Capgemini’s RAISE framework signifies a new era of advancement and evolution for generative AI. It embraces the fast-evolving and highly experimental development of the technology and adapts to the rapid pace of innovation. RAISE delivers accelerators and learnings to harness the power of AI and generative AI while focusing on sustainability, scalability, and trustworthiness.

Generative AI has emerged as a groundbreaking force, democratizing innovation across industries. Open source and commercial models alike have become widely available, leveling the playing field for those eager to harness their potential.

However, as approachable as generative AI may seem, navigating its complexities is no small feat. Within just a year, the field has seen seismic shifts in paradigms and underlying technologies. Organizations are exploring generative AI, recognizing its value as a catalyst for innovation and revenue growth. The Capgemini Research Institute underscores this trend, revealing that nearly 90% of organizations plan to prioritize AI, including generative AI, in the next 12 to 18 months. The question is: are organizations ready to transition from mere exploration to achieving tangible results?

Generative AI generates new content, ideas, or solutions by learning from vast datasets. This extends from creating realistic images and text to generating code and innovative solutions across various fields.

As generative AI solutions are being constructed, decoupled development has led to redundancy and inefficiency. This disconnected approach gives rise to multiple issues: identical open-source models running on separate GPUs, increasing costs and complexity; commercial APIs used in disparate applications, preventing better vendor deals due to split volumes; and the repeated development of similar applications without performance comparison or monitoring.

The year 2023 was a time for experimentation; 2024 is the year for scaling. To address these challenges and herald a new era of development, we crafted the RAISE framework to enforce sustainability, reusability, and trustworthiness throughout the code, establishing a robust AI partnership with our clients. The RAISE framework streamlines the development process, ensuring that generative AI solutions are built on a solid, efficient, and cohesive infrastructure.

Exploring the RAISE framework

RAISE is a gateway to a future of trustworthy, scalable, and sustainable AI. While many companies concentrate on solutions, RAISE shifts the focus to infrastructure – the foundation of both the service and solution layers.

RAISE’s is built on modularization. It promotes the development and deployment of reusable components as independent services, covering an array of AI and generative AI models, tools, and data services. This includes both commercial APIs and open-source models.

At the core of the RAISE framework is a uniform pipeline structure, ensuring cohesion and efficiency throughout development and deployment. It emphasizes efficient deployment of open-source models, leveraging shared GPUs for optimal resource utilization, minimizing environmental impact, and maximizing performance. Its unified management system facilitates easy comparison, efficient deployment, and thorough monitoring of both open-source models and commercial APIs, improving scalability and enabling cost savings as new models emerge.

Adaptability and experimentation are critical, as is embracing a diverse mix of technologies and staying open to future changes. As H. James Harrington noted, “Measurement is the first step that leads to control and eventually to improvement,” highlighting the importance of enforced evaluation. Such processes not only enable rapid comparison of changes in the experiment stage but also ensure smooth transitions to new ideas and models.

RAISE continually identifies and builds reusable components to enhance trust, efficiency, and performance in AI and generative AI applications. Its current offerings include model cascading for cost savings, prompt optimization for performance improvement, and RAG services for scaling enterprise text retrieving service. The framework is committed to evolving and refining its services, empowering its users to embrace ultramodern technology.

RAISE provides a path towards:

• Trustworthy AI, through rigorous evaluation, testing, and monitoring
• Scalable AI, by embracing modularization, DataOps, MLOps, DevOps, and governance
• Sustainable AI, focusing on cost optimization, reusability, and efficiency.

Pioneering the future

The RAISE framework is setting the pace towards a future where AI’s potential is fully unleashed. This next phase in RAISE’s evolution is pioneering tomorrow’s innovations.

Innovative deployment and customization. The future of RAISE is marked by an even greater emphasis on customization and efficiency. Leveraging the latest advancements in large language models (LLMs), RAISE is poised to offer an even more refined infrastructure setup. This includes specialized pipelines for deployment, fine-tuning, and data management, designed to streamline the AI development lifecycle from conception to deployment.

Tailored data and model training. A standout feature for RAISE is its enhanced capability for organizations to craft their own high-quality datasets for training or evaluation purposes. This ensures the data meets the specific needs of each project and also elevates the quality of model training. Coupled with the RAISE training framework, organizations will have the flexibility to develop custom models, pushing the boundaries of what’s possible with generative AI.

Cost-effective model selection. A novel aspect of the RAISE framework is its intelligent model selection service, designed to optimize resource allocation by matching the most suitable model to each task. This reduces costs and amplifies the effectiveness of AI initiatives.

Leading the way into tomorrow

RAISE is pioneering innovative solutions that address today’s challenges and anticipate tomorrow’s opportunities. As large models advance in capability, emerging agents face the challenge of dynamically decomposing tasks and choosing the right tools for completion. RAISE provides these agents with an ever-growing toolbox, incorporating a comprehensive catalog of information and standardized endpoints.

We invite you to reach out. Together, let’s shape the future of AI, leading the charge into uncharted territories of possibility and success.

INNOVATION TAKEAWAYS

MODULARIZATION FOR EFFICIENCY: Break down AI development into reusable components with RAISE, optimizing resources and streamlining processes for enhanced efficiency.

TRUST AND SCALABILITY: Ensure trustworthiness and scalability in AI solutions with RAISE’s evaluation, testing, and monitoring mechanisms.

