ISG Provider Lens™ Generative AI Services - Strategy and Consulting Services - Large Providers - Global 2024

GenAI-led disruption is inevitable, but its success needs a thoughtful, data and human-centric strategy

The GenAI market is experiencing a transformative period of rapid growth, unlocking new possibilities for enterprises to innovate and streamline operations. From an enterprise standpoint, this growth signals a pressing need to invest in GenAI now to stay competitive. Businesses understand that the automation and efficiencies offered by AI can be  leveraged to reduce operational costs, improve accuracy in tasks and drive new revenue streams through innovative products and services.

• For enterprises, the focus is on identifying the right use cases that deliver clear RoI and align with long-term strategic goals such as enhancing CX or improving operational efficiency.
• For decision-makers, it is essential to invest in adaptable AI technologies that can evolve as a business grows, ensuring sustainability and relevance in the long term.

The current state of the GenAI market — PoC to delivery

In the past 12 months, the GenAI landscape has witnessed a significant surge in the number of projects — right from PoCs and minimum viable products (MVPs) to full-scale deployment pipelines. Enterprises across industries are increasingly exploring GenAI to handle complex challenges, enhance operational efficiencies and drive innovation. The initial PoC phase is critical for them to evaluate the potential of GenAI in solving specific business problems. PoCs allow organizations to assess the feasibility and practicality of GenAI applications before committing to substantial investments. They help organizations develop a functional version of the AI application that can be tested in realworld conditions. As an increasing number of enterprises experience the benefits of AI, the pipeline of GenAI projects will continue to grow, leading to a future where AI will be deeply embedded in business operations. The successful implementation of these projects will not only drive business outcomes but also shape the future of innovation across industries.

Developments across modalities and applications

The field of GenAI will progress rapidly both in terms of research and commercialization, but use cases are emerging rapidly in the enterprise landscape. Some of the recent GenAI applications have proven how this new-age technology can help with innovation and creativity, indicating usability for both businesses and individuals. While the current GenAI landscape is dominated by text-based and bot applications, there is a growing anticipation for other modalities — such as image, video, data and audio-based applications — to reach their full potential in the near future.

Text-based applications have seen the most rapid development in the field of GenAI, primarily due to their simple interface, rapid RoI and immense utility. Chatbots, in particular, have become a ubiquitous tool across industries. Powered by large language models (LLMs) such as OpenAI’s GPT, chatbots can provide personalized assistance, customer support and automated communication at an unprecedented scale. Companies use chatbots/copilots to enhance customer service, streamline business operations and facilitate conversation. For instance, GenAIdriven chatbots are revolutionizing customer service centers, enabling businesses to provide 24/7 support, instant issue resolution and multilingual capabilities — all while reducing human workload.

However, despite the significant growth in text-based AI, the full scope of GenAI’s potential remains untapped as other modalities — such as image, audio, video, and data-based applications — are still evolving toward mainstream adoption. The full potential of multimodal GenAI is yet to be realized, but the ongoing advances in neural networks and deep learning indicate that the future is bright. As models become increasingly capable of handling diverse inputs — text, images, audio, video and data — applications will become considerably integrated and sophisticated.

Model development using RAG versus finetune  versus SLM and LLM

The development of GenAI applications is undergoing a significant transformation, with IT service providers at the forefront of this change. While current methods such as retrieval-augmented generation (RAG) and fine-tuning LLMs have proven to be effective, the future points toward a more specialized approach. Fine-tuning LLMs will remain essential, but the focus will increasingly shift to the development of small, customized language models that cater to specific applications. As AI technologies mature, there is growing recognition that small, more customized language models — fine-tuned for specific applications — can provide significant advantages in terms of cost, scalability and performance. Small, domain-specific versions of LLMs can perform equally well in targeted tasks without requiring the massive infrastructure needed to train and deploy full-scale LLMs.

