Just look at how Microsoft’s commitment to AI features created
Still, some AI features remain beneficial. The challenge for product managers, CEOs, or anyone in digital business is knowing when your product actually needs AI.
Products Solve Problems
I already wrote about the importance of
Successful products do it well, and failing ones do not. It is because of this ability that they become valuable, and a company making them can capture part of that value to turn a profit. There are examples of AI features making meaningful business improvements to increase product value.
Large Language Model (LLM) applications in documenting meetings, writing corporate emails, creating design or code prototypes, improving customer service chatbots, and many other similar applications are successful. The impact of AI features is undeniable and is only likely to grow.
Yet, many generative AI pilot initiatives fail to deliver any meaningful impact. The underlying reason is often the inability to solve problems relevant to users or employees, but even when that is the case, other challenges must be addressed.
- Technical limitations. Most AI features require large, high-quality, and domain-specific data. Proxy servers and web scrapers for collecting such data are costly, and integrating it into existing systems takes time.
- Trust erosion. Even occasional hallucinations in critical flows can destroy user trust. It’s even more important for products in finance, health, and related areas, but all products must ensure that AI features are reliable.
- Data privacy concerns. Sending logs and user data to third-party APIs creates risks of unauthorized access or data collection. Some companies even prohibit the use of AI tools for sensitive data.
- Adoption challenges. AI implementation can simply be a poor User Experience (UX) choice for a product that wasn’t built for it. Developers must consider whether their product and overall brand positioning align with AI implementations.
- Monitoring. Generative AI features in existing products require constant oversight to avoid potential crashes, PR crises, legal risks, or other issues. Evaluation methods must be set to ensure that generative AI works the way it should.
- Unknown challenges. Generative AI is still so new that many of the challenges and issues it may cause are unknown, so there’s an inherent additional risk in implementing it.
Most of these challenges are nothing new to entrepreneurs scaling digital products. Almost any feature, AI or not, needs to solve similar pain points. The difference lies in managing the expectations that come with the AI label. There’s so much hype surrounding generative AI implementations that companies fear being left behind.
Competitive FOMO
Fear of Missing Out (FOMO) applies to CEOs and other decision makers as much as it does to consumers. When ChatGPT was released, many carefully planned strategies were abandoned to chase the visibility of staying ahead.
Competitive pressure for AI features resembles an arms race. One company announces AI integration, and others must respond with something similar or presumably risk their product appearing outdated. Never mind that such a feedback loop might not reflect the needs of consumers.
The calculation seems to be that a critical mass will start choosing products primarily for their AI features. It’s undeniably true for investors, since AI has become the dominant growth narrative in markets.
Any public or venture-capital-backed company now faces pressure to implement at least some AI features. Investors expect AI initiatives, and analysts track AI announcements. Without them, the company seems to be missing out on the most important technological innovation since the Internet.
The same reflects in marketing as companies are quick to rebrand existing automation or machine learning workflows as AI-powered. Such a buzzword helps to drive attention and justify premium pricing, and up to a point, it works.
Why Users Might Prefer Products With AI?
Studies show that people tend to choose products with more features, even if they won’t actually use them. If cost isn’t an issue, consumers almost always choose a product that can do more, and the AI label is the ultimate signal for this.
However, consumers also experience feature fatigue, a frustration stemming from feature overload. Feature fatigue has been known since at least when common household appliances, such as fridges and TVs, became more complicated.
An AI-powered browser, as in the case of recent Microsoft Edge updates, is an easy way to market the product as more capable than competitors. Yet the AI features risk complicating the product's core purpose: browsing the web.
Privacy-conscious users are also suspicious of AI features as just another way to collect user data for the company's own benefit. That’s where the "AI slop" label comes from, and why it means more than just reduced usability.
Some products are trying to reverse the trend, promising no LLM AI features to stand out for their user segment. Vivaldi, a privacy-focused web browser, is doing just that by advertising their intentions to
The approach may be too extreme, as most users will still choose AI-enabled products, at least initially. The solution should be to find the right balance and learn when your product needs AI.
Does Your Product Need AI Features?
We are already seeing that traditional product development frameworks, such as RICE (Reach × Impact × Confidence ÷ Effort), are being adapted to consider the challenges that AI brings.
The technical limitations, user trust erosion, adoption, monitoring, and data privacy concerns we mentioned earlier must be included in typical RICE calculations. It might also be useful to give them more weight by using a multiplier for AI complexity. Doing so forces you to think about AI features as features during the development stage.
Weighing the difficulties of implementing AI does not help with determining whether AI features will actually help increase consumer satisfaction. The Kano model addresses this by defining three product feature types.
- Must-be features are core functionalities of the product. All users expect them by default, and losing them causes an adverse reaction from customers.
- Performance features increase customer satisfaction to a certain extent. Faster load times, more storage, or similar improvements will increase customer feedback up to a point.
- Attractive features help to compete with other brands and exceed customer satisfaction by going the extra mile. Lacking them does not cause dissatisfaction, but their presence might attract a segment of users.
Many generative AI features, at least in the early stages, fall into the performance and attractiveness categories. In e-commerce, customer support chatbots, contextual product summaries, and similar features can enhance appeal or improve performance, but they hardly replace good product selection and UI.
That's not to say some AI features don't have the potential to become must-haves. AI search with at least basic spell check in e-commerce is almost a necessity nowadays. Similarly, AI-fueled web unblockers for online data collection are becoming increasingly crucial, too.
As websites block data access with AI tools, web scrapers need to leverage AI tools to manage proxy pools, browser fingerprints, and other details. Finding such must-have examples in your niche is key to determining whether your product needs AI features at all.
Conclusion
AI feature failures are product management failures. A product that starts with technology and then thinks of what it might solve is bound to fail. The same applies to all AI features. We must ask which problems AI can solve for our users and implement it only when it is a fit.
It’s undeniable that generative AI is a transformative technology, but the real winners will be those who resist the hype and stick to traditional product management frameworks. Other approaches risk getting labeled as just another AI slop.