Investment Should Follow the Evidence

As enterprises finalize their 2027 budgets, they have far more evidence than a year ago about where technology creates value and where it creates additional requirements. AI provides the clearest example. Two years of pilots and deployments have produced useful applications, disappointing ones, new infrastructure demand, security concerns, data requirements, operating changes, and a much better understanding of what it actually takes to make the technology useful.

The difficulty is turning those lessons into 2027 commitments without pretending the learning will stop once the budget is approved.

Budgets convert strategy into allocations. Management decides which outcomes matter, makes assumptions about how to achieve them, and assigns people, money, infrastructure, and time accordingly. Those assumptions are necessary because the evidence is incomplete. The weakness appears later, when commitment to the outcome becomes confused with commitment to the original allocation.

An enterprise can remain committed to improving customer service while discovering that another investment in models produces less value than fixing the data, redesigning the process, strengthening identity, or giving experienced operators better tools. It can remain committed to using AI while moving spending from experimentation toward infrastructure, integration, cybersecurity, or domain expertise. The strategy has not necessarily changed. The enterprise has learned more about what it takes to achieve it.

That suggests a different way to construct the 2027 budget. Commit to the outcome, the amount the enterprise is prepared to invest, and the important constraints. Treat the initial distribution of that investment as the best current hypothesis about how the outcome will be achieved.

Execution then tests the hypothesis.

The budget for a strategic priority should therefore carry more than an amount and an owner. Management should know which assumptions justified the allocation, what evidence it expects to emerge as the work progresses, and which changes would justify moving the next increment of investment. The same decision should establish who has authority to make those adjustments within the agreed strategy and which changes are consequential enough to return to the executive team.

That is increasingly becoming an executive finance question as well as a technology one. Deloitte's Finance Trends 2027 research found that 54% of finance leaders now play a leading role in cross-enterprise AI and technology allocation, while optimizing capital allocation and investment decisions is among their leading priorities for fiscal 2027.

Dynamic budgeting becomes useful when the enterprise can move resources because the evidence behind the original allocation has materially changed. Its purpose is not to reopen the budget every quarter. The annual process can set a resource envelope around an important outcome while letting investment move within it as management learns what actually works.

Current AI spending already shows why that flexibility matters. IDC found that 76% of surveyed technology leaders ran AI-agent spending above forecast, while 63% have at least a quarter of that spending supported by promotional credits or introductory rates scheduled to expire by early 2027. A budget can become wrong even when the strategic objective remains unchanged.

The Decision Support System I wrote about previously provides part of the mechanism. It can retain the assumptions behind the original decision and flag when execution begins to contradict them. Dynamic budgeting gives management the ability to act on that evidence.

The quality of the evidence still matters. Moving investment every time a metric changes would replace rigid budgeting with unstable budgeting.

Technology value rarely develops on one timetable. Saving two hours of work creates capacity, but its enterprise value depends on what happens next. That time may serve more customers, improve quality, avoid another expense, or create something new. It may also disappear into the working day. The productivity gain can be real while the eventual value remains unresolved.

Some technology investments also look weak early because they require complementary changes before broader value can emerge. Processes change, skills develop, architecture evolves, and new operating practices have to become usable. The productivity J-curve provides useful historical context for that delay. Brynjolfsson, Rock, and Syverson showed that general-purpose technologies can require substantial complementary investment before conventional productivity measures reflect the assets they create.

The evidence required should change as the investment develops. Early in the cycle, progