AI Implementation: Key Considerations from 3 Google Cloud Customers

Why AI adoption still lags 

It may be surprising that in Devoteam’s AI Survey, only 60% of respondents report using AI in their business. [Source: a preview of Devoteam’s AI Survey Results] This raises the question: why? Why is there such a gap between the potential of AI and its actual implementation in an organisation? At a recent Google Cloud Webinar, experts from different businesses (like Strawberry, Cegid, and Lampenlicht) discussed how to implement AI in Business.

What are the Key Considerations for Successful AI Implementations? 

Before we start, let’s remind you that successful AI adoption requires more than just technology. It involves a strategy for people, processes, and a supportive culture.

• People: Foster a culture of innovation and ensure employees have the necessary skills and training.
• Process: Adapt and refine processes to integrate AI effectively.
• Technology: Invest in robust data infrastructure, security measures, and reliable AI tools.

How to Implement AI in Business

Considering people, processes, and technology, here are 5 key takeaways for effectively implementing AI in business and driving meaningful growth:

1. Identifying Business Challenges and Opportunities

Before starting your AI journey, it is essential to identify specific business challenges or opportunities where AI can add real value. Start by asking:

• What are the pain points in our current processes?
• Where do we see bottlenecks or inefficiencies?
• Are there areas where we can leverage data to gain insights and improve decision-making?

For example, Strawberry, a Scandinavian hotel company, faced challenges with legacy systems, employee turnover and the need to streamline training. This led them to explore AI Use Cases that make training simpler and information more accessible. 

Cegid needed to manage a high volume of support tickets related to product usage and aimed to provide 24/7 support. They sought ways to move from a support call approach to self-service. 

Lampenlicht has a small marketing team but needed to scale its content production for a large European market. They were looking to overcome localisation issues and reduce translation costs. Also, the procurement team was looking for solutions to become more efficient. 

2. Building a Strong Data Foundation

While GenAI has reduced the need for very curated training data sets, data foundations are still critical for performance. Why? Because AI models rely heavily on data quality. Building a robust data foundation is essential for successful AI implementation. This involves:

• Data Quality: Ensuring data accuracy, completeness, and consistency.
• Data Governance: Establishing clear procedures for data collection, storage, and management. Read more on why data governance is key to your AI Transformation. 
• Data Security and Compliance: Implementing security controls and complying with regulations like GDPR.

Lampenlicht successfully implemented AI solutions by migrating to Google Cloud, creating a flexible environment for AI development and establishing a strong data pipeline.

3. Build vs. Buy: selecting the Right AI Solutions

Once you’ve identified your business challenges and established a data foundation, you can explore potential AI solutions. Consider whether a “build” or “buy” approach suits your needs.

• Build: Developing custom AI solutions in-house provides greater control but requires significant expertise and resources. Cegid initially explored both options with a pilot project but discovered that a “build” approach was necessary to truly leverage the potential of AI. 
• Buy: Leveraging existing AI platforms and tools can be faster and more cost-effective but may offer less customisation. Strawberry partnered with Devoteam to develop their AI chatbot, leveraging platforms like Dialogflow CX and Langchain. Buying off-the-shelf solutions may however not be sufficient for the continuous improvement necessary with AI. 

4. Ensuring user buy-in

Implementing AI often involves changes to existing workflows and user habits. To ensure successful AI adoption, focus on:

• Meeting users where they are: Integrate AI tools into familiar workflows and platforms. Strawberry integrated their chatbot into Google Chat for easy accessibility. They also connected their AI to Google Drive and ticketing systems. 
• User Experience: Design AI applications with user-friendliness in mind. Did you ever consider the importance of a tone of voice and personality for your AI chatbot? Strawberry chose a sassy tone for their chatbot to increase engagement. 
• Communicating the benefits of AI: Promote the benefits of AI and provide adequate training and support. For example, help users understand how to interact with the AI using natural language prompts instead of keywords. 

5. Embracing Continuous Improvement and Iteration

AI implementation is an iterative process that requires continuous improvement and adaptation. Regularly monitor and evaluate your AI systems, gather user feedback, and make adjustments as needed. Cegid recommended cultivating a mindset of continuous improvement. Cegid had to shift their team’s mindset from viewing AI as a traditional technology purchase to seeing it as something that needed constant training and adjustment. They also highlighted the need for realistic expectations. AI implementation is not “magical”, it requires time and effort to achieve strong results. It will need continuous improvement rather than being perfect out of the box. 

For example, Cegid started training their AI model on 137 top questions, which yielded 85% correct answers. When they moved to 5000 real questions, their correct answer rate dropped to 19%, but they were able to reach 85% again after two weeks of improvement. 

Real-World Examples of AI Implementation

• Strawberry: Their AI chatbot, Scout, reduced support ticket volume and improved knowledge accessibility.
• Cegid: Their virtual agent enhanced customer support quality and efficiency.
• Lampenlicht: AI-powered content generation at Lampenlicht increased content production and customer engagement.

Future Trends and Directions in AI

So how will AI evolve in the future? Google focuses on improving AI models, multimodal capabilities, and features for explainability, monitoring and security. 

• Advancements in Generative AI: Generative AI models are becoming increasingly powerful, opening up new possibilities for businesses.
• Multimodal AI: AI systems capable of processing and generating information across multiple modalities.
• Explainable AI: Ensuring transparency and understanding of AI decisions.
• Responsible AI: Addressing concerns around data privacy, bias, and ethical implications.

Lampenlicht envisions a future in which each department will have its own AI Agent. Strawberry has plans to connect its AI to more fulfilment systems. Cegid anticipates the emergence of new roles, such as conversational architects, as AI becomes more integrated into business processes. This also means upskilling human agents to handle more complex tasks, while AI manages basic questions. This shift in roles and responsibilities is a direct consequence of AI’s growing capabilities in customer support and other business areas. 

Implementing AI solutions can transform your business, but it needs a strategic approach and a commitment to always improving. If you follow the steps in this article, you can successfully navigate the complexities of adopting AI and drive impactful results across your organisation.