Build It, Buy It, or Both? The AI Software Decision in 2026

In 2020, a modernization conversation often started with a shopping list: a CRM, a cloud migration, a better website, perhaps a new app. In 2026, the conversation has become more interesting: which parts of the business need to keep evolving, and which technologies should we own to make that happen?

There is usually one more question, asked quietly after the demo: “Do we really need to build this? It appears to be available for $30 a month.” Fair question. The answer takes a little more work than the demo suggests.

From occasional AI projects to everyday business tools

The shift is visible in the numbers. In McKinsey’s 2020 survey, 50% of respondents said their organizations had adopted AI in at least one business function. In 2025, that figure reached 88%. Yet only about one-third of respondents in the 2025 survey said their organizations had begun scaling AI across the enterprise. AI has become common; turning it into a company-wide business advantage is still a work in progress. McKinsey, 2020; McKinsey, 2025

The market for ready-made software was expanding well before generative AI arrived. In July 2020, Gartner forecast worldwide SaaS spending of approximately $104.7 billion. Its November 2024 forecast put spending for 2025 at $299.1 billion. These forecasts were produced at different times using evolving methodologies, so they illustrate the scale of change rather than provide a precise comparison of actual sales. Gartner, 2020; Gartner, forecast for 2025

Generative AI then added a new wave of products. Menlo Ventures estimates that enterprise generative AI spending grew from $1.7 billion in 2023 to $37 billion in 2025, including $19 billion on applications. This measures a particular market segment—not all software spending or custom development services. In Menlo’s research, the share of enterprise AI solutions purchased rather than built internally rose from 53% in 2024 to 76% in 2025. The shelf of ready-made options has become considerably better stocked. Menlo Ventures

The sales conversation has changed, too

Around 2020, buying enterprise software typically involved demonstrations, license negotiations and an implementation project. Custom development started with requirements, estimated hours and a team.

Today, an employee may discover an AI tool, try it, introduce it to colleagues and create demand long before procurement joins the conversation. Menlo Ventures found that 27% of AI application spending came through product-led growth, compared with 7% in traditional software. This is evidence from a specific study of US enterprise buyers, rather than a rule for every market. Still, it suggests that “let me show you what I already use” has become a meaningful sales channel. Menlo Ventures

Services remain a substantial business. Gartner forecast worldwide IT services spending of $1.69 trillion for 2025, although that broad category includes much more than custom development and cannot be compared directly with SaaS. Capgemini reported that generative and agentic AI represented more than 7% of its bookings in the second quarter of 2025. In the first half of 2026, it linked growth in its Applications & Technology business partly to accelerating legacy modernization projects. Gartner; Capgemini, 2025; Capgemini, 2026

For development providers, the question is increasingly: “What can you deliver that we cannot get by subscribing?” A good answer explains the client’s data, business rules, integrations, reliability requirements and expected results. The work often involves connecting purchased products to the way a particular business actually operates.

As Microsoft CEO Satya Nadella wrote in January 2025: “2025 will be about model-forward applications that reshape all application categories.”

That is a bigger change than adding a sparkling AI button to an existing interface. Gartner subsequently forecast that up to $234 billion in enterprise application spending could be exposed to disruption from agentic AI by 2030. This is a forecast about potential changes to software economics, not evidence that SaaS spending has already disappeared. Microsoft; Gartner

What successful companies actually do

Walmart builds where its own knowledge matters. It developed Wallaby, a family of retail-specific language models trained using its data, and combines these with other models for customer-facing experiences. This illustrates strategic control over technology supporting the shopping experience. The announcement does not establish a specific revenue increase attributable to Wallaby. Walmart

Amazon built DeepFleet for its robotic logistics. According to Amazon, the model coordinates more than one million robots and improves their travel efficiency by 10%. Developing a system around such distinctive infrastructure makes sense. Most businesses, fortunately, do not need to coordinate a million robots before lunch. Their opportunity will be different, and so will their economics. Amazon

