2 min read time

Not All AI Solutions Are Created Equal in Functional Testing

by   in DevOps Cloud

AI should be able to make your life easier, right? Whether it's machine learning, natural language processing, or computer vision, it safe to assume functional testers employ AI solutions with the expectation it will improve quality, and ultimately make their work-life less burdensome. One could be forgiven for thinking all the AI models being offered render the same results, but challenges remain for many organizations to implement a reliable AI model that can recognize even the smallest changes without risking the integrity of multiple scripts. Anything less, and you’re back running maintenance as if the AI never existed. So, what goes into an ideal AI model that can truly reduce and frequently eliminate tedious test maintenance making life easier for your testers? Let's take a look at two key capabilities to look out for when implementing AI into functional testing efforts:

1. Objects Identified Using VISUAL Characteristics - Not PROPERTIES

There are plenty of vendors who can implement an AI model for functional testing, but many options are still reliant on object properties. This method involves capturing a screenshot, attempting to locate a similar screenshot, and then deciphering object properties for script updates. Even worse, if you were to pair such an AI model with non-standard OCR, the results could be disastrous. Testers often report that this type of solution will identify the same characters differently on immediate, subsequent interrogations, which means they will spend just as much time and effort (or even more) on maintenance as before any AI implementation.

In contrast, an ideal AI strategy leverages visual object recognition to streamline testing processes. Rather than fixating on identical images and properties, a superior AI model identifies objects based on their visual characteristics. Once located, it intelligently interacts with the object by capturing its screen position, thereby eliminating the need for extensive property management, and multiple steps for complex objects (like search fields, calendars, etc.), handling those interactions naturally, like a human would. This approach not only enhances accuracy, but also promotes flexibility in handling diverse interface designs and updates.

2. Plays Well With Others

Another key drawback of some custom-built AI solutions lies in their susceptibility to inconsistencies. These solutions often struggle to maintain consistency in character recognition across multiple interrogations, leading to unreliable test outcomes. In contrast, a better AI model will have the ability to work with configurable tools such as ABBYY, Google OCR, Tesseract, or Baidu. This flexibility allows testers to adapt the AI to different environments and optimize performance according to specific requirements, ensuring consistent and reliable results.

---

In the realm of software testing, using AI offers immense potential to streamline processes and improve efficiency. However, not all AI solutions are created equal. There is a significant difference between models that can identify objects on visual characteristics and those that cannot. Additionally, consistency is key, and a model that can work with other configurable OCRs helps a lot. Ultimately, the better the AI model the more capable it will be at ignoring innocuous changes in the user interface – which means less broken tests and less maintenance for everyone involved in your organization’s functional testing effort.

Labels:

Functional Testing