Key Takeaways
- Buy an existing tool when the problem is common and the tool already solves most of it well.
- Choose integration when the capability exists but needs to connect to your CRM, inbox, documents or internal process.
- Build custom AI when the workflow is valuable, genuinely specific and important enough to justify ownership, testing and ongoing support.
- Most small businesses do not need a model built from scratch. A tailored application using established models is still a custom solution.
- Start with a pilot and real examples. Do not pay for a full build until you understand what the ready-made options cannot do.
Most UK small businesses should try an off-the-shelf AI tool first. Move towards custom AI only when the standard product cannot fit a valuable workflow, cannot connect safely to the right data or creates limits that cost more than a bespoke solution would.
That does not mean every business should settle for generic software. It means you should earn the complexity of a custom build. A good custom AI solution solves a defined business problem that existing products cannot solve properly. It should not be an expensive way to recreate features you could already subscribe to.
There is also a useful middle ground: integrate established AI models and software into a workflow designed around your business. For many small firms, that hybrid approach delivers most of the value without the cost and risk of building everything from zero.
First, Be Clear About What “Custom AI” Means
The phrase can create the wrong picture. A custom AI system does not necessarily mean training a giant language model from the ground up. That would be unnecessary and unaffordable for most small businesses.
Custom work might include:
- A secure internal assistant that searches approved business documents.
- A quotation workflow that reads enquiries, checks rules and prepares a draft for review.
- A customer portal that uses AI to explain account information in plain language.
- A document-processing application with a tailored review screen and audit trail.
- A connected sales workflow that combines forms, email, CRM records and human approval.
The underlying model may come from an established provider. The custom value sits in the application, data connections, rules, permissions, user experience, monitoring and fit with the business.
The Three Practical Options
1. Buy an Off-the-Shelf Tool
This is software designed to solve a common problem for many customers. Examples include meeting transcription, writing assistance, customer-support platforms, CRM features and general AI workspaces.
Choose this route when the process is fairly standard, the product covers most requirements and your team can adapt without losing something important. You gain faster setup, an existing support team, regular updates and a predictable subscription.
The trade-off is fit. You work within the supplier's interface, rules, roadmap and pricing. The product may include many features you do not need while missing one detail that matters to your operation.
2. Integrate Existing Tools
Integration is often the best small-business answer. The AI capability already exists, but you connect it to the systems and approval points that make the workflow useful.
For example, an existing model can summarise an enquiry. An AI integration can then place that summary in the correct CRM record, mark missing information, create a follow-up task and send anything uncertain to a person.
You avoid rebuilding the underlying AI while making the process feel specific to your business. The risks are dependency on several suppliers and the need to maintain the connections when their systems change.
3. Build a Custom Solution
Custom development makes sense when the workflow is central to how you compete or deliver, the requirements are unusual, and the value justifies a proper product lifecycle.
You gain control over the user journey, data flow, business rules and roadmap. You can design the system around the people doing the job rather than forcing the job into a generic interface.
You also take on more responsibility. The system needs discovery, testing, security, documentation, hosting, monitoring, user support and continued improvement. Custom does not mean finished forever.
Six Signs You Should Start With Off-the-Shelf
- The task is common. Many businesses already need meeting notes, basic writing help or standard support-ticket summaries.
- You have not tested the use case. A subscription is a cheaper experiment than a build.
- The process changes every week. Stabilise it before encoding it in software.
- The data is not ready. A custom interface will not repair missing or inconsistent records.
- The value is modest. Do not build a product to save a few minutes a month.
- Nobody will own it. A bespoke system without an internal owner becomes an orphaned system.
Our honest review of AI tools that save time is a useful starting point before commissioning anything.
Six Signs Custom AI May Be Worth It
- The workflow is genuinely specific. Your process, documents or decision path are not served well by a standard tool.
- The workflow has material value. It affects revenue, service quality, capacity or risk often enough to justify investment.
- Several systems must work together. The user needs one reliable flow rather than copying information between products.
- Control matters. You need particular permissions, hosting, audit records, approval points or retention rules.
- The experience is customer-facing. A generic interface would weaken your service or force customers through unnecessary steps.
- The limitation has been proved. You can show exactly where existing software fails, rather than assuming bespoke must be better.
Compare the Full Cost, Not Just the Subscription
Off-the-shelf software can look cheap because the first price is a monthly seat. Calculate the cost across the period you expect to use it, including extra seats, usage, premium integrations, onboarding, migration and the staff time spent working around its limitations.
Custom software has a larger upfront cost, but that does not automatically make it expensive over time. It can remove several subscriptions or support a process that a standard product cannot. It still needs hosting, model usage, maintenance and improvements.
As of 18 July 2026, our current custom AI service uses these starting points:
- Discovery: £499 plus VAT. A problem workshop, feasibility work and a prototype or proof of concept.
