Two years ago, almost every business conversation I had about AI started the same way. “Which tools should we be using?” People wanted a shortlist. ChatGPT for this, Copilot for that, a voice bot for the phones. Give me the stack and I’ll get started.

That phase is mostly over now. The businesses that moved early have a stack. They have some tools in place. And now a lot of them are coming to me with a different question.

“We’ve got AI tools, but they don’t quite fit. Should we build something custom?”

This is the right question to be asking. And the answer is genuinely not obvious. I’ve helped businesses go both ways — leaning into off-the-shelf platforms and building from scratch — and I’ve seen both approaches succeed and fail. Here is how I think about it.

The Case for Buying

Off-the-shelf AI tools have got genuinely good. When I say “off-the-shelf” I mean the category of purpose-built AI platforms — tools like HubSpot’s AI features, Salesforce’s Agentforce, Microsoft Copilot baked into 365, or any of the dozen vertical SaaS products with AI built in.

Two years ago these were demonstrations of capability. Today, many of them are actually useful. If your business runs on standard processes and you are in a sector that AI vendors have targeted, you can probably find something that covers 80% of what you need.

The advantages of buying are real:

You deploy faster. A platform built for your use case — say, AI-assisted proposal generation for professional services — has already solved the core problem. You configure, you train, you go live. Weeks, not months.

You stay updated automatically. When the vendor improves the model or adds features, you get that improvement without doing anything. Off-the-shelf tools compound.

The support infrastructure exists. There are help docs, user communities, dedicated customer success teams. You are not alone.

And most importantly: you know what you are getting. The product has been used by thousands of other businesses. The edge cases have been found. The obvious failure modes have been addressed.

Why Buying Eventually Runs Out

Here is what I have consistently observed with businesses that go all-in on off-the-shelf AI tools and nothing else.

They hit a ceiling.

The ceiling is not always the same. Sometimes it is a workflow that the vendor never anticipated. Sometimes it is proprietary data that the tool cannot access or handle securely. Sometimes it is a combination of tasks that the tool can do individually but cannot connect into a single smooth operation.

The most common version of this ceiling I see is the integration problem. A business has a CRM, an accounting system, a job scheduling tool, a customer portal. Off-the-shelf AI works beautifully within one of those systems. But the business process that would actually save time — or create real value — requires moving data across all of them in a specific way that no vendor has built for.

The second version is the differentiation problem. If you and your three main competitors all use the same AI CRM with the same AI features, AI becomes table stakes, not competitive advantage. You have automated, which is good. But you have not differentiated.

The third version is the data lock-in problem. You have built workflows and automations inside a vendor’s platform. When the vendor raises prices, changes their model, or gets acquired, you have a problem. Everything you built lives in their ecosystem, not yours.

What Custom AI Actually Is

When people hear “custom AI,” they often imagine something exotic. A team of machine learning engineers training a proprietary model on your internal data.

That is one version of custom AI. It is also the expensive version, and for most businesses, it is the wrong version.

What we actually build at Omni by Enterprise DNA is different. It is custom applications built on top of existing foundation models — Claude, GPT-4, Gemini — combined with your business’s specific data, processes, and integrations. The model itself is not custom. The layer around it is.

This approach captures most of the value of custom AI at a fraction of the cost. You get:

Workflows designed for your exact process. Not a general-purpose AI assistant that handles a range of tasks. A specific agent that handles your job-costing process, or your client intake flow, or your internal knowledge queries, exactly as your business runs it.

Access to your proprietary data. Your historical contracts, your pricing logic, your client notes, your operational procedures. Off-the-shelf AI cannot touch this data because it was not trained on it. Custom-built applications can use retrieval-augmented generation to make your internal knowledge available to the AI in real time.

Integrations that do not exist off the shelf. Your industry-specific systems — the ERP you have used for fifteen years, the scheduling tool that only 2,000 businesses in your sector use — probably do not have native AI integrations. Custom builds can bridge these gaps.

Brand and experience control. Your customers experience something that feels like your business, not like a vendor’s product with your logo on it.

How I Actually Make the Decision

When a business comes to me with this question, here is the process I walk through.

Step one: what are you actually trying to solve?

This matters more than any other factor. Businesses often come in with a solution in mind (“we want to build an AI chatbot”) when the real problem is something different (“our team spends four hours a day answering the same twenty client questions”). Sometimes the solution to that is a chatbot. Sometimes it is a better FAQ. Sometimes it is automating how those questions are routed before a human touches them. Understand the problem before you decide on the approach.

Step two: does an off-the-shelf solution exist that handles this?

If yes, use it. Seriously. Do not build what you can buy. The engineering effort, ongoing maintenance, and update burden of custom AI is significant. If a $500-per-month SaaS tool solves your problem, that is the right answer.

Step three: what is the data situation?

If the value you are trying to unlock lives in your proprietary data — your specific client relationships, your institutional knowledge, your operational history — you probably need a custom build. That data is not available to any off-the-shelf product. It is your competitive advantage. Build the system that uses it.

Step four: what is the integration requirement?

If the workflow you need to automate requires connecting more than two or three systems in non-standard ways, custom is usually the more practical path. Trying to build complex cross-system integrations on top of off-the-shelf tools is often harder and more expensive than just building the integration directly.

Step five: how important is differentiation?

For commodity processes — invoice generation, meeting summaries, basic customer support — buy. The differentiation from doing those things better than competitors is low.

For core business processes — how you scope projects, how you price jobs, how you manage client relationships — consider building. These are the processes where your methodology, judgment, and institutional knowledge live. That is exactly what you want AI to amplify, not replace with a vendor’s generic logic.

The Hybrid Path

Most businesses end up in the middle. They use off-the-shelf AI for standard tasks and build custom solutions for the processes that are uniquely theirs.

This is smart. There is no reason to build a custom AI assistant for writing marketing emails when excellent off-the-shelf tools exist. But there is every reason to build a custom agent that handles your specific end-to-end client intake process, integrates with your project management system, and documents your proprietary methodology throughout.

The division is roughly: buy where the process is standard, build where your competitive advantage lives.

The Skills Question

One thing I have not mentioned yet: regardless of which path you take, your team needs baseline AI literacy to work effectively with AI tools — custom or off the shelf.

This is something I see skipped constantly, and it consistently causes problems. Businesses deploy an AI system, and the team uses it at 20% of its potential because nobody understands how to write effective prompts, interpret outputs critically, or spot when the AI is producing plausible-sounding nonsense.

Whether you buy or build, invest in the data and AI skills of your people. The tool is only as useful as the human behind it.

Where to Start

If you are sitting on this question right now, here is my honest recommendation.

Start with the clearest problem you have. The task that consumes the most time, or the workflow that frustrates people the most, or the customer experience point that consistently fails. Get specific about what you would actually want an AI system to do there.

Then check whether an off-the-shelf solution addresses it well. If it does, stop there. If it does not — if the fit is awkward, the integrations are missing, or the data situation makes it impossible — that is your signal to consider a custom build.

If you want to work through this properly, we do this as part of our Omni Advisory engagements. The first conversation is mapping your current processes against what AI can actually do for your specific business. From there, we make the build-vs-buy call together, with a clear picture of the cost, effort, and value on both sides.

The question is genuinely worth answering carefully. The difference between buying the wrong tool and building the right one is often the difference between AI being a cost centre and AI being a competitive advantage.

Get that decision right, and the rest follows.