GenAI applications span several areas, each requiring different levels of rigor and investment.

GenAI’s ability to canvas and synthesize information is already helping manufacturers jumpstart the creative process, whether it’s brainstorming new concepts, expanding on ideas, or fine-tuning product details and messaging. When used effectively, GenAI has the potential to accelerate product innovation cycles, pulling from a vast repository of cross-category insights that help manufacturers come up with their next innovation even faster.

“We’ve seen remarkable progress with the capabilities of these models, even from only five years ago. Yesterday’s models were difficult to customize with our own research and expertise and generated less than stellar output. Today, we’re conducting our own in-house experimentation to develop solutions that will help accelerate the innovation process — with promising results.”

Mark Flynn, Head of Product, NIQ BASES

Which means it’s unlikely that AI will replace innovators’ jobs — instead, innovators who embrace AI as an enabler to do their job more efficiently and effectively will be the ones who flourish.

But as with any new business initiative, before diving headfirst into the GenAI landscape to develop your next great innovation, you must first lay a solid foundation to ensure your organization is set up for success.

The following questions can help guide your AI task force toward solutions that are safe, effective, and a good fit for your business needs.

Many companies are jumping onto the GenAI bandwagon simply because the technology exists and everyone seems to be using it. This phenomenon is not a new one: A 2019 survey by venture capitalist firm MMC found that 40 percent of European startups that are classified as AI companies don’t actually use artificial intelligence in a way that is ‘material’ to their businesses.

With so many businesses rushing to capitalize on GenAI’s popularity, organizations must perform their due diligence in vetting potential vendors, not only on their application of AI, but whether they’re truly using AI at all.

Don’t let hype drive your decision making. Instead, ask yourself: What problem am I trying to solve? And what’s the most direct, simple, and straightforward answer to this problem?

In many cases, a statistical or other, non-generative machine-learning model may be more appropriate. If you’re summarizing comments from your customers, GenAI is a great solution. But if you’re trying to forecast sales or use it for data extrapolation, machine learning and regression toolkits might be a better fit.

And even certain use cases, like text creation, must be followed up with validation. GenAI is very good at summarizing information you provide because you can control the input and how you want it summarized. But if you want GenAI to create new information for you, your AI team must have the data to verify whether the information the model provides is accurate.

So don’t forsake your current analytic tools that are already working well. Instead, explore what new use cases GenAI opens for you to innovate better and faster.

Now that you’ve established your use case for GenAI, you’re ready to explore solutions. Asking the right questions allows you to make educated decisions so you can get the most out of your technology investment and keep your data safe.

Just as you would vet other types of R&D modeling research, you’ll apply similar standards and best practices to AI investments. The following questions can help you find the right fit.

Ideally you would build models that are incredibly accurate and easy to scale to new use cases, geographies, and trends. However, it’s more likely that you’ll need to make tradeoffs when picking the right GenAI investments.

Generalized models have become so advanced that you probably won’t need to build a model from scratch — your AI team can use domain-specific training to customize the model to your use case so you can apply accurate, relevant industry data to solve your problem.

Technically, you can customize a model with any type of data — but is this worth the time and cost? Ultimately, it requires a balance between accuracy and scalability.

For example, perhaps you have a certain tone that you’d like to use in your ad copy, so you’d like to use a custom-trained LLM to give you an output in that tone. But what if you work across several different brands and countries? Would the tone for ad copy promoting laundry detergent in the US be the same for promoting haircare in Japan?

While you might be tempted to ask your AI team to train separate models for each of the brands and countries that you manage, this would become time- and cost-intensive, resulting in diminishing returns. You may get a slightly more final product in doing so, but you’d likely lose the efficiency that you were hoping to gain through using an LLM in the first place.

When investing in these models, you’ll need to make conscious decisions about whether it’s worth customizing to your exact use case instead of using more generic models and then curating how you input information and utilize the outputs.