- Custom Build: £2,999+ plus VAT. A focused production system with integration, documentation and initial support.
- Enterprise: £9,999+ plus VAT. A more complex ecosystem with stronger security, multiple components and ongoing operational needs.
These are MattDarm prices, not a market average. A quote should explain what is included, which third-party costs remain and what happens after the initial support period.
A Simple Build-or-Buy Scorecard
Ask these questions and write down the evidence.
Does an Existing Tool Solve 80% of the Job?
If yes, test it. The missing 20% may be manageable through process changes or integration. If the missing part is the part that creates your value, custom work deserves a closer look.
Is the Problem Stable and Measurable?
You should be able to describe the starting event, inputs, desired output, exceptions and success measure. If the brief is “we want our own ChatGPT”, discovery is not complete.
Is the Data Available and Lawful to Use?
List the data sources, owners, permissions, retention needs and sensitivity. The ICO's AI and data-protection guidance applies whether you buy or build. A supplier's security page does not replace your own assessment of how the tool will be used.
What Happens When the AI Is Wrong?
Every AI system will produce uncertain or incorrect output at some point. Decide how that is detected, what a user can correct and when a person must approve the result. If an error could affect someone's money, rights, safety or access to a service, the control needs to match the risk.
Who Owns and Maintains the Result?
For a subscription, ask about export, termination and data deletion. For a custom build, make the contract clear about code, data, configuration, prompts, documentation and third-party licences. “Bespoke” does not automatically mean you own the underlying commercial model.
Security and Supplier Due Diligence
The UK government's AI Cyber Security Code of Practice recommends documenting the business requirement, assessing threats, limiting permissions, testing, monitoring and planning for the end of the system's life.
Those principles apply to both choices. With off-the-shelf software, review the supplier, contract, data use, admin controls and exit route. With a custom system, review architecture, secrets, dependencies, logging, patching, backups and who responds to incidents.
Our guide to protecting business data when using AI gives a practical checklist for staff and managers.
Three Example Decisions
Example 1: Meeting Notes for a Consultancy
The need is common: record an approved meeting, create a summary and list actions. Start with an established product. A custom build is difficult to justify unless the firm has an unusual regulated workflow or must store recordings in a specific environment.
Example 2: Enquiry-to-Quote Workflow for a Specialist Installer
The business receives emails, drawings and site details, then prepares a quote using its own rules. A hybrid may be right. Use established models for extraction and drafting, but build the integrations, review screen and quote logic around the real process.
Example 3: Document Review for a Regulated Firm
The system must search approved documents, respect client boundaries, show sources, record activity and escalate uncertainty. A custom or carefully configured private solution may be justified because permissions, evidence and auditability are part of the product, not optional extras.
Pilot Before You Commit
Use a small set of real examples and define the test before building. Measure accuracy, time saved, number of exceptions, review effort and whether the team actually uses the result.
A pilot should answer one decision: can this approach improve the workflow safely enough to justify the next stage? It does not need polished branding or every feature. If the result is weak, stop or change direction. That is a successful discovery outcome because it prevented a larger waste.
An AI strategy session can help frame that test. Our AI-readiness checklist can also show whether process and data work should come first.
Frequently Asked Questions
Should a small business build custom AI?
Only when the problem is valuable, specific and poorly served by existing tools. Most small businesses should test an off-the-shelf or connected solution first, then move to custom development when the limitations are clear and measurable.
How much does a custom AI solution cost?
MattDarm currently starts with a £499 discovery, custom builds from £2,999 and complex enterprise work from £9,999, excluding VAT. The final cost depends on data, integrations, security, testing, hosting and ongoing support.
What is the difference between custom AI and AI integration?
Integration connects existing AI capabilities to your systems and workflow. Custom AI involves building more of the application, logic, interface or model behaviour around your unique requirements. Many strong small-business solutions are a hybrid of the two.
Do we own a custom AI system?
Ownership depends on the contract, third-party models and licensed components. The agreement should state who owns the code, configuration, data, prompts, documentation and deployment. Do not assume bespoke work automatically means complete ownership of every dependency.
How should we test an AI idea before a full build?
Use real examples in a limited pilot, define success and failure before testing, keep a human in the loop and measure quality, time saved, exceptions and user adoption. A working prototype should answer a business question rather than merely look impressive.
The Bottom Line
Buying is usually the right first move for a common problem. Integration is often the best next move when useful tools need to work together. Custom AI earns its place when the workflow is valuable, specific and important enough to own properly.
Do not start by choosing a technology. Start with the job, the evidence and the cost of the current limitation. Then choose the simplest approach that can deliver the result safely.
If you want a neutral build-or-buy assessment, get in touch. We will tell you when an existing product is enough and scope custom work only when it solves a problem worth owning.