FUTURE-PROOF INNOVATION: Stay ahead of the curve with RAISE’s commitment to adaptability and customization, empowering organizations to pioneer tomorrow’s AI solutions.

Interesting read?

Capgemini’s Innovation publication, Data-powered Innovation Review | Wave 8 features contribution from leading experts from Capgemini and esteemed partners like Dassault Systèmes, Neo4j, and The Open Group. Delve into a myriad of topics on the concept of virtual twins, climate tech, and a compelling update from our ‘Gen Garage’ Labs, highlighting how data fosters sustainability, diversity, and inclusivity. Embark on a voyage of innovation today. Find all previous Waves here.

ServiceNow helps eliminate friction in human resources by seamlessly integrating people and processes 

Companies are fundamentally composed of people. While this may seem obvious, the complexity of recruitment, onboarding, training, and retaining employees reveals just how intricate managing this essential resource can be.  

As technology continuously evolves, transforming the nature of work and the workplace, it is crucial that the workflows supporting the employee journey evolve as well. However, outdated technology, inefficient processes, and siloed teams often hinder companies from delivering a seamless employee experience. This is where ServiceNow comes in. By providing an integrated platform and employee portals, ServiceNow enables companies to revolutionize their investment in human capital. 

Envision a transformative employee culture

Human resources encompasses a broad spectrum where people and processes intersect at various touchpoints. However, amidst the analysis of terabytes of data and assessment of key performance indicators (KPIs) is “human.” To foster a positive and transformative company culture, the employee experience must be human-centric, composed of individual moments that matter, and seamlessly connected. 

For instance, companies often rely on various distinct systems from third parties, such as Microsoft, Oracle, SAP, and others, to manage key steps in the employee journey. These systems handle aspects like day-to-day objectives and KPIs, compensation and benefits, learning opportunities, and overall career roadmaps. However, navigating through these multiple systems can be time-consuming and inefficient, leading to frustration, especially among new hires, and wasting valuable company resources. 

ServiceNow is an integrated platform that provides human-centric advantages, effectively breaking down siloes and creating cohesive workflows in the back end, where third-party applications and data are housed. Simultaneously, ServiceNow’s employee experience portal offers user-friendly access to key information that can be personalized for an individual, based upon their role, responsibilities, interests, and even geographic region. 

Identify points of friction

The actual employee experience often differs significantly from what many company leaders in IT and HR might expect. According to Capgemini’s 2022 research of 2,250 employees across 750 organizations, less than one-third (30 percent) of employees are happy at work, yet 92 percent of managers mistakenly believe they are content. Even more worrying is that 68 per cent of employees did not know what they are expected to do in their role.  

For instance, employees reported they did not have access to data to make informed decisions. They lack clear, quantifiable objectives and do not understand how those objectives are aligned to the company’s overall targets.

Not surprisingly, these types of barriers can skew an employee’s perceived performance, leading to dissatisfaction and increased attrition. Our survey found that a staggering 20 percent of employees left the company within 45 days of being hired.  

Automate and optimize the employee experience

Enabling the right technology can help solve these issues, ensuring employees have the best experience throughout their career at the company, starting from the recruiting stages and continuing immediately after hiring. 

Consider the actual onboarding process, for example. It requires collaboration across departments and involves numerous stakeholders, including managers, facilities, security, and finance. Less than half of these activities or touchpoints are related to HR, which can lead to a fragmented employee journey. ServiceNow’s platform brings it all together to clearly integrate these crucial steps. This helps companies provide a smooth end-to-end experience for new employees, right from when they first join an organization.  

Capgemini also has a pre-packaged ServiceNow HRSD product that optimizes and automates manual processes and workflows. It provides managers (and new hires) with a detailed list of every onboarding activity required, from sending a welcome email and assigning a desk to ensuring the appropriate forms are filled out and training tasks are set up. This way new hires can get up to speed and become productive more quickly.  

ServiceNow’s capabilities continue across the employee journey, extending from “hire to retire,” culminating in a unified, consistent people experience. Capitalizing on such advancements in technology can streamline workflows, automate processes, and enable data-driven decision-making for companies. This helps managers make the most of “human” resources, while promoting a productive and positive workplace where people truly come first. 

ServiceNow can help create an environment that fosters employee engagement and growth  

People are a company’s greatest asset. However, maintaining enhanced employee and customer experiences has remained a challenge for organizations, especially those struggling to implement technological changes. With advancements in artificial intelligence to optimize and automate organizational workflows, companies at any stage of their digital transformation journeys can focus on an often-overlooked investment – employees – to gain a competitive edge. 

Promoting a people-first strategy through technology

Happy employees are more productive, and that leads to a wide range of quantifiable benefits across an organization. 

• 59 percent reduction in turnover 
• 41 percent lower absenteeism 
• 21 percent improvement in profitability 
• 20 percent increase in sales 
• 17 percent increase in productivity 
• 10 percent improvement in customer ratings 

Enabling the right technology is a foundation that can be built on to transform a company’s performance culture, allowing it to aim for these types of improvements. However, for many organizations, the employee journey from recruitment to retirement is a disconnected process rather than one that flows without friction. Instead of being proactive in promoting change and making strategic decisions based on data, company leaders often find themselves bogged down by trouble-shooting specific issues.  

A more holistic approach can create an environment that fosters employee engagement and growth. This involves adopting a combination of strategy and culture, technology, and operations that work together to deliver impactful moments and seamless employee experiences. Organizations that already rely on ServiceNow to deliver key business solutions can also use its integrated platform to advance how they manage people and processes quickly and easily.  