LLMs make logical predictions for each word, choosing what’s most likely to follow based on learned patterns from its training data. But they have the potential to generate inaccurate or nonsensical outputs — a phenomenon called “hallucination.” Because these responses are conveyed with confidence, they appear to be true and are therefore difficult to detect.

Hallucination is when a language model produces information or data that is not grounded in reality or factual accuracy, effectively “making up” details that may sound plausible but are not true.

The risk of hallucinations can vary according to use case. For example, GenAI used for text-based data summarization tasks is less likely to hallucinate, but if it’s generating new information for you — like new benefits or claims — you must validate the outputs to confirm accuracy. Similarly, with the latest enhancements in technology, users can call upon certain functions to execute numeric data analysis, but using LLMs directly for complex numeric data analysis is not recommended.

Indeed, the technology is rapidly advancing, as evidenced by the 40% improvement in generating factual responses between GPT-3.5 and GPT-4. However, a cautious approach is crucial. Inquire whether and how your AI team can provide data to verify the accuracy of their outputs. Reliable data sources, primary research for validating AI outputs, and the capability to constantly monitor accuracy are industry best practices. If you are unable to verify whether your model’s output is correct … should you be using it?

Protecting your data and IP is critical when using GenAI applications. You likely already understand the importance of protecting against unauthorized access to your data. However, you may not be aware that you could actually be providing access to your data when you use GenAI or work with vendors to build AI models.

There are a number of different ways to access GenAI. In fact, there is a proliferation of vendors producing plug-ins and other types of easy interfaces that allow users to simply paste in data and receive something back — but it’s not always clear what the guidelines and privacy considerations are. Using a free, open source LLM, like ChatGPT, means anything that you input as a question or prompt can potentially be used for future model training. In fact, Google DeepMind recently published a paper about its ability to extract training data from ChatGPT, raising new questions about data security.

Without due diligence, the information you share with an open source LLM — including that proprietary new concept your team has been developing — could ultimately be provided as part of an answer in a future model to anyone, including your competitors.

To protect your information, we encourage investment in safe ways to leverage GenAI — for example, through a premium membership, or a “walled garden,” which contains your information and keeps it inaccessible to the public or any other third-party. Be sure to ask potential vendors how your data and the information entered into prompts are being stored, shared, and used.

If you simply prompt the model to come up with a great product idea, you’ll probably come up short (mosquito-repellent roast potatoes and cookie-vegetable stir fry aren’t likely best sellers). Similarly, if you put in a generic prompt, you’ll receive an equally generic idea. However, if you craft a specific, insightful prompt leveraging foundational research, then you can enable the model to pull together a unique combination of innovative and interesting ideas. Let’s illustrate with an example.

A manufacturer wants to create new, innovative packaging for dish soap and hopes to generate ideas by comparing packs in other categories. LLMs can manage this kind of analogical thinking with the right prompts, but at a much larger scale. With a vast repository of data, it has the ability to make infinite combinations across industries and categories without getting tired. What if the manufacturer asked the LLM to generate dish soap pack ideas inspired by breakfast cereal boxes? Or ice cream cartons? By combining the right prompts using foundational research — for example, your brand strengths, personas, and category knowledge — even seemingly off-the-wall comparisons can lead to interesting insights, which you can keep running in the prompts until you’ve struck gold.

Note that this exercise is meant for using GenAI as a tool for ideation and inspiration, not for exact outputs or fully-baked ideas — especially when pulling from open data repositories like the internet. Remember that, on its own, an LLM doesn’t come up with truly novel ideas. In addition to some of the aforementioned watchouts about bias and hallucinations, using exact outputs also risks potential intellectual property conflict. An important role of your AI task force will be to monitor the latest legislation around this topic and determine best practices for effectively using GenAI

Ultimately, GenAI will change how you innovate — but it won’t do all the work for you. For now, it’s safer — and more effective — to use it as a tool for inspiration with honed prompts that play to your strengths and help differentiate you from the competition.

Want to learn about our AI exploration in the product innovation space?