In the future, IT service providers are likely to adopt a hybrid approach, leveraging the strengths of each method, including finetuning, RAG approach, LLMs and small language models (SLMs). For instance, LLMs could be used for general tasks or applications that require broad linguistic capabilities, while SLMs could be deployed for more focused and specialized tasks. This hybrid strategy will offer a balance between scale, cost-efficiency and task-specific performance, enabling providers to offer GenAI solutions that are both powerful and practical.

Implementation using bespoke versus existing accelerators

GenAI has taken center stage in transforming industries by automating tasks, generating content and providing intelligent insights. While open-source GenAI models offer broad capabilities and accessibility, enterprises are increasingly gravitating toward bespoke approaches and customizations tailored to their unique needs. A bespoke approach to AI development entails creating customized GenAI models that are fine-tuned, secure and optimized for specific business use cases. This shift is driven by various factors such as the need for privacy and security, specialization of tasks, cost-efficiency requirements and the ability to integrate AI seamlessly into existing operations. While enterprises are focusing on customized GenAI models, they are not starting from scratch; many organizations are leveraging existing AI platforms, toolsets and accelerators to streamline the development process. These resources provide a foundation upon which bespoke GenAI solutions can be built, reducing development time and costs.

While the technology outcomes from Gen AI are driving increased adoption, enterprises continue to face key challenges that must be addressed. Service providers play a crucial role in overcoming these barriers, enabling organizations to fully harness the potential of Gen AI. By addressing these hurdles, adoption can be accelerated, and the transformative benefits of the technology can be realized more broadly.

Enterprise challenges and provider recommendations

As GenAI continues to reshape industries, enterprises are increasingly looking to leverage its capabilities to achieve innovation, operational efficiency and a competitive advantage. While the potential of GenAI is significant, the journey to adoption is fraught with challenges. From data privacy concerns to integration complexities, the road to realizing AI’s full potential can be a difficult one. Enterprises face mounting challenges in assessing business needs, evaluating RoI, ensuring technology compatibility and prioritizing change management. By partnering with relevant service providers, enterprises can make informed decisions that maximize the impact of their AI initiatives, while addressing the following challenges.

1. Data strategy and governance challenges

To implement GenAI successfully, enterprises must have a well-defined data strategy and governance framework as AI systems rely heavily on high-quality, well-organized data to generate accurate predictions, insights and solutions. Enterprises that fail to establish strong data governance face issues such as data silos, poor data quality and  compliance breaches — all of which can derail AI projects.

2. Data availability and accessibility

For GenAI models to function effectively, enterprises need to ensure that relevant data is accessible and available in sufficient volumes. AI thrives on large datasets, and enterprises must collect, store and organize data across various functions and systems to fuel AI initiatives. Several enterprises struggle with siloed data across different business units, which can deter AI adoption.

Provider contribution needed to mitigate the above challenges

• Breaking down data silos: Providers should offer solutions that enable seamless integration of data from multiple sources, including CRM systems, ERP platforms and IoT devices. GenAI solutions need access to real-time and historical data. Therefore, it is essential for enterprises to implement data architectures that promote sharing and accessibility across teams.
• Building a unified data infrastructure: Providers should support enterprises by recommending the adoption of unified data infrastructures such as data lakes or data warehouses to centralize data and ensure it is available for AI consumption. Cloudbased solutions often provide the flexibility and scalability needed for this purpose. By organizing data into a unified structure, enterprises can optimize AI performance and minimize delays in data processing.

3. Data quality management

The quality of the data fed into GenAI models directly impacts their performance. Poor-quality data can lead to inaccurate insights, flawed predictions and biased outcomes. Enterprises need a robust data quality management strategy to ensure that the data used by GenAI models is accurate, complete and up to date. To this end, they need mechanisms for data cleansing and validation. Data bias is a significant risk when implementing AI systems. Biased data can lead to biased AI outcomes, which can have serious implications on decision-making, particularly in areas such as recruitment, financial functions and customer segmentation.