Implementing these changes might sound complex; however, Capgemini has a new pre-packaged ServiceNow HRSD product. It’s designed to help companies address everyday challenges, such as absenteeism and productivity, and enable data-driven decision-making to produce positive outcomes and, in turn, boost profitability. We can work collaboratively to identify your company’s points of friction, find opportunities to enhance existing process, and tailor solutions to your specific human resources needs. 

Producing impacts across the employee experience

Integrating technology is an essential element in creating a seamless people experience, from onboarding to offboarding. ServiceNow’s platform supports a number of third-party systems typically used by human resources and IT teams, which can be accessed through a user-friendly employee experience portal.  

Even that seemingly small entry point for employees can make a big impact. Having a single self-service interface immediately promotes autonomy and reduces friction; instead of needing to log in to a number of portals just to access tools for productivity and collaboration, everything is all in one place. Additionally, rather than working in siloed programs, managers can connect jobs and roles to outcomes, providing employees with clear and quantifiable objectives related to their day-to-day work, as well as their career growth potential. 

Consider your organization’s intranet as one type of employee portal. It’s probably populated with all types of information individuals might want to access, from financial results to company events. Having access to a wealth of information, however, doesn’t guarantee an inclusive employee experience. Data intelligence derived from analyzing patterns on intranet portals reveals that a tiny percentage of the information is ever looked at, much less used. Rather than being a resource, these internal pages are difficult to navigate when employees are seeking information and resources to do their jobs. 

Instead, envision how portals and collaborative spaces can be personalized for each employee. Only information relevant to their role’s objectives, the tools and technology they need for their work, and access to key data so they can track individual KPIs and other metrics are surfaced. ServiceNow can facilitate this type of targeted communication and assemble touchstones to drive business outcomes across the employee journey.  

Preparing for the future of work 

Using analytics and insights to continuously improve all aspects of the employee journey is already transforming the future of work. As generative AI technology continues to mature, manual processes can be improved through automation, freeing up staff to work more strategically, while maintaining the “human” in a people-first strategy. It’s accelerating the transformation of the employee experience, allowing people to be much more efficient in the ways they interact within the organization. 

Soon, employees will be able to access a personalized chat bot to find specific policies or learning opportunities. Consider how technology built upon large language models will soon deliver features such as scanning an employee’s list of skills, experience, and even language abilities to recommend new job opportunities or refine their career path within the organization. This also aids in succession planning and adds value to a company’s long-term investment in human capital. 

Such possibilities are already in the works, with generative AI bringing together all this complexity and delivering an end experience in a very simple and accessible way.  This type of environment fosters employee engagement and growth, a valuable resource that brings myriad benefits to an organization, ultimately giving it a competitive edge. 

Most companies that rely heavily on document stores for knowledge sharing and team collaboration often end up with many pages created by users.

The rapid growth and constant evolution of these knowledge bases pose significant challenges in finding relevant content. Despite diligent documentation, navigating to the pertinent information remains difficult – one either knows where the document is or the exact keywords to find it.

Real customer scenario

At one of Capgemini’s clients, a team operating and building a new data platform was entangled in customer support, reducing its ability to create new functionalities.

Allow me to briefly explain what a typical support request entailed:

1. The customer raises a question to the platform team.
2. A dedicated person from the platform team browses available Wiki documentation and searches for relevant information.
3. Several minutes or even hours later, the person passes the information (a link) to the requestor.
4. The answer may not be clear, prompting the need for another question to be asked.

The weekly effort spent on customer support is increasing, and it is projected to reach 2.5 “FTE” permanently occupied with customer support activity by the end of 2024, as the number of platform users grows. Moreover, the response time for support requests is too long, leading to a poor customer experience.

Talk to your data with Gen AI

The client uses several cloud technologies, including Snowflake, as the core database and data warehouse solution. Capgemini experts were quick to consider Snowflake Cortex AI technology as the key to creating a cutting-edge solution for tomorrow, addressing the client’s issues with operational costs.

Why not ramp down on operational costs and ramp up customer interactions to a new level like this?

1. Go straight to the chatbot and ask the question.
2. Still have a question? Ask another question.
3. Is the chatbot not able to answer your question? Contact a relevant person from the platform team.

With this vision in mind, Capgemini set out to implement a Gen AI-based chatbot that could answer customer questions efficiently. The chatbot, powered by the company’s extensive knowledge repositories, ensured that the provided answers were accurate and relevant. Additionally, the chatbot referenced the source Wiki documentation link as part of its responses, making it easier for users to find the information they needed.

The solution worked 24/7, ensuring that customers could get help at any time of the day or night. This innovative approach aimed to reduce the burden on the customer support team and enhance the overall customer experience. By leveraging the power of Cortex AI and Retrieval-augmented generation “RAG”-based Gen AI, Capgemini was poised to revolutionize how customer support was handled, paving the way for a more efficient future.

High-level architecture

The RAG architecture Capgemini proposed for the Cortex AI chatbot consisted of three service types:

•      Cortex AI Functions for large language model (LLM) support: EMBED_TEXT_768, VECTOR_L2_DISTANCE, and COMPLETE)
• Snowpark Container Services for retrieval front-end.
• Snowflake tables as a vector store (native support of vectors as data types in Snowflake).

