When Building AI In-House Makes Business Sense

AI has made custom software cheaper to produce, and enterprises are acting on it. In McKinsey’s 2026 State of AI survey, 32% of respondents said their organizations had decided against buying at least one software product or feature because they could build it with agentic coding tools. The share reporting financial impact from AI has not moved: 37% attributed any EBIT impact to AI, essentially the same as in 2025.

Lower development costs change what a build project costs to start. They do not settle whether the result is worth owning. Building pays off when a capability depends on process knowledge competitors cannot buy and when the company is prepared to operate what it creates.

The Companies Building Most Are Also Redesigning Workflows

McKinsey’s survey shows building is most common among the companies getting the most from AI. Nearly half of its AI high performers have skipped a software purchase to build in-house, compared with 31% of other organizations. The same group is far more likely to redesign workflows around AI: nearly three-quarters do so, compared with about a quarter of others.

These figures show a pattern, and high performers differ from their peers in other ways too. The pattern still points to where building pays off: in companies that change how work gets done. A custom tool that automates an unchanged process adds software to maintain and little the business could not buy.

Building also adds running costs. About one in five respondents say AI operating costs, including tokens, already limit how much their organizations use AI. A vendor spreads integration, updates, and maintenance across many customers, while an internal team carries them alone. Standard capabilities, such as document summarization or invoice processing, rarely justify that burden.

Building Pays Off Where the Process Knowledge Is Proprietary

The strongest build cases rest on knowledge a vendor cannot package and sell. McKinsey’s interviews with leaders at 15 AI-centric companies produced a simple rule: build only what makes the company distinctive. For the operations stack, McKinsey recommends starting with off-the-shelf tools that can be integrated and swapped out later.

In an SAP landscape, that knowledge often sits in pricing logic, exception handling, approval paths, plant-specific scheduling rules, and negotiated customer terms. These rules shape how a company runs its processes, and competitors cannot license them.

The build boundary is often narrow. A company can buy the model and the development platform, then build the agent logic that encodes its own rules. SAP is positioning Joule Studio for this layer, and says agents built there are grounded in the live business data, processes, and business semantics already in a customer’s landscape.

A Build Decision Is Also an Operating Decision

A built agent needs an owner after go-live. Someone has to monitor its output, retest it when SAP releases change the underlying processes, and manage model updates and usage costs.

SAP is taking on part of that work. Its managed Joule Studio runtime handles runtime configuration, cluster management, storage, and model access, and SAP is offering free design-time access through the end of 2026, with AI-assisted development capabilities under fair-use limits. Those terms lower the entry cost and remove some infrastructure work. The customer still owns how the agent behaves inside its processes.

Roadmap timing adds a second check. At SAP Connect, SAP said most of its new Joule capabilities arrive in Q4 2026 or Q1 2027, with several finance assistants limited to early adopters at first. An agent built today may duplicate a capability SAP delivers within months, leaving the company to maintain software it no longer needs.

What This Means for SAPinsiders

Build only the layer holding proprietary context. Buy the models and platform capabilities that competitors can also access. Internal effort then goes where it creates an advantage.

Check SAP’s roadmap before approving a build. Compare each proposed agent with SAP’s announced delivery dates. That comparison avoids paying to build and maintain a capability SAP will ship.

Name the operating owner before the build begins. Assign monitoring, retesting, and model updates up front. The agent then stays reliable after the project team moves on.