Last CallLast Call Pool
Product & Management Stage

AI or not AI: making sense of the hype

September 18, 2025 2:45 PM
Power Talk
🇬🇧 English
Bratysława 1

About

Is AI truly needed in your product, or is it just a buzzword? This talk helps product teams make informed decisions by distinguishing real AI value from hype. We’ll explore when AI significantly outperforms traditional methods, the risks of “AI for AI’s sake,” and key success metrics. Through practical frameworks and case studies, we’ll cover cost, complexity, and fast hypothesis testing with MVPs. The takeaway? Stay rational—focus on user needs, not trends, to build impactful products.

‍

Talk by

Photo of Veronika Tamaio Flores, Product Lead at Railsware
Veronika Tamaio Flores
Product Lead at Railsware

Watch the full talk

Watch this WaysConf session, then continue with related talks or explore the current programme.

Explore the 2026 agendaMore WaysConf recordings
From the recording

Talk in brief

In this WaysConf 2025 presentation, Veronika Tamaio Flores, Product Lead at Railsware, addresses the massive gap between artificial intelligence hype and actual business value. Drawing on her experience implementing AI at Coupler.io, she highlights that while venture capital and startup trends heavily favor AI, many enterprise users have yet to see tangible returns. Flores outlines a practical framework for determining when to implement AI—such as automating tedious tasks or accelerating time-to-action—and when to avoid it, such as for simple queries or scheduling. She shares real-world performance metrics from Coupler.io's AI insights widget, demonstrating that successful AI integration relies on high-quality data, starting with small-scale heuristics, and focusing strictly on solving user problems rather than chasing technology for its own sake.

Key takeaways

  1. 01

    The Hype-to-Value Gap

    Many organizations rush to implement AI features out of fear of missing out, yet very few can actually measure the return on investment or prove these features solve real customer problems.

    Watch from 2:45
  2. 02

    Prerequisites for AI Implementation

    Before integrating AI, a product team must have a well-defined task, the necessary infrastructure, specialized talent, and, most importantly, high-quality structured data.

    Watch from 5:06
  3. 03

    When to Avoid AI

    AI should be avoided if a simpler technology can solve the problem, if the organization lacks basic data management, or if the feature does not deliver clear, measurable value to the user.

    Watch from 5:49
  4. 04

    Real-World Cost and Performance Metrics

    Coupler.io's AI insights widget demonstrated that practical AI features can be highly cost-effective, averaging less than ten cents per generation with a 90% user satisfaction rate.

    Watch from 11:09
  5. 05

    The "Start Small" Testing Framework

    Product teams should validate AI concepts by starting with simple heuristics, utilizing cheaper models, or employing "Wizard of Oz" testing before committing to complex, expensive engineering.

    Watch from 16:24
Read edited transcript highlights

These concise notes were edited from automatic captions and checked against the talk structure. They are not a verbatim transcript.

The Disconnect Between Hype and ROI

During her presentation, Veronika Tamaio Flores asks the audience how many have launched AI features and, more importantly, how many can actually measure their return on investment. The lack of hands in the room illustrates a widespread industry issue: the massive gap between the pressure to deploy AI and the ability to extract real business value from it. Many product teams are driven by the fear of missing out rather than a clear understanding of how these features solve actual customer pain points.

Watch from 2:45

Foundational Requirements for AI Success

Implementing AI successfully requires far more than just calling an API. Product teams must first establish a well-defined task that genuinely benefits from machine intelligence, supported by robust infrastructure and the right talent to understand its limitations. Most critically, AI is useless without high-quality, well-managed data. Organizations that attempt to build AI features without basic data management and clean data sources are bound to fail.

Watch from 5:06

Why Customers Do Not Care About the Tech Stack

A common pitfall for product teams is assuming that labeling a feature with "AI" will automatically make it more attractive or valuable. Research indicates that users are not willing to pay more simply because a feature is branded as AI. Customers ultimately care about saving time, reducing costs, and achieving their goals efficiently. The underlying technology is irrelevant to them; they only care about the practical value and usability of the solution.

Watch from 7:00

Metrics from Coupler.io's AI Insights Widget

To illustrate a successful implementation, Flores shares data from Coupler.io's AI insights widget. Across approximately 700 generations for over 100 users, the total API cost was only 62 dollars, averaging less than ten cents per query. With an average generation time of just over 20 seconds and a user satisfaction rate near 90 percent, this feature proves that targeted, prompt-based AI can deliver high value at a very low operational cost.

Watch from 11:09

Choosing Manual Workflows Over AI

Coupler.io intentionally avoids AI in several key areas where human precision or simpler logic is superior. For instance, data flow scheduling remains manual because users simply want their data refreshed at predictable times. Schema mapping and dashboard templates also rely on manual configuration and human expertise, as AI can fail catastrophically in these areas. Additionally, simple actions like sorting and filtering are much faster to execute with two clicks than by writing a prompt.

Watch from 14:32

The "Start Small" Product Strategy

When introducing AI, the best approach is to start with the smallest possible scope. Product managers should consider using basic heuristics or cheaper, smaller models to solve the most fundamental user problems before scaling up. Techniques like releasing to a small batch of early adopters or using "Wizard of Oz" testing—where human effort mimics automated AI processes behind the scenes—allow teams to validate demand and functionality without over-engineering.

Watch from 16:24
Explore WaysConf 2026

View the agenda

Get tickets