Building AI Capability
Choosing the right Approach for your Business
Artificial intelligence is no longer a futuristic concept, it is an essential tool that businesses can leverage for efficiency, insight and competitive advantage.
However, building an AI capability requires investment into infrastructure, expertise and strategy. Depending on the type of AI you want to deploy, your business will need to make different considerations.
This guide explores what it takes to establish AI capability for three classes of AI: Generative Language Models (LLMs and SLMs), Generative Image Models, and bespoke Analytical AI solutions.
Generative Language Models - Deploying AI for text Processing
Modern Natural Language Processing-based AI, such as chatbots, automated translation, and AI-driven research tools, relies on two key approaches: Large Language Models (LLMs) and Small Language Models (SLMs). Understanding the difference between these approaches is crucial for choosing the right solution.
LLMs (Large Language Models)
LLMs are powerful, general-purpose AI systems capable of generating human-like text, summarising documents, and answering complex queries. LLMs are only accessible as a service for SMEs as they require substantial infrastructure.
Infrastructure and Costs
Two or more high-performance servers for development and deployment
Each server requires multiple high-performance GPUs (costing between £5K–£10K+ per GPU)
High-speed internet via a leased line
Performance and Scalability
Highly resource-intensive: large deployments require clusters of GPUs
For enterprise-level performance, cloud-based AI services are often used instead of on-premises hardware
Fine-tuning an LLM requires vast amounts of data and significant computing power, making it costly
Manpower Requirements
Requires specialised AI engineers for deployment, fine-tuning and maintenance
Needs constant monitoring and optimisation to ensure accuracy, prevent bias, and reduce hallucinations
Best suited for businesses requiring advanced, large-scale language processing
SLMs (Small Language Models)
Infrastructure and Costs
Can run on a single high-performance server, reducing initial investment, between £5K – 10K per server.
High-speed internet, £50 for broadband, £250 for leased line, per month.
Performance and Scalability
A single GPU can support three concurrent users with real-time performance
Up to ten users can be supported with slightly reduced response times
Open-source server application software is available to keep costs manageable such as Meta’s LLaMa.
Manpower Requirements
Can be managed by a technician with basic AI training (e.g. Udemy courses)
Requires less frequent intervention, making it ideal for automated customer support or structured text processing
2. Generative Image Models – Creating Content at Scale
Generative Image Models encompasses models that create images. For the purpose of this comparison, which is focused on SMEs operating private AI services, the solution is scoped on the basis that work will be completed with a one-week turnaround.
Infrastructure and Costs
Same as for SLMs
Performance and Scalability
A single GPU can handle five concurrent offline clients
Open-source AI tools such as Stable Diffusion 1.5 and XL are available, but implementation may require technical expertise
Manpower Requirements
Typically used as a value-add service alongside professional offerings
Requires AI expertise for internal support and optimisation, but not constant monitoring
3. Bespoke Analytical AI Solutions – Tailored AI for Specific Business Needs
Analytical AI models combine domain expertise with AI algorithms, making them highly specialised and more complex to implement than utilising open source pretrained generative models.
Infrastructure and Costs
Similar hardware costs to Generative AI (£5K–£10K per server)
Bespoke software development using open-source AI tools such as Tensorflow, Pytorch and Keras, and data manipulation libraries such as numpty and opencv.
Performance and Scalability
Analytical AI often optimises efficiency and accuracy by combining machine learning with traditional analytical methods
Typically, analytical AI achieves between 10 – 100 x greater efficiency (and therefore reduced energy consumption).
Manpower Requirements
Analytical AI solutions require deep domain knowledge in addition to AI skills
Typically developed by specialists, engineers, or scientists to address highly specific use cases
Long-term sustainability depends on maintaining and refining both the AI models and the underlying subject expertise
Final Thoughts
Establishing an AI capability in your business isn’t just about buying hardware and installing software. It requires a strategic approach tailored to your business needs. While Generative AI can be implemented with relatively lower expertise and intervention, Bespoke Analytical AI solutions demand a deeper level of commitment and specialisation. SMEs should develop a detailed business case that quantifies the benefits of AI in terms of increased sales or margin.
