncomfortably fast. This means keeping an internal tempo that pushes teams and individuals to squeeze the most from every day, week, and month. The ubiquitous 2-week sprint that keeps rhythm for software organizations is not optimal. Sprint-centric management yields low production node capacities and high cycle times that are exacerbated by the demands of AI. While it is tempting to run AI initiatives using Scrum-based Agile frameworks, it is far better to adopt a sprint-less process framework. Sprints are not efficient packaging for AI engineering workloads, as every investment in planning (even for 2-week periods) becomes process waste.

Select a continuous-pull (Kanban-styled) process framework. Keep the process framework lightweight, shed sources of garbage time, and establish a quick tempo that is rigorously enforced. We recommend having a demo every week to cover the product’s evolving realization of the vision.

In the rush to adopt AI, organizations often fail to anticipate a realistic ROI.

It is easy to get starry-eyed about industry-defining autonomous agents that will one day change everything. It is also easy to get disillusioned by costly AI initiatives where the return side of ROI never catches the investment side.