Provider intervention required to mitigate the above challenges

• Ensuring data accuracy and completeness: Providers should offer data cleansing and preprocessing tools that automatically detect and fix issues such as missing data, duplicate entries or inconsistencies. Additionally, AI systems can be integrated with real-time validation tools to continuously assess and improve data quality.
• Managing data bias: Providers should support enterprises by offering tools and frameworks to detect and mitigate bias in data, ensuring that GenAI models remain fair and inclusive.

By focusing on data quality management, enterprises can ensure that their GenAI models are delivering trustworthy and actionable insights.

4. Data privacy and security concerns

GenAI models require large datasets, which often include sensitive information, for training and deployment. For enterprises in regulated industries such as healthcare, finance and retail, protecting customer and business data,
while complying with stringent regulations such as EU AI Act, GDPR, HIPAA, and CCPA is critical. Data protection should be at the core of the solution adopted, with encryption protocols for data storage, processing and transfer. Data breaches or incorrect handling can result in legal repercussions, hefty fines and damage to a company’s reputation. Additionally, AI-driven systems must be designed to meet user privacy standards, especially when data is collected and used for personalized outputs or decisions.

Provider arbitration needed to deal with the above challenges

• Compliance-driven GenAI solutions: Service providers should prioritize data privacy by ensuring their GenAI models are compliant with local and global regulations such as GDPR and EU AI Act in the European Union or HIPAA for healthcare data in the U.S. AI systems should be designed to respect consent and allow enterprises to manage data according to regulatory guidelines.
• End-to-end encryption: Providers should implement anonymization or tokenization methods to ensure that sensitive data cannot be misused even if unfairly accessed.
• Secure cloud infrastructure: AI systems deployed in cloud environments should leverage secure cloud services with strong identity and access management (IAM) tools, ensuring that only authorized users can access sensitive data. Providers should align with industry-standard security certifications such as ISO/IEC 27001 and SOC 2 to build trust with enterprises.
• Explainable AI (XAI): To address the concern of data misuse, providers should develop explainable GenAI solutions that offer transparency into how decisions are made. For example, financial institutions would benefit from AI systems that can explain why certain credit decisions were made based on customer data, reducing the risk of discriminatory practices.

5. Integration with existing systems and workflows

Several enterprises have complex, legacy IT infrastructures. Incorporating GenAI systems into these environments can be challenging due to compatibility issues, disruption to established workflows and the cost of overhauling existing technology stacks. As GenAI solutions often require input from multiple data sources, ensuring  interoperability between different data formats and systems is critical. Without the right integration, enterprises risk failed AI implementations that do not deliver the desired value, causing inefficiencies.

Provider support to mitigate the above challenges

• Customizable AI platforms: Service providers should offer highly flexible, modular AI platforms that allow enterprises to tailor solutions to their existing environments, without the need for extensive system overhauls. This advantage would enable enterprises to plug GenAI solutions into their workflows, ensuring minimal disruption.
• Comprehensive API integration: Providers should build GenAI solutions with robust APIs and middleware support to integrate seamlessly with CRM, ERP, HR and other enterprise functions. This would enable data to flow smoothly between AI tools and legacy systems.
• Consultative approach to integration: Providers should offer integration support services to assist enterprises throughout the process. This support could include mapping existing workflows, identifying integration points and conducting phased rollouts to ensure that a GenAI system harmonizes with ongoing operations without disrupting day-to-day activities.
• Data interoperability: Providers should offer solutions that can standardize and harmonize data inputs, preventing data silos and improving overall system efficiency.

6. Lack of AI expertise and talent

The shortage of skilled AI professionals, such as data scientists, ML engineers and AI ethicists, makes it difficult for enterprises to build, deploy and maintain GenAI solutions internally. This talent gap can slow down AI adoption or lead to poorly managed implementations that fail to deliver value. Additionally, many business users and  decisionmakers lack the technical understanding of AI capabilities, limiting the organization’s ability to capitalize on GenAI investments.