Let me explain some basic terms:

• RAG is an architectural approach that enhances the capabilities of large language models by incorporating an information retrieval system. This system retrieves relevant data or documents and provides them as context for the LLM, improving the accuracy and relevance of the generated responses.
• Snowflake Cortex AI is an intelligent, fully managed service within Snowflake that allows businesses to leverage the power of artificial intelligence (AI) that enables users to quickly analyze data and build AI applications without the need for extensive technical expertise.
• Snowflake Cortex AI Functions are a set of pre-built LLM functions that allow users to perform advanced data analysis and AI tasks directly within the Snowflake platform. These functions include capabilities such as text completion, sentiment analysis, and text summarization.
• Snowflake Container Services is a fully managed container offering that allows users to deploy, manage, and scale containerized applications within the Snowflake data platform.

By implementing a Gen AI chatbot based on Snowflake Cortex AI technology, evaluated by Capgemini, the client can streamline the customer support processes, reduce operational costs, and enhance customer interactions. This innovative solution leverages the power of AI to provide accurate and timely answers, ensuring that users can easily navigate through vast knowledge bases and find the information they need.

Cortex Search

I described the way Capgemini built a search tool for the client’s use case. The latest introduction of Cortex Search replaces the need for standalone vector tables and a self-managed embedding process with a fully managed RAG engine. This advancement not only streamlines development but also elevates the quality of outcomes with sophisticated retrieval and ranking techniques that merge semantic and lexical search. This effective approach is undoubtedly a game changer in building Gen AI RAG-based solutions.

Capgemini and Snowflake

The collaboration between Capgemini and Snowflake leverages Snowflake’s AI data cloud to enable businesses to unify and connect to a single copy of all data with ease. This partnership allows for the creation of collaborative data ecosystems, where businesses can effortlessly share and consume shared data and data services.

Capgemini and Snowflake are collaborating to develop generative AI solutions that leverage Snowflake’s advanced AI Data Cloud technology to drive innovation and enhance business outcomes across various industries.

This strategic relationship has led to Snowflake naming Capgemini the 2023 EMEA Global SI Partner of the Year.

Last year, we predicted that post-quantum cryptography (PQC) would be a defining technology trend in 2024 with far-reaching implications for organizations.

Following the release of NIST’s post-quantum encryption standards a few weeks ago, the race to secure IT systems for the quantum era has accelerated. Nowhere is this urgency more pronounced than in the financial sector, where sensitive data, stringent regulations, and vast datasets demand a rapid shift to quantum-safe systems

As #quantum computing advances, it presents both opportunities and risks for the financial sector. On one hand, quantum computing could revolutionize financial processes such as market trading, risk management, and secure communication through technologies like quantum key distribution. On the other hand, it could create significant exposure, particularly to public-key cryptography, which underpins the security of digital communications. Cyber actors may use quantum computers to break current encryption methods, creating a scenario where sensitive financial data becomes vulnerable. The concept of “harvest now, decrypt later” is particularly concerning, as threat actors might intercept encrypted data today, with the aim of decrypting it once quantum computers mature.

Recently, the G7 Cyber Expert Group published a very interesting statement that highlights the dual nature of this technology.

The G7 is urging financial institutions to start planning for post-quantum cryptography (PQC) as soon as possible to safeguard future communications. Financial institutions are encouraged to assess their own quantum risks, build inventories of vulnerable systems, and implement governance processes to mitigate emerging threats.

Personally, I would go beyond the G7’s recommendations and urge organizations across all sectors to start investigating and navigating the complex quantum landscape.

There are many actions that CxOs can take today to start preparing for a quantum future: such as auditing current cryptographic systems, investing in quantum-resistant algorithms, and ensuring that quantum readiness is integrated into long-term IT roadmaps.

Quantum computing is advancing faster than initially predicted, and when it reaches critical maturity, it will be too late to start preparing. Post-quantum cryptography implementation will not be easy, so a gradual migration with careful planning will be essential. Starting now will prevent unwelcome surprises and allow an orderly migration. The actions we take today will determine whether we are resilient or exposed when quantum supremacy becomes a reality.

The modern HR landscape is rapidly evolving, and at the heart of this transformation is the pursuit of exceptional Employee Experience(EX).

EX is not only a key driver of employee satisfaction, retention, and productivity, it is also a strategic advantage that enables organizations to attract and retain the best talent in the market. As a result, HR leaders are increasingly seeking innovative AI-based solutions to make EX a reality in their own organizations.

According to the latest Zendesk EX Trends Report, “81% of employee service leaders believe AI enables employees to handle complex tasks effectively, while 79% think it enhances their competitive advantage.”

Couple this with the rise of Generative AI (GenAI) and its ability to transform how HR teams deliver personalized and engaging experiences to employees, and organizations are left with a need to foster better employee interactions to keep pace in the market.

Capgemini and Zendesk’s strategic partnership delivers all of this to HR teams by combining Capgemini’s strategy and transformation expertise with Zendesk’s cutting-edge technology. Ultimately delivering effective and empathetic HR solutions to clients, that ensure employees feel valued and supported, leading to increased satisfaction, retention, and productivity rates across any organization.

Transformation with innovation at the core

In an era where change is the only constant, Capgemini’s Intelligent People Operations (IPO) solution stands at the forefront of HR transformation, championing a future where innovation is not just a buzzword, but the driving force behind every strategic move.

