Most businesses started using AI the same way. Someone signed up for ChatGPT, tested it on a few tasks, and shared the results in a team meeting. Then came Copilot, Gemini, and a wave of AI-powered SaaS tools promising to automate everything from customer service to financial forecasting.
45% of UK SMEs reported using at least one AI tool by the end of 2024, up from 25% in 2022, and recent British Chambers of Commerce research has the figure climbing past 50% in 2026. The adoption curve has been steep. But a gap is opening between businesses that use AI and businesses that benefit from it. Gartner forecast last year that at least 30% of generative AI projects would be abandoned after proof of concept by the end of 2025, citing poor data quality, inadequate risk controls, and unclear business value as the main reasons.
The issue is not that AI tools are bad. Many of them are genuinely useful. The issue is that bolting AI onto existing processes and hoping it transforms the business is not the same thing as building AI into how the business actually operates.
What buying AI looks like
When a business subscribes to an AI SaaS tool, it gets access to a model trained on general data, a set of pre-built features, and a user interface designed for the broadest possible market. ChatGPT can draft emails, summarise documents, and answer questions. Copilot can generate spreadsheet formulas and help with presentations. Jasper can write marketing copy. These are useful tools that save time on routine tasks.
What they cannot do is understand your business. They do not know your product catalogue, your customer history, your pricing logic, your compliance requirements, or your operational workflows. They work with whatever text you paste into the prompt window. The output is generic by definition because the model has no access to the context that makes your business different from every other business.
For individual productivity, this is fine. A marketing manager using ChatGPT to draft a blog post outline is getting genuine value. But for business-critical operations – demand forecasting, client risk scoring, automated quality control, intelligent document routing – generic AI tools fall short because they are not connected to the data that matters.
What building AI in looks like
Building AI into a business means embedding machine learning, natural language processing, or predictive analytics into the software that the business already runs on. Not as a separate tool that someone opens in a browser tab, but as a layer within the systems that manage operations, serve clients, and process data.
The difference is specificity. A generic AI chatbot can answer questions about your industry. A chatbot trained on your knowledge base, connected to your CRM, and aware of each client’s account history can answer questions about that specific client’s order status, outstanding invoices, or service history. One is a novelty. The other replaces a support team member’s workload.
In manufacturing, this looks like predictive maintenance models trained on your own machine performance data, flagging anomalies before they become failures. In professional services, it looks like intelligent document processing that extracts data from client submissions and routes it into the right workflow based on rules specific to your practice. In retail, it is demand forecasting built on your own sales history, seasonal patterns, and supplier lead times rather than a generic model that has never seen your product range.
Each of these requires two things that off-the-shelf AI tools cannot provide: access to business-specific data, and integration with the systems that act on the AI’s output.
The data question
This is where most AI projects succeed or fail, and it has nothing to do with which AI model you choose.
AI tools are only as good as the data they work with. A large language model trained on the entire internet can write fluently about almost anything, but it cannot tell you which of your product lines is trending downward this quarter because it has never seen your sales data. A predictive model can forecast demand accurately, but only if it has clean, structured historical data to learn from.
For businesses that have spent years running on disconnected spreadsheets, siloed systems, and manual processes, the AI readiness gap is not really about AI at all. It is about data. Before any AI layer can be useful, the data needs to be consolidated, cleaned, and structured in a way that the AI can actually use. This is not the glamorous part of an AI project, but it is the part that determines whether the project delivers anything meaningful.
Businesses that invest in getting their data foundations right, often through custom-built software that connects their existing systems and creates a unified data layer, find that AI integration becomes a natural next step rather than a separate initiative. The software provides the structure. The AI provides the intelligence on top of it.
Where the real value appears
The businesses seeing measurable returns from AI are not the ones with the most AI tools. They are the ones where AI is embedded into a specific workflow solving a specific problem with access to specific data.