Provider intervention to address the above challenge

• Training and education programs: Providers should offer AI training and certification programs for non-technical
  employees and business leaders, thereby increasing AI literacy within an organization. Training should focus on
  helping employees understand AI concepts, interpret AI-generated insights and make data-driven decisions.
• Managed AI services: To address talent shortage, providers should offer fully managed AI services, where they take responsibility for developing, deploying and maintaining GenAI models on behalf of enterprises. These services allow enterprises to benefit from AI without the need for in-house expertise.
• AI-as-a-service platforms: Providers should develop self-service AI platforms that abstract the technical complexities of model training and deployment. These platforms should be no-code/low-code, enabling users to interact with AI tools, customize models and run AI-driven processes without the need for deep technical knowledge.
• Prebuilt GenAI models: Providers should offer pretrained, industry-specific GenAI models that enterprises can deploy with minimal customization. For example, a financial services company might use a pretrained fraud detection model, requiring only minor adjustments to suit its specific data and regulatory context.

7. High initial costs and unclear RoI

GenAI solutions often involve significant upfront investments in technology, infrastructure and talent. For enterprises,
especially the ones with tight budgets, this can be a deterrent to AI adoption. Compounding this challenge is the uncertainty around RoI from AI projects. In many cases, enterprises struggle to quantify the long-term benefits or see the immediate value, leading to a reluctance to commit to large AI initiatives.

Provider arbitration to handle the above challenge

• Flexible pricing models: Providers should ease the cost burden by offering flexible pricing options such as pay-as-you-go, subscription models or usage-based pricing. Such options enable enterprises to scale their GenAI investments gradually, based on performance, rather than face prohibitive upfront costs.
• PoC programs: Providers should offer low-cost or risk-free PoC programs to allow enterprises to assess the feasibility of GenAI solutions in their real business environment. This would help enterprises evaluate RoI before making significant financial commitments.
• RoI benchmarking and business case development: Providers should help enterprises develop business cases for GenAI investments by offering clear benchmarks from previous deployments. RoI examples from similar industries and use cases can provide guidance and confidence to decision-makers considering GenAI investments.
• Phased implementations: Service providers should recommend phased or modular AI adoption approaches. This flexibility would allow enterprises to start small, with pilot projects, and gradually scale up AI initiatives as measurable benefits become evident, thereby managing financial risks.

8. Ethical and bias concerns

GenAI systems can unintentionally generate biased or unfair outputs, which can lead to ethical concerns, particularly in sectors such as recruitment, finance and healthcare. For example, GenAI models trained on biased data may perpetuate existing stereotypes, leading to discriminatory practices in hiring or lending decisions. Enterprises fear that deploying biased GenAI models can result in legal and credibility risks, especially with growing regulations around the ethical use of AI.

Provider arbitration needed to intercept the above challenge

• Bias detection and mitigation: Providers should build GenAI solutions with integrated bias detection tools that can analyze data for potential biases and ensure that the models’ outputs are fair and equitable. Offering transparency in the way GenAI models are trained and continuously monitoring them for such biases is essential.
• Fairness audits: Providers should offer services such as fairness audits, where GenAI models are regularly tested for fairness and adherence to ethical guidelines. Enterprises can receive reports on model behavior, ensuring that AI tools align with a company’s values and compliance requirements.
• Explainability and transparency: Providers should develop AI systems that can explain their decision-making processes in ways that are easy to understand. This facility would allow enterprises to evaluate the fairness of AI outputs and identify any potential bias. Explainable AI is particularly valuable in regulated industries, where  accountability is crucial.
• Ethical AI frameworks: Providers should offer GenAI solutions that are built around ethical AI principles, emphasizing fairness, transparency and accountability. For example, GenAI solutions could include builtin safeguards to prevent biased outputs and provide enterprises with the tools to audit and modify models over time.
• Ensuring accountability: Providers should help enterprises with the creation of AI ethics committees and governance boards, ensuring that decisions around AI use are made with ethical considerations in mind.