The benefits of this transformation are manifold, as they impact not only the efficiency and effectiveness of the HR function, but also the overall performance and culture of the organization:

• Organizations that leverage Capgemini IPO solution can expect to see a significant reduction in operational costs, through the value and optimization it delivers to HR processes through its GenAI capabilities
• Additionally, a transformed HR function also means a more engaged workforce, with employees benefiting from more personalized and seamless interactions with HR, leading to higher employee satisfaction and retention rates which are critical in today’s competitive talent market
• While a transformed HR function enables organizations to become more agile and responsive to the changing needs and expectations of their employees, customers, and stakeholders. By leveraging data and analytics, organizations gain insights into their workforce and talent trends and can leverage them to make informed decisions and drive innovation.

Zendesk’s role in this partnership is pivotal. Its robust customer service platform is the perfect complement to Capgemini’s transformation strategy, providing the technological backbone needed to support a modern, digital-first HR function.

With Zendesk’s tools, organizations can deliver a more personalized and efficient service to their employees, fostering a culture of continuous improvement and innovation across any HR team.

A journey of transformation: Building lasting partnerships

Embarking on a transformation journey is a voyage that requires not just a clear vision and a strategic map, but also a committed partner who can navigate through the complexity of change at speed.

This is where the enduring partnership between Capgemini and Zendesk becomes invaluable. Together, we stand as reliable partners on any transformation journey, ensuring that every step moves organizations towards achieving their long-term goals.

Capgemini and Zendesk understand the challenges of this journey and are committed to helping our clients benefit from more agile, resilient, and equipped HR operations, enabling them to handle the demands of today’s ever-changing business landscape.

Together, we do not just help organization reach their transformation goals, we also help them thrive during the journey.

What are multi-agent systems?

Multi-agent systems with AI are those systems where autonomous agents are equipped with AI capabilities, working together to achieve the desired outcome. An agent in this context can be as generic as an entity which is acting on another entity’s behalf. In multi-agent AI systems, AI agents (bots) cooperate to help achieve the goals of people owning processes and tasks.

How do tokens work?

Put simply, a token is a piece of a word or text that can be used as input for a large language model (LLM) like ChatGPT. All passages of text are broken into tokens, but not every word is a token, some are broken down further. For example, the word “chat” is one token, but a longer word like “tokenization” might be broken into multiple tokens.

When you input text into ChatGPT, for example, the text is converted into a sequence of tokens in a process called tokenization. The model processes these tokens and generates a sequence of output tokens, which are then converted back into text.

Why does understanding tokenization matter?

The reason why it matters is because models have token limits, and models also have pricing determined by the number of tokens in the input and output.

Models like GPT-3.5-turbo have a maximum number of tokens they can process in a single request. For instance, GPT-3.5-turbo can handle up to 4096 tokens which is around 3,000 words (both input and output combined). These limits are put on the models to ensure they work effectively and can respond quickly.

The number of tokens processed affects the computational resources required and the cost of using the mode, so the more tokens, the more cost.

Token limits are one of the reasons why retrieval-augmented generation (RAG) involves the use of traditional search tools to help subset the relevant information into the prompt to enable vast quantities of information to be processed in efficient ways.

What are the costs?

The costs on paper may not seem high, but they can mount up especially in multi-agent systems. Let’s explore the costs a little.

Here are the costs of two common models offered by OpenAI. You can see that there are substantial differences in cost between the two models (prices correct August 2024).

What do tokens look like in reality and how do they count up?

Here is a single-agent example with a total of 13 input tokens and 15 output tokens, which is not many:

Prompt: “I think we should replace the sensor immediately to avoid any issues.” (13 tokens)

Answer: “I agree, let’s replace the sensor right away to prevent any problems.” (15 tokens)

Here is a multi-agent example with a total of 1005 input tokens (approximately 77 times as many as the single-agent example because of the need to input the responses of all agents to all agents each round) and 153 output tokens (around 10 times as many as the single-agent example):

Round 1 prompt: “I think we should replace the sensor immediately to avoid any issues.” (13 tokens × 5 agents = 65 tokens)

Round 1 answers: Initial opinions

• Agent 1: “I think we should replace the sensor immediately to avoid any issues.” (13 tokens)
• Agent 2: “I suggest we first run a diagnostic test to determine the fault.” (13 tokens)
• Agent 3: “We might need to recalibrate the sensor instead of replacing it.” (11 tokens)
• Agent 4: “Let’s review the sensor’s error logs before deciding.” (10 tokens)
• Agent 5: “Perhaps the issue is with the wiring, not the sensor itself.” (11 tokens)

Round 2 prompt: Includes all the past answers from round 1 and the initial prompt (71 tokens × 5 agents = 355 tokens).

Round 2 answers: Arguments and counterarguments

• Agent 1: “Replacing the sensor ensures we eliminate the fault quickly.” (10 tokens)
• Agent 2: “Diagnostics can reveal if the sensor really needs replacement.” (10 tokens)
• Agent 3: “Recalibration is cost-effective and might fix the issue.” (9 tokens)
• Agent 4: “Error logs can provide insights into the root cause.” (8 tokens)
• Agent 5: “Checking wiring first can save time if it’s not the sensor.” (9 tokens)

Round 3 prompt: Includes all the past answers from rounds 1 and 2 and the initial prompt (117 tokens × 5 agents = 585 tokens).

Round 3 answers: Further discussion and attempt to reach consensus.

