Adding AI features to software through API integration costs 8,000 to 25,000 for development, plus 100 to 1,000 per month in ongoing API usage. This covers practical AI features like document processing, email classification, customer chat, and automated reporting. For custom AI model training (rarely necessary for UK SMEs), costs start at 40,000 and can exceed 100,000. The right approach for most businesses is API integration: it is cheaper, faster, and sufficient for the vast majority of use cases.
We help UK SMEs add AI features to their software with clear, upfront pricing. This guide breaks down the costs so you can budget accurately.
What Are the Two Approaches and Their Costs?
There are two fundamentally different approaches to adding AI, with very different cost profiles:
Approach 1: API Integration (Recommended for Most SMEs)
You connect your software to an external AI service (OpenAI, Anthropic, Google AI) through their API. Your system sends data, receives AI-generated responses, and uses them. No model training, no AI infrastructure, no specialised ML expertise needed.
- Development cost: 8,000 to 25,000 per feature
- Ongoing API cost: 100 to 1,000 per month
- Timeline: 2 to 6 weeks
- Best for: Document processing, classification, chat, summarisation, search
Approach 2: Custom Model Training
You train or fine-tune a custom AI model on your specific data. This requires ML expertise, training infrastructure, and significant data preparation. The model runs on your infrastructure or a managed platform.
- Development cost: 40,000 to 100,000+
- Ongoing cost: 500 to 5,000 per month (compute and maintenance)
- Timeline: 8 to 16 weeks
- Best for: Domain-specific tasks where general models underperform
For 90 percent of UK SME use cases, API integration is the right choice. It delivers the same outcomes at a fraction of the cost and time. See our guide on adding AI to existing software for the technical details.
What Is the Cost Breakdown for API Integration?
The development cost for a single AI feature breaks down as follows:
- Use case definition and data assessment: 1,000 to 2,000. Understanding the task, assessing data availability, defining success metrics.
- API integration development: 4,000 to 12,000. Writing the code to connect your system to the AI API, process requests and responses, and handle errors.
- User interface or workflow integration: 2,000 to 6,000. Adding the AI feature to your existing interface or building an automated workflow.
- Prompt engineering and testing: 1,000 to 3,000. Tuning the AI prompts or parameters for accuracy in your specific context. This is iterative and requires real data.
- Monitoring setup: 500 to 2,000. Setting up cost tracking, accuracy monitoring, and alerts for failures.
Total for a single feature: 8,000 to 25,000. Multiple features share some of the infrastructure cost, so two features do not cost twice as much as one.
What Are the Ongoing API Usage Costs?
AI API providers charge per token (roughly per word processed). The cost depends on the model and usage volume:
- Low volume (hundreds of calls per month, short documents): 50 to 200 per month
- Medium volume (thousands of calls per month, moderate document length): 200 to 800 per month
- High volume (tens of thousands of calls per month, longer documents): 800 to 3,000 per month
Example: Processing 200 invoices per week through an AI API. Each invoice is about 1,000 tokens. At typical pricing of 0.005 per 1,000 tokens, that is about 1 per week, or 4 per month. Very affordable compared to the 10 hours of manual processing it replaces.
The key is monitoring usage. Without monitoring, API costs can spiral if a feature is called more than expected. Set up cost alerts from day one.
How Does the Cost Compare to the Value?
The ROI calculation for AI features is straightforward:
- Current cost: Staff time spent on the task. At the ONS average of 35,000 per year for a UK knowledge worker, 10 hours per week is worth about 7,500 per year.
- AI cost: Development cost amortised over the expected life of the feature (typically 3 to 5 years), plus ongoing API and maintenance costs.
- Net saving: Current cost minus AI cost.
Example: Automated invoice processing. Current cost: 10 hours per week = 7,500 per year. AI development: 12,000. API cost: 100 per month = 1,200 per year. Maintenance: 1,500 per year. Year one total AI cost: 14,700. Year two and beyond: 2,700 per year. Break-even at approximately year two. From year three, net saving of about 4,800 per year.
For higher-volume tasks or more expensive staff, the ROI is faster. See our guide on AI use cases for UK SMEs for which tasks deliver the best ROI.
When Is Custom Model Training Worth It?
Custom model training is worth the investment only when:
- Domain-specific accuracy is critical: General AI models perform well on most tasks, but may underperform on highly specialised content (medical records, legal documents, technical specifications in a niche field).
- You have enough training data: Training a custom model requires hundreds or thousands of labelled examples. Without sufficient data, the model will not perform well.
- API costs are prohibitive: If your volume is extremely high (millions of calls per month), the API cost may exceed the cost of running your own model.
- Data cannot leave your infrastructure: For highly sensitive data in regulated industries, self-hosted models keep data entirely within your control.
For most UK SMEs, none of these conditions apply. API integration is cheaper, faster, and performs well. Custom training is an option for the future if your needs evolve.
What Hidden Costs Should You Watch For?
Beyond the development and API costs, factor in:
- Data preparation: AI features need clean, structured data. If your data is in disorganised formats, preparation work is needed before AI can process it. Budget 2,000 to 8,000 for data preparation.
- Accuracy tuning: AI is not perfect out of the box. Prompt engineering, parameter tuning, and testing with real data take time. This is included in the development cost but may need additional rounds.
- Integration maintenance: AI APIs update their models, change pricing, or deprecate features. Budget for quarterly review and occasional updates. Typically 1,000 to 2,000 per year.
- Training and change management: Staff need to learn to use the new AI features. Budget for training time and documentation.
Ready to add AI features to your software? Book a free discovery call to discuss your specific use case and get a fixed-scope quote, or see our services for what we offer.