9. Change management concerns

Adopting GenAI often leads to significant changes in the way an enterprise operates, making it essential to address organizational resistance and implement strong change management strategies. The adoption of AI can lead to concerns about job security, shifts in work responsibilities and apprehension about innovative technologies.  Enterprises must manage these fears, while ensuring a smooth transition to AI-powered operations.

A primary concern among employees in the context of AI is the fear of job displacement, with the notion that AI will automate their tasks and make their roles redundant. Providers need to help enterprises frame AI as a tool for workforce augmentation rather than replacement. AI has the potential to automate repetitive, low-value tasks, freeing employees to focus on high-value, strategic work. It is critical for enterprises to communicate this message clearly and frequently to their workforce. Employees are more likely to embrace AI when they understand the need
for AI implementation and its alignment with business goals.

Provider support required to deal with the above challenge

• Reframing AI as an enabler of productivity: Providers should help enterprises position AI as a tool that enhances human capabilities rather than replace them. For example, AI can assist customer service agents by handling routine inquiries, allowing agents to focus on complex or high-touch interactions. This narrative helps employees understand that AI will improve their day-to-day work rather than eliminate their roles.
• Highlighting job transformation examples: Providers should support this messaging that AI can bring about job transformation by showcasing case studies where AI has done so without causing job losses. For instance, companies in sectors such as healthcare, finance or manufacturing might use AI to reduce manual data entry efforts, enabling employees to focus on critical tasks such as decision-making, customer care or creative problem-solving.

Overcoming these challenges will enable organizations to unlock the full potential of Gen AI more effectively and at a faster pace. Service providers that have successfully addressed these issues are already setting new trends, which are expected to significantly shape the Gen AI landscape in 2025 and beyond.

Key trends impacting 2025

• GenAI is moving beyond text to embrace multimodal AI models that can process and generate outputs across various data types such as text, images, audio and video. This capability allows for more robust and sophisticated AI applications across industries.
• As the market matures, there is a growing demand for domain-specific AI models tailored to industries such as healthcare, finance, legal and manufacturing. These models are trained on specialized datasets to meet the nuanced needs and regulatory requirements of each sector.
• The democratization of AI is a key trend, with no-code/low-code platforms allowing non-technical users to leverage GenAI. These platforms provide user-friendly interfaces for creating and deploying AI models, without the need for deep technical expertise, enabling widespread adoption across industries.
• GenAI models are being utilized to create synthetic data; this data is artificially generated based on real datasets to improve AI training, while addressing data scarcity and privacy concerns. Synthetic data can be used in place of sensitive, real-world data, ensuring improved data privacy while enhancing model accuracy.
• There is a rise in leveraging agentic workflows, where AI systems function as autonomous agents to handle complex tasks with minimal human intervention. These agents can make decisions, execute tasks and learn from outcomes, making them ideal for automating intricate processes.
• With the proliferation of IoT devices, there is growing interest in running AI models on edge devices that operate in decentralized environments. To do this efficiently, model compression and optimization techniques such as quantization, pruning and knowledge distillation are utilized to reduce the size and computational requirements of GenAI models but without sacrificing performance.
• As GenAI models grow in complexity, there is an increasing focus on optimizing AI infrastructure to support distributed training and inference across multiple nodes and graphics processing units (GPUs). This allows enterprises to scale the development of massive AI models and reduce the time taken to train and deploy them.
• The complexity of managing AI workflows, from data preparation to model deployment and monitoring, has given rise to AI orchestration platforms and the adoption of Large Language Model Operations (LLMOps) frameworks. These tools ensure that the entire life cycle of AI models — from development to production — is managed efficiently and sustainably.

Notes on quadrant positioning: The market has been segmented into Large, Midsize and Specialists to showcase the varying analytics requirements of enterprises based on their size, scale and industry dynamics. It also reflects providers’ strategy to align their portfolio, industry verticals and offerings to suit market demands and enterprise needs.

Access to the full report requires a subscription to ISG Research. Please contact us for subscription inquiries.