• Agent 1: “If diagnostics show a major fault, we can then replace it.” (12 tokens)
• Agent 2: “Agreed, diagnostics first, then decide the next steps.” (9 tokens)
• Agent 3: “Can we perform diagnostics and recalibration simultaneously?” (10 tokens)
• Agent 4: “Yes, and while doing so, review the error logs.” (9 tokens)
• Agent 5: “And inspect the wiring as a part of the diagnostics.” (9 tokens)

Conclusions and the challenges of scale

We can see that the multi-agent example uses 1005 input tokens and 153 output tokens, which is a huge increase on the 13 input tokens and 15 output tokens of the single agent. While the value of the multi-agent system may be worth the extra cost for the additional value in response, we can see that costs can increase quickly when they are determined by token volume, and therefore, the architectural design of multi-agent systems should consider this spiralling token use and cost. This becomes a much bigger issue when we have proactive agents seeking out work and having discussions with each other frequently.

If we have one sensor and we need to make a decision every one minute, what do these costs look like?  

• Single-agent (tokens per day and cost):
  • Input tokens: 13 × 60 × 24 = 18,720 (GPT-4o non-API 18,720 * ($5 / 1,000,000) = $0.09)
  • Output tokens: 15 × 60 × 24 = 21,600 (GPT-4o non-API 21,600 * ($15 / 1,000,000) = $0.32)
• Multi-agent (tokens per day and cost):
  • Input tokens: 1005 × 60 × 24 = 1,447,200 (GPT-4o non-API 1,447,200 * ($5 / 1,000,000) = $7.24)
  • Output tokens: 153 × 60 × 24 = 220,320 (GPT-4o non-API 220,320 * ($15 / 1,000,000) = $3.30)

So, we see the cost per day of these two systems being $0.41 for the single-agent and $10.54 for the multi-agent system which is approximately 26 times more expensive. The difference in cost becomes even greater when viewed by week or month and the number of sensors may well push volume and costs up even further. So, do we abandon multi-agent systems, or can we mitigate these spiralling token costs?

Top tips

• Use GPT3.5 turbo instead of GPT4. This is a good option for simple tasks, and we already see that costs can be much lower for simpler models.
• Use a model hosted by someone for free. This can be offered from services such as Groq.
• Use a local model such as LLaMA 7B. This involves downloading a model and running it locally so the compute costs are on your own infrastructure and, therefore, can be managed yourself and could be cheaper. However, simple/smaller LLMs are those available for download today, so a compromise on performance might have to be made for this option.
• Use token limits. A lot of LLMs have settings for limiting the output tokens of an LLM and this can have a significant downstream effect especially if you are giving the entire dialogue to the next agent in a multi-agent system.
• Be careful when you use applications like CrewAI, as they employ a quality and context self-checking and updating mechanism that updates the context and runs a query to check if the agent has answered the question properly. This can double all the token use in the system as well.

About Generative AI Lab:

We are the Generative AI Lab, expert partners that help you confidently visualize and pursue a better, sustainable, and trusted AI-enabled future. We do this by understanding, pre-empting, and harnessing emerging trends and technologies. Ultimately, making possible trustworthy and reliable AI that triggers your imagination, enhances your productivity, and increases your efficiency. We will support you with the business challenges you know about and the emerging ones you will need to know to succeed in the future. One of our three key focus areas is multi-agent systems, alongside small language models (SLM) and hybridAI. This blog is part of a series of blogs, Points of View (POVs) and demos around multi-agency to start a conversation about how multi-agency will impact us in the future. For more information on the AI Lab and more of the work we have done visit this page: AI Lab.

How Capgemini customers get the benefit of a partnered approach to Gen AI innovation and scaling

Effectively rolling out and scaling generative AI initiatives to be impactful to your business expectations can sometimes be like trying to navigate a maze in the dark. Gen AI is full of promise but, as the technology and use cases evolve very fast and often shift direction, marketing efforts tend to amplify its complexity by overblowing it. This results in distinct challenges for organizations seeking immediate value, including finding the most impactful use case relevant to your industry, trying to ease or disrupt your own processes while staying within safe guardrails, building business cases, data management and security, and the availability of skilled resources and talent to make it happen. These are true challenges. 

Yet enterprises are forging ahead: our research shows that 80 percent of organizations have increased their investment in Gen AI in the last year, and it’s starting to pay off. Some organizations have seen productivity boosts of up to 25 percent over the past year. 

Businesses are exploring innovative ways to harness Gen AI’s capabilities and chase business value. In a recent conversation with Google Cloud CEO Thomas Kurian, Capgemini Group CEO Aiman Ezzat revealed four areas in which Gen AI technology will drive the most organizational value: streamlining and improving core internal processes, improving productivity, innovating customer interfaces, and building new customer experiences. 

But to fully harness this value, organizations need access to the right resources: proven Gen AI use cases, ability to demonstrate business ROI, guardrails and scalable plans for the in-production phase, skilled advisors and engineers, and of course a strong data foundation. According to our research, only 51 percent of data executives say their companies have processes in place to manage data integration, and only 49 percent of data sources are exploitable to support Gen AI scenarios efficiently. The rest are siloed or reside in local servers, presenting data accessibility challenges.

Working to solve these challenges is essential to unlocking value with Gen AI. Partnering with Google Cloud has been effective at accelerating value creation. Here’s how we’re doing it, and the benefit it is delivering to our customers. 

Growing global talent and Gen AI solutions 

The first step is mastering the technology and understanding its roadmap. 

