AI in Mobile Apps Is Booming, But Many Brands Still Miss The Mark

AI has quickly transitioned from novelty to expectation in just about every digital product, including mobile apps.

Yet many app developers and brands have integrated AI in ways that aren’t yielding the expected impact.

In fact, Gartner predicts that over 40% of agentic AI projects will be canceled by the end of 2027 due to escalating costs, unclear business value, or inadequate risk controls.

Meanwhile, a separate Gartner report revealed that organizations will likely abandon 60% of AI projects unsupported by AI-ready data in 2026.

These figures are hard to ignore, especially since more and more companies are accelerating AI adoption.

But why exactly are these AI projects running into issues?

Finding the answer to that question sits at the center of Jerzy Biernacki’s work.

As Chief AI Officer at Miquido, he’s led dozens of AI implementations across industries, providing him with keen insights into what makes AI integration successful.

“It’s because most companies start with the technology, not the problem. The conversation typically goes, ‘We need AI in our app,’ and then someone goes looking for a place to put it,” he said.

“That's backwards.”

In our interview, Biernacki explains why so many AI features fail to improve mobile apps, what companies consistently get wrong, and how to build systems that deliver measurable value.

Who Is Jerzy Biernacki?

Jerzy Biernacki, PhD in Computer Science and Chief AI Officer at Miquido, focuses on leveraging technology, especially artificial intelligence, to solve complex business challenges. Since 2018, he has led the company’s AI initiatives, delivering nearly 50 commercial AI implementations across various industries. In his role, Jerzy not only consults, designs, and supports clients but also educates and drives innovation, with a primary emphasis on using AI to transform businesses.

Why AI Features Fail in Mobile Apps

When AI fails, it’s easy to point the finger at the platform itself.

Maybe the model just isn’t advanced enough. Perhaps this is the limit of AI as it is today.

However, Biernacki offers a more sobering explanation:

“The root cause is almost never the AI model itself,” he said.

“It fails because someone skipped the boring work of figuring out whether the problem was real, whether the data existed, and whether users actually wanted this solved differently than what they already had.”

This misalignment creates what Biernacki describes as the “Experience Expectations Gap,” which he defines as “the distance between what a user expects from an AI feature and what they actually get.”

“This gap kills products. In fact, I've experienced it myself,” he explains.

“I once followed the development of an underwater species identification app for months. When it finally launched, it recognized a handful of fish, covered a few dive sites, and the recognition barely worked.”

“The idea was brilliant. The execution created disappointment.”

And that disappointment results in a loss of trust, which may end up in fewer downloads and higher uninstall rates.

The Mistakes That Keep Killing AI Initiatives

Biernacki points to three common mistakes he’s seen come up again and again when it comes to integrating AI into mobile apps:

1. Building AI where it adds no value

A classic (and very expensive) example of this is when developers add AI where ChatGPT already works fine.

“The value is in domain-specific knowledge baked into a purpose-built tool,” Biernacki said.

“If your AI feature does what a user can already do by opening ChatGPT, you haven't added value. You've added cost.”

2. Treating testing as proof of reliability

Most companies and developers test their AI feature internally, get decent results on a handful of scenarios, and ship it.

But just because the AI integration worked in a few controlled test cases, it doesn’t mean it’ll work flawlessly for hundreds of thousands of users.

After all, real users have ways of finding use cases that test teams hadn’t anticipated. This can lead to bizarre, embarrassing, or even harmful outputs.

3. Ignoring compounding failure

On paper, 99.9% accuracy sounds like near perfection, and it definitely is. But if that number was achieved through a small sample size, it’s a misleading figure.

As the sample size grows, that .01% error rate compounds exponentially.

For example, that same 99.9% reliability rate drops to roughly 45% once a few tests become 800 consecutive calls in an AI workflow.

4. Underestimating cost structures

AI doesn’t follow traditional software economics.

The real cost of an AI call is easily 2x to 4x what you'd estimate from the model's pricing page, once you factor in retries, RAG embeddings, moderation, and observability logging.

“Every user query costs real money in tokens and compute, and those costs grow linearly with your user base,” Biernacki said.

“If you haven't modeled this before launch, you're building a product that gets less profitable as it succeeds.”

