The Results of the Last Six Months and the 5x More Efficient AI Development

Last year, we embarked on an ambitious endeavor to make the entire AI development process more efficient without compromising the performance of the models. To achieve this, we formed teams of four and three members. The team of three worked based on a new “Synergy” methodology, while the team of four followed the traditional agile methodology (they were entirely unaware of the new methodology).

Why is this important?

• AI became the biggest buzzword in 2023-2024, making it imperative for companies to adopt these solutions.

• AI and ML models can provide a significant competitive advantage, especially in sectors where time is a critical factor.

• Saving money and time by efficiently moving models from the PoC phase to the production phase is essential.

• According to our survey 76% of organizations fail with AI with Agile methodogy
  

How do we execute this? 

In conventional AI applications, roles are usually very distinctly separated. Typically, there are four roles: Data Engineer, Data Scientist, ML Engineer, and Project Manager. These roles are quite costly, and full readiness is not always necessary. In our case, these roles were less divided, blending the Data Scientist and ML Engineer positions into an Integrated AI Specialist position. This is because it's not enough to simply create good PoC models; they also need to be taken into production.

Another problem is the misinterpretation of agility as a concept. Agility is not about knowing what the final product will be and the exact steps to get there, but rather about setting a goal or KPI that you suspect can be achieved with a particular end product and iterating multiple times to ensure that the end product is indeed necessary. Many projects become costly because iteration cycles are not planned well or not planned at all, as managerial-level participants are often not familiar with the AI development process.

Technological Methodology: After organizing the team and project, it’s worth mentioning the technological innovation. AI projects fundamentally consist of very similar components that can be standardized, saving time in the long run. During the evaluation period, we developed an AI architecture that can integrate with existing systems, effectively integrate unique use cases, and ensure code quality.

Sectors where we tested the methodology

• Banking and Insurance Sector: Customer segmentation – the goal was to identify clients who are receptive to taking out loans, which benefits them. Within various segments, we were able to identify parameters indicating what the client is inclined towards (e.g., vehicle, home, travel), aiding in crafting personalized offers and sales processes.

• Manufacturing: Quality assurance – components from different suppliers affect the overall product quality. Our unique AI solution could predict the type of defect an item would have within a specific period, which supplier was responsible for the defect, and how to improve it.

• Agriculture: Operations – the growth rate of plants significantly depends on environmental factors like temperature, humidity, and soil quality. Our task was to build a software system capable of measuring, predicting these quantities, and scheduling regulation based on the predictions.

• Telecommunications: Personalized offers – instead of providing the same or very similar packages to all customers, we tailored these based on user habits, which increased customer loyalty. As marketing data shows, acquiring a new customer is four times more expensive than retaining an existing one.
  

Lessons Learned? Results Compared to the Control Group

Team Dynamics and Task Distribution:

• Meetings: The group following the old methodology spent 52% more time in meetings and status updates than the group following the new methodology.

• Independent Development: In the old methodology group, developers could work more independently, but throughout the development process, there were more frequent miscommunications and technological mismatches. In the new group, there were more pair coding sessions (about 7% of the total development time), which helped maintain an overview of the entire code, making it easier to delegate tasks during absences.

• Meeting Framework: The old methodology group held the following meetings: Standup, Planning, Demo, and Retro. Results: Standup meetings often disrupted developers' daily routines, making them feel less productive. Planning meetings often revolved around navigating the project manager through the technological landscape, frustrating developers. Retro meetings saw more frequent clashes for the above reasons. The new “Synergy” methodology group held regulated meetings: Impact Report and Review Rally. There was no negative feedback from this organization, and the absence of standups had a positive effect on developers as they could focus on actual work.

Development Process

• Task Allocation

• New "Synergy" Methodology Group: Focused on task standardization, with less emphasis on bug fixing and debugging.

• Old Methodology Group: Primarily concentrated on bug fixing and debugging, with less focus on standardization.

• Initial Start

• New "Synergy" Methodology Group: Spent the first week planning and produced testable results by the end of the second week.

• Old Methodology Group: Began development on the second day and had testable results by the end of the first week.

• Endgame

• New "Synergy" Methodology Group: Easily incorporated changes or feature requests due to thorough planning.

• Old Methodology Group: Faced more challenges with integrating changes or new features.

Results

• Time spent on development at the end of the first project:

• Traditional Agile methodology group: 66 days

• New “Synergy” methodology group: 57 days

• Time spent on development at the end of the last project (combined):

• Old methodology group: 391 days

• New “Synergy”  methodology group: 192 days

• Time spent in meetings:

• Old methodology group: 37 days

• New “Synergy”  methodology group: 17 days

• Number of developed features:

• Old methodology group: 214

• New “Synergy” methodology group: 568

• Stakeholder satisfaction (process, demos, documentation, reliability):

• Old methodology group: 7.3

• New “Synergy” methodology group: 9.2

• Team comfort and feedback:

• Old methodology group: 8.3

• New “Synergy” methodology group: 8.9

Final Results:

• Qualitative Results:

• Quantitative Results:

• Based on the number of developed features and time spent on development, the new “Synergy” methodology is 5.4 times more efficient than the old one.

• There is no significant difference in team comfort levels.

• Stakeholders were over 26% more satisfied with the overall process following the new “Synergy” methodology.

Conclusion

It is evident that, regardless of the industry, we can organize human teamwork in the IT field in a way that provides a significant advantage. The fact that a team of the same size can work 5 times more efficiently than a team following another methodology indicates that we don’t necessarily need new tools but rather a shift in perspective. Among other things, we can assist with this through the Syntheticaire team. Whether it’s education, workshops, process optimization, or specific development (UI/UX design, backend, frontend, and AI development), we can certainly help. Contact us through the form, via email at info@syntheticaire.com .

Because unique solutions and proven results are what your organization deserves.