We’ve worked with our partners at Google Cloud to understand how best to harness the power of this technology for our customers since Gen AI first showed promise. We’re committed to helping clients embrace this potential, and we’ve welcomed the opportunity to expand our expertise and offerings – and have seen market demand rise sharply in response.

We invested in the creation of a global Center of Excellence focused on Gen AI, which brings together our subject matter experts from our practices ranging from strategy to data science to software engineering, to cover all angles of Gen AI applications. This deeper relationship with Google Cloud has given us an inside view of how Gen AI is evolving. For example, we participated in Google’s trusted tester program, and enjoyed early access to Gen AI technologies and solutions ahead of their market releases. This means we can better take our clients’ needs into consideration.   

Next, we needed to act on this knowledge at scale. We created a series of hackathons designed to produce innovations, use cases, and Gen AI MVPs in a compressed time frame. We then turned these results into a suite of demos and horizontal and vertical accelerators, to support our clients in exploring concrete Gen AI possibilities.  

When Google launched Gemini at the end of 2023, Capgemini responded by launching Gemini Week. In doing so we reached our goal of tripling the number of our team members who could help us meet the market demand for Gemini-powered applications.   

By the end of June 2024, our specialization in Google’s Gen AI platform was truly cemented when Google Cloud invited Capgemini to undertake its Gen AI Specialization designation. Achieving the specialization required an extensive third-party assessment process, and we’re proud to say we were among the first of Google’s partners to have achieved it.   

Our year of Google-powered Gen AI partnership successes was capped off with the award for Google Cloud Global Industry Solution Partner of the Year for Gen AI services.  

Between the new Gen AI Specialization and the Partner of the Year designations, we’ve been audited to truly qualify and quantify our expertise and customer impact. These steps have sharpened our assets to bring maximum value to customers.  

RAISEing the baseline on Gen AI  

Enterprises need a reliable accelerator that supports the industrialization at scale of their Gen AI projects. The risk is that the ecosystem can become disconnected, with multi-agent systems evolving across multiple platforms and without cost controls and consistent trust across the process. That’s why we developed Capgemini RAISE (Reliable AI Solution Engineering) to bring disparate solutions under one roof. 

Built using Google Cloud native services, Capgemini RAISE augments Google’s existing solutions. It is based on our deep understanding of the platform and our breadth of domain experience. It is a modular solution that enables clients to customize tools to successfully scale any Gen AI workload.  

Capgemini RAISE comes with trust, cost, and scale controls, as well as reusable custom Gen AI templates and integrations to speed up Gen AI for enterprise adoption. With these building blocks, Capgemini RAISE provides additional value on top of GCP native services for model training, evaluation and deployment, prompt libraries and optimization, chatbots, and more. Its suite of modular services means clients can pick and choose the most relevant tools to ensure they achieve the cost, scale, and trust factors they need.  

Our strategic partnership with Google Cloud and its Gen AI initiatives is working to drive value for our clients in the short- and long-term, enabling organizations to foster innovation and achieve their business transformation objectives. With the power of Conversational AI, we have already started to see our customers reap tangible business benefits. This collaboration is enabling organizations to foster innovation and achieve their business transformation objectives. We are excited to see how this will continue to evolve as generative AI advances. 

Interested in exploring Generative AI with Google Cloud Platform? Contact us for more information and an assessment – googlecloud.global@capgemini.com   

Learnings and best practices based on successful Google Analytics data platform implementations 

Effectively leveraging enterprise data is one of the best ways to grow a business. Data-driven  businesses are 23 times more likely to surpass their competitors when it comes to acquiring customers, nearly seven times more likely to retain them, and 19 times more likely to stay profitable. They’re also improving operational efficiencies and saving costs. 

An increasing number of organizations are successfully monetizing data and leveraging it to boost their top lines. In the last four years, organizations have, on average, improved in activating data, unlocking its value, and scaling infrastructure, platforms, and tools that enable them to leverage data more effectively.  

Today, nearly two in three executives agree that their organizations use activated data to introduce new products or services or to develop entirely new business models. But there are challenges when it comes time to actually implement cloud systems. 

Numbers from HFS Research show that only about one third of organizations are realizing cloud-implementation ambitions even though 65 percent have made strategic investments. So, how should companies build towards the results they need? 

Creating a structured and secure data ecosystem

The ability to share data across an ecosystem of stakeholders is a core component of turning data into insights and action that make it useful, marketable, and widely consumable. But many companies make one big mistake when moving to a cloud platform: they treat data the same way they did when it was on premises, overlooking opportunities to operate in new, more efficient ways. The beauty of cloud data platforms is their ability to manage data in ways that weren’t possible before – as long as the data ecosystem has been set up securely and effectively.  

A data analytics platform on the cloud manages, processes, and analyzes large amounts of data in a more scalable and flexible way than on-premises infrastructure, and businesses can use them for everything from data ingestion and storage to data processing, analytics, and visualization. Ultimately, cloud data platforms enable enterprise organizations to become data-powered enterprises thanks to better decision-making abilities and more efficient operations.  

Eight strategies for data analytics implementation

For the last 10 years, Capgemini has worked with Google Cloud Platform’s data analytics services and team to serve some of the world’s largest enterprises in retail, manufacturing, finance, and more to modernize data estates and leverage cloud and AI/ML tools to extract and share actionable insights – reducing costs and increasing top-line revenue. Here are eight of our key learnings. 