Fortunately, these mistakes can be avoided simply by asking yourself a few questions:

1. Is the problem real? When you get the AI hammer, suddenly every problem looks like a nail. Match solutions to needs, not the other way around.
2. Can AI solve it better than the alternatives? If the task requires judgment on unstructured data, AI probably adds value. If it follows a clear decision tree with structured inputs, traditional logic is cheaper, faster, and more reliable.
3. Do we have the data? If you don't have the data to train, fine-tune, or provide as context, the best model in the world won't save you.
4. Is the organization ready? Internally, process owners must have the focus and time for change, as well as a plan for what to do with the time AI saves.
5. What's the priority? Plot everything on an impact-vs-difficulty matrix. Start with high-impact, low-difficulty initiatives that generate fast results.

The Underrated Role of Product Design in AI

It’s common to see app developers flaunting their new AI-powered features to their users.

Yet the best AI features are often the ones that feel like a natural extension of what the user was already trying to do. These features should never feel intrusive or confusing.

This is where clever product design in terms of UX and CX plays an important role in mobile app development strategies.

But Biernacki feels that this side of mobile app development is often an underrated and underinvested element.

When designing an AI-powered feature, it’s important to start by understanding how it respects the user's time, money, and energy.

Those are the three most valuable resources any user has, and every AI feature should be evaluated against them.

If your design fails on any of these three fronts, then there’s a fundamental flaw in the design.

And that’s a huge problem since product design also functions as a trust calibration layer between the AI and the user.

How Miquido Integrates AI in Its Projects

So what does all of Biernacki’s advice look like in action?

One of their latest projects, a mobile app for Diagnostyka, Poland’s leading laboratory diagnostics provider, is a fantastic case study.

New Diagostyka App

Miquido and Diagnostyka worked together to introduce two AI-powered preventive healthcare features:

• LiDia: A virtual AI assistant that organizes preventive healthcare in a transparent, fact-based manner.
• Profilaktometr: A proprietary gamified dashboard that helps users start or improve their preventive health activities.

Both of these features weren’t implemented just to have AI in the app. Miquido leveraged Diagnostyka’s decades of medical knowledge and experience to create both features.

This helps the app deliver recommendations powered by real medical knowledge, which is then paired with understandable context.

This allows users to see why actions are suggested rather than asking them to trust advice blindly.

At the same time, these features go beyond the basic functions of purchasing tests and downloading results.

It also acts as a daily companion for users in their personal preventive health journeys.

Both deliver clear and impactful value to anyone who uses the app.

More importantly, this value is something ChatGPT cannot provide, since standard models might suggest tests that are incompatible with one another or are no longer performed.

“Your real competitive advantage in this flood of AI slop isn't the model you use. It's your knowledge, your know-how, and your unique approach,” Biernacki said.

Measuring Degrees of Success

Of course, it isn’t enough to know how to integrate AI into a mobile app. There’s also the matter of measuring whether an AI-powered feature is actually delivering on its promise

At Miquido, Biernacki and his team use three layers of measurement.

“If you can't measure it on all three layers, it's a toy, not a product feature,” he said.

These measurement layers include:

1. Direct user value

Is the feature you’ve added actually saving users time, money, or energy compared to how they were doing things before?

If your AI recommendation engine doesn't outperform a simple "most popular" list, you're burning tokens for vanity.

2. Business impact

The feature should move the metrics that actually matter, such as conversion rate or revenue per user.

Pick the metrics that matter most to you and measure them honestly.

And if your AI feature doesn't show up in a business metric within a quarter or two, you need to reassess.

3. AI-specific trust and quality metrics

Biernacki points to four trust and quality metrics in particular:

• Truthfulness
• Groundedness
• Coverage
• Relevance

All four help evaluate whether your AI's outputs are factually correct, grounded in the data you provided, comprehensive enough, and staying on topic.

These metrics are so crucial that Miquido has built these evaluators into its AI Kickstarter framework as part of its standard production toolkit.

Build Less, Prove More

The next phase of AI in mobile will not be defined by how much is built, but by what proves its value.

Biernacki’s advice to achieve this remains consistent:

1. Start with a real problem.
2. Build a focused solution.
3. Measure outcomes honestly.
4. Scale only when the results justify it.

“The most future-proof AI investments aren't in models or fancy interfaces. They're in "boring" foundations like clean data, clear ownership, and disciplined processes,” he concluded.

And in a market full of AI-powered everything, the smartest products may be the ones that know when AI should be left out of the picture.