Capita bought Microsoft 365 Copilot and developed its use across the business. In a Microsoft-published customer story, Capita reported saving 9,000 employee hours in one month. Employees also created more than 169 agents for their own tasks. A purchased platform became the foundation for company-specific workflows. These are customer and vendor-reported results, rather than an independent comparison with building an alternative from scratch. Microsoft / Capita

1-800Accountant used Salesforce Agentforce for customer service. Salesforce reports that it autonomously resolved 70% of the company’s chat engagements during tax week in 2025. The result applies to that channel and period; it does not describe every support interaction. It does show how a ready-made platform can help with repeatable questions during a busy season. Salesforce

So, who wins? These cases support different decisions for different jobs. Custom software offers control over capabilities that distinguish a business. Purchased software can deliver value sooner when the problem is widely shared. Neither route removes the need to change how people work: McKinsey found that organizations achieving the strongest AI impact were nearly three times as likely as others to fundamentally redesign workflows. McKinsey

Where buying makes sense, and where building earns its place

Ready-made products are sensible starting points for email and documents, CRM, routine customer support, marketing automation, analytics, coding assistance and standard internal knowledge search. Buying is especially attractive when time matters, security and access controls fit your requirements, and adapting the process will not weaken your competitive advantage.

Custom development is more compelling for a distinctive customer product, a specialized portal, unusual supply-chain processes, complex pricing, industry-specific document handling, forecasting on proprietary data, legacy integration or an AI agent that must safely act across several systems. In these cases, the combination of data, rules and customer experience creates the value.

Often, the practical answer is to buy the foundation and build the missing pieces. Keep the ERP and CRM, use an existing AI model, and develop the workflow that connects them, with clear rules for handing exceptions to people.

An Intetics case involving construction invoice matching illustrates this approach. Following deployment, daily manual effort for the relevant invoice-processing work fell from more than four hours to typically two or three. The system currently matches 21% of all processed invoices. That is a measurable benefit with room to expand coverage: a useful reminder that progress can be valuable before automation is complete.

What might it cost, and how long might it take?

The following are illustrative planning ranges, not published market averages or quotations. They assume existing cloud models and a blended team rate of approximately $75–150 per hour, covering analysis, development, testing and management. Difficult integrations, regulatory requirements and poor-quality data can change the budget substantially.

These budgets cover an initial scope. Operations, cloud services, model usage, security and further development require separate allowances. A pilot and a production system serving the whole business are different financial commitments, even if their screenshots look remarkably similar.

For purchased products, the entry price is easier to see. Microsoft publishes a price of $30 per user per month for Microsoft 365 Copilot, paid annually, with a separate qualifying Microsoft 365 license required. For 100 employees, that means $36,000 a year for Copilot alone.

HubSpot Sales Hub Professional starts at $90 per seat per month with an annual commitment: approximately $54,000 a year for 50 seats, plus the published mandatory Professional onboarding fee of $1,500. Complex migration and integration are additional. Microsoft; HubSpot

A simple product pilot may take days or weeks. Data migration, permissions, training and ERP connections can take weeks or months. For early planning, I would allow roughly $10,000–50,000 for a straightforward integration and $50,000–250,000 or more for a complex one. These are illustrative estimates, not vendor prices or industry averages.

The useful comparison is therefore the total cost over two or three years for the same expected outcome: subscriptions and consumption, implementation, support, data work, risk and the ability to change suppliers.

“$30 a month” is a price tag. Making it work for your business is a project.

A practical way to decide

Start with the result you want: faster order processing, lower support costs, fewer documents requiring manual checks or quicker delivery of new features. Test a ready-made product on real data within a limited workflow.

If it does the job with reasonable configuration, buy it and invest in adoption. If essential business logic falls outside its capabilities, build the missing layer. If that logic creates your competitive advantage and needs to evolve continuously, invest in your own product and the team that can maintain it.

Modernization in 2026 looks less like a project with a finish line and more like a business capability you keep exercising. New models and products will continue to arrive. The companies best placed to benefit will know how to test value quickly, retain control of their data and adapt their architecture as the business grows.

That is a useful kind of confidence: being ready to make the next decision, even before the next impressive demo arrives.