In the first part of the blog series “Leading with AI – think strategically, implement effectively” we took a closer look at the diselva AI Excellence Framework. This framework consists of the elements Discover, Validate, Prototype, Scale, and Sustain.
From our perspective, AI is not just a technological trend but a new step in digital evolution. This process is characterized by its high speed. To keep pace at the right tempo, it is crucial to think strategically and to align with the needs of your organization.
This second part, “Leading with AI – implement effectively,” shows how strategic thinking helps you become a leader in the AI space at high speed and implement new solutions.
Strategic Experimentation
The impressive capabilities of Large Language Models (LLMs) like ChatGPT were not first defined as requirements and then implemented. The models came first. Through extensive experimentation, their fascinating possibilities were explored, and based on the results, larger and better models were developed.
As an organization, it is therefore essential that you enable and encourage your employees to experiment with AI. At the same time, it is indispensable to train the correct handling of internal data, especially personally identifiable information (PII). To do this, the appropriate tools with enterprise licenses and employee logins must be evaluated and acquired. These future-oriented investments need to be made quickly yet sustainably, which poses a challenge for many organizations.
AI solutions from software vendors
As a central technology trend, most software vendors are developing and integrating AI support into their software solutions. Many of these functions are already available, meaning your organization is very likely already using software with built-in AI capabilities. Therefore, review which AI features are already present and ensure that your employees know about them and are able to use them. Often, these new AI functions come with additional costs. In such cases, we recommend a careful yet swift evaluation and decision for or against using the feature.
We also recommend staying informed about the AI-related roadmaps of the software solutions you use. This ensures that you do not develop your own AI solutions that will soon be provided by one of your existing software tools.
Intelligent Solution Architecture
Individual AI solutions require an intelligent solution architecture. These solutions access internal enterprise data and applications, require authentication and authorization mechanisms, and must not negatively impact the daily work with applications and data. It must be ensured that data processing is clearly defined and complies with legal regulations. Even when a new AI solution is implemented as a prototype, no compromises may be made in critical areas.
Security must operate deterministically
Security is a critical success factor for any AI solution and, depending on the data integration, often business-critical. It is important to keep in mind that a prototype using production data is subject to the same requirements as standard production software solutions. All security mechanisms must operate deterministically. This specifically means that pure system prompts/inputs for an AI model, or text-based guardrails, cannot provide guaranteed access protection.
Developing AI solutions with AI coding agents
AI models such as LLMs are very good at generating software code and are already used daily by most software engineers. To implement AI solutions effectively, the use of AI coding agents is therefore recommended. On the one hand, this increases efficiency; on the other, it builds experience in using AI within software development. AI does not replace experienced software engineers, but it can make them faster.
AI projects are different
At first glance, AI projects look like normal software development projects. They are often sold, approached, and executed just like software projects, only to lead to disappointing results. But AI projects are different and must be approached starting from the desired outcome. AI models do not deliver deterministic results. Ask ChatGPT the same question twice, and you will receive two different answers. AI projects therefore require a new mindset at all levels of an organization.
Implementing AI effectively also means being able to decide whether an AI initiative is successful and should be continued, or whether it does not deliver the expected value. Determining which new success factors should guide such decisions is a challenge for every organization.
You can find more about “Leading with AI – think strategically, implement effectively” in our first article, “Leading with AI – think strategically.”
For more information about diselva’s AI-related offerings, we look forward to hearing from you.