The 3 Most Common Mistakes in Product Innovation with AI

Some machinery and plant manufacturers are already intensively exploring how to leverage AI for their next product generation. Their goal is to equip their machines with intelligent digital services to better meet customer needs. 

They themselves benefit from digital business models that open up innovative, sustainable revenue streams. Furthermore, the consistent integration of AI into the product portfolio helps expand technological leadership and capture new market share. For AI to become a key driver of your company's development, you should avoid certain common pitfalls. Read on to discover the 3 errors we encounter most frequently when implementing AI projects.

Inflated Expectations Without Clear Objectives

When we speak with executives and senior managers about AI projects, we often find that they overestimate AI's capabilities, particularly regarding its performance in the early stages. This is partly due to a lack of clear understanding about how the technology works. Additionally, when reading many marketing articles on AI, one often gets the impression that AI is a silver bullet—you simply feed all available data into it and instantly gain any desired functionality effortlessly. However, outside of highly idealized laboratory environments, it's not that simple. It requires significant expertise and focused effort to bring your ideal AI solution to life.

Once expectations are inflated, disappointment can be particularly acute if there's no clearly defined objective. Absent or vaguely defined goals prevent proper evaluation of AI results. When an AI project starts with this toxic combination, measurable results don't materialize and no sense of success emerges. What remains is a bitter aftertaste and uncertainty about whether AI can ever truly help your company.

Our tips for cultivating the right mindset to prepare for successful AI projects are:

1. Let AI success stories inspire you, but don't let them mislead you. The supposed AI miracle is the result of continuous effort and doesn't simply fall from the sky.

2. Develop an understanding of value-creating combinations of your existing products and AI. Ask your customers, for example, what additional features would make their lives easier. Look for simple wishes ("low-hanging fruits").

3. Always connect product-AI combinations with potential business models. Only the prospect of a successful business keeps AI projects alive during difficulties. No one persists with technology for its own sake.

4. Derive measurable objectives from the product-AI combinations and underlying business models you've identified, at minimum in the form of hypotheses, such as "the AI will detect more than half of all daily process deviations" or "2 out of 5 test customers will use the new AI service daily."

Data Does Not Match the Objective

You likely already know that the functionality of an AI depends on the volume and quality of available data. However, you can't generalize how much data is enough or how high data quality must be for an AI project to succeed. Rather, this must be assessed on a case-by-case basis, and the data situation may need to be improved before the AI project can begin.

Additionally, the data must fit the use case—that is, the AI functionality with which you want to achieve an objective or generate value. The central question is: "Does my data contain all the information relevant to the AI?" For example, if you want to know how the configuration of setpoint parameters affects the availability of your machine or system, the dataset for training the AI must contain both the setpoint values and the operating state (ON, OFF) over time. However, this dataset alone is still insufficient for reliable AI availability forecasting if, for instance, material supply or the condition of used equipment, such as tools, have a decisive influence.

To prevent demoralizing delays at the start of the project, we therefore recommend:

1. Address your data infrastructure early on: which data do you have in which systems at what resolution?

2. Which data matters for which use case? For example, conduct surveys with your machine operators—they know best. However, avoid lengthy discussions about presumed causal relationships; let the AI handle that instead.

3. Collect data systematically and manage it, where possible, in open databases and IT systems.
   
   

Too Short-Term a Perspective

Many companies launch an AI pilot project without a clear plan for the next steps. In case of success, it remains unclear how further development toward a marketable AI service should proceed and when investments will pay off. And if the pilot results are not fully satisfactory, you cannot assess the scope and effort required for improvements. What should become the actual starting point of your AI future can thus, in the worst case, become a dead-end road.

A well-structured AI roadmap can help you avoid this dead-end road. It makes reasonable assumptions about the expected time-to-market for your newly developed AI service and the development steps required to achieve it. Additionally, the necessary financial and personnel resources can be tied to it. An AI roadmap helps you assess development progress and protects you from ill-considered knee-jerk reactions.

A sustainable AI roadmap should contain at least: 

1. Working hypotheses about the AI use case with which value will be generated, as well as the underlying business model. Both can change during implementation, but there should always be a solid reference point.

2. Development plan from demonstrator to prototype to market-ready product to rollout and scaling. The technical details can initially be highly simplified in later phases, but the objectives of each phase should always be clearly defined.

3. Time, cost, and resource requirements, particularly internal resources and required digital infrastructure. For later development phases, suitable test scenarios and sales activities should also be considered.

Stay in Control at All Times

Even though it concerns AI and this field is likely unfamiliar to you, in our view you'll keep the steering wheel firmly in your hands at all times if you follow a few key points: Ensure that all stakeholders have a clear understanding of the AI use case and expected value creation. Define simple objectives with which you can monitor progress and evaluate AI results. Formulate clear expectations for interim results that you expect in various project phases. Place each interim result in the AI roadmap and make adjustments to project execution or the AI roadmap when there are deviations.