• Create a data platform architecture that drives success. Everything you can achieve with cloud-based data analytics will revolve around your core platform architecture. And to function effectively, three conditions must be met.  
  • The ability to host all data formats: This includes real-time and batch data.  
  • The segregation of business domains: These should be linked using a common key. 
  • The right file format: While most data is in CSV files, better formats include Avro and Parquet, as they both provide better compression and the fastest read and write experience to the database. 
• Data ingestion must be in a common format across all domains to enable reuse. Cloud data systems ingest new data much more efficiently and they enable data harvesting from sources such as Google Ads and social media platforms, which provides a more holistic view into customer behavior and feedback. But data ingestion must occur in a common format across all data domains. Again, I recommend going with Avro or Parquet.  
• Use cloud-native partitioning and indexing. In most cases, moving data to a cloud platform is not just a lift-and-shift operation. It requires a process of data transformation that converts one style of business logic into another. Many enterprises, when moving to the cloud, look to replicate legacy partitioning and indexing to their new cloud database. However, cloud-native databases such as BigQuery and Bigtable are designed differently than many older on-premises systems. Defining the business key and the partitioning are critical for optimizing the query, which leads to lower costs.
• Consider data security from the beginning. Data security is often omitted in the design phase when moving to a new platform, and this can lead to many issues down the road when the organizations have to meet security compliance objectives like GDPR and data privacy and access. Some tools integrate data security parameters right out of the box, such as Google DLP, an excellent tool to identify and secure PII data. This means data security has to be included as part of design and not bolted on as an afterthought.  
• Data lineage is key for resolving production issues, and has to be part of design. Data cataloging is critical to making data accessible across the entire organization. Sharing data products through a data catalog, so there is a consistent definition of the data products across all the dashboards and applications, ensures enterprise users are speaking the same language. 
• Enable generative AI. Is your analytics platform Gen AI-enabled? Gen AI is invaluable in helping organizations leverage the full potential of data analytics. It’s also useful for unlocking new insights and reducing time to market on new use cases. For example, BigQuery can use Google’s Gemini AI for more user friendly, prompt-based searches. 

AIOps design is now available to quickly identify the core issues leading to high cloud cost and quickly deploy the solution. (For example, poorly design BigQuery SQL can be quickly highlighted for the Ops team.) 

• Include FinOps and AIOps to BusinessOps. One of the major issues faced by customers is the high cost of moving to the cloud. This is often due to poorly designed FinOps modes, or lack of a design. FinOps has to be part of the design process so that we can then feed the usage data of the cloud services into the FinOps dashboard for active monitoring of the usage and cost parameters.
• Understand how your data analytics platform will drive value. Many organizations struggle to extract business value from their data. Realizing true value depends on the following technical requirements: 
  • Data domains must be linked using business keys, and this has to be part of the design 
  • Designing Data-as-a-Service or Data-Products-as-a-Service allows common definitions and reduces duplications 
  • Data platform design must include the ability to ingest new data sources  
  • Setting lifecycles for data products is important as new variations are created 
  • Many successful organizations provide an end of life for a data product. 

The critical factor in all of this is that choosing and implementing a data analytics platform is a long-term decision. In addition to developing the right architecture and databases for the current requirements, consider future use cases your organization may face five or 10 years from now. The ideal platform is robust and flexible enough to accommodate changing needs.  

For customers who do not want to wait for a new design, we typically recommend a re-platforming strategy. However, many organizations are choosing to combine both a new design and a re-platforming approach to achieve the best of both worlds. And that and our other advice gives them a much better chance of joining the group of companies that realize cloud-based ambitions. 

Interested in exploring Google Cloud Analytics data platform implementations? Contact us for more information and an assessment.   

A year ago, we predicted that agentic systems would be one of the main technology trends of 2024.

It seems our predictions were spot-on, with AI agents being announced everywhere and becoming the new business conversation topic (some hype maybe?).

Indeed, the concept of multi-agent systems, where multiple #AI agents interact and cooperate to achieve defined objectives is very promising. No longer limited to simple task execution, AI agents are now evolving towards greater autonomy, capable of making decisions, learning from their environments, and performing complex actions without continuous human intervention.

But as we step into this future, I can’t help but ask myself: What will govern the interactions between AI agents when they become autonomous?

To understand this question, we can draw a parallel with human social behavior. As individuals, our interactions are shaped by character, social norms, cultural values, learned behaviors, and a myriad of other rules that are implicitly followed by all (at least in theory!). These mechanisms allow us to collaborate, make decisions, and solve conflicts when we disagree.

AI currently lacks this framework to navigate complex and unexpected situations. As an interesting example, my friend Brett Bonthron shared how his driverless taxi got frozen in place when faced with the chaos of a traffic accident in front of it: https://lnkd.in/eGTCuMgS
An unexpected situation which would have been easily navigated by the average human utterly confounded the AI systems of his car (funnily enough, Brett eventually had to exit his driverless taxi and call for a good-old human driven one).

In the future, what will happen when several AI Agents run into each other and that they have to get to a clear outcome but their assigned tasks happen to be in contradiction? Who will go first? Who will have to step back and give priority to the other?

If you want to learn more about this, our colleague Jonathan Aston from our Capgemini Generative AI Lab recently posted a very interesting piece exploring how Game Theory may provide some of the answers:
https://lnkd.in/e_efTnY9

In the physical world, individuals essentially follow three main tracks to resolve such conflicts: we endorse the rules of courtesy, we negotiate, or we go to war (figuratively or not). Will AI agents follow a similar reasoning?