A logistics company that built a custom route optimisation engine using its own delivery data, traffic patterns, and vehicle capacity data reduced fuel costs by a margin no generic routing tool could match, because the model was trained on their routes, not everyone’s routes.
An accountancy practice that integrated AI-powered document extraction into its client onboarding workflow cut the time to process a new client’s financial records from two days to four hours. The AI was not a standalone tool – it was built into the practice management system, so extracted data flowed directly into the right accounts without manual re-entry.
A healthcare provider that embedded predictive scheduling into its appointment system reduced no-shows by analysing patterns in its own patient data: time of day, appointment type, patient history, weather, distance from clinic. A generic scheduling tool could not have done this because it did not have access to the data that made the predictions accurate.
These are not science fiction use cases. They are operational improvements built by AI implementation specialists working with real business data in real business systems.
The economics
Off-the-shelf AI tools are cheap to start with. ChatGPT Plus is $20 per user per month. Microsoft 365 Copilot for Business runs between $18 and $30 per user. Most AI SaaS tools fall in the $50 to $500 per month range. For a small team using them for general productivity, the ROI is positive almost immediately.
The picture changes as adoption scales. Recent research found that 60 to 68% of mid-sized firms regretted a software purchase made in the past year, and many are now actively cutting their app portfolios as the cost of running parallel subscriptions across teams becomes harder to justify. The cheap monthly bills add up to expensive annual ones.
Custom AI integration costs more upfront. A bespoke system with AI capabilities built in starts at around £25,000 and can reach six figures for complex implementations. The comparison is not like-for-like, though. The off-the-shelf tool is a productivity aid. The custom implementation is a business system that changes how the operation works.
Forrester’s 2024 Total Economic Impact study on the Microsoft Power Platform, which includes AI Builder and custom AI integrations, found a 224% three-year ROI for a composite organisation – $118m in benefits against $36m in costs. Businesses that embed AI into operational workflows rather than using it as a standalone tool consistently report higher returns, because the AI acts on business data and produces business outcomes, not just text outputs.
There is also the competitive angle. When every business in your industry has access to the same ChatGPT, none of them has an advantage. When you build AI into a system trained on your own data and optimised for your own processes, that is proprietary capability. It is an asset that competitors cannot replicate by subscribing to the same tool.
When to buy and when to build
Both approaches have their place, and the decision is not binary.
Buy when the task is general, the stakes are low, and speed matters. Drafting content, summarising meetings, generating code suggestions, answering general knowledge questions. These are perfect use cases for off-the-shelf AI tools, and there is no reason to build custom solutions for them.
Build when the task is specific to your business, the data is sensitive or proprietary, the output feeds into operational decisions, and accuracy matters more than speed. Client-facing systems, financial processes, compliance workflows, quality control, demand planning. These are areas where generic AI is not precise enough and where the business case for custom implementation is strongest.
Most businesses will end up using both. The mistake is assuming that buying a set of AI subscriptions is the same as having an AI strategy. The subscriptions handle the commodity tasks. The strategy is about where AI can create value that is unique to your business, and that almost always requires building rather than buying.
The direction this is heading
The AI market is maturing fast. The initial wave of “add AI to everything” is giving way to a more practical question: where does AI actually improve outcomes for this specific business?
Mistral CEO Arthur Mensch’s prediction at this year’s AI Impact Summit in New Delhi – that more than half of current enterprise SaaS spend could shift to AI – points in the same direction. As AI becomes capable of replacing generic software functions, the value shifts to bespoke systems that are built around the specific data, processes, and competitive advantages of individual businesses.
For SMEs that have started with off-the-shelf AI tools and seen the limitations, the logical next step is not more tools. It is infrastructure: the data layer, the integrations, and the custom software that turns AI from a productivity aid into a genuine operational advantage.
The businesses that build that infrastructure now will be the hardest to compete with in two years. The ones that stay on generic tools will find that their competitors have the same tools and the same results. The difference is in what you build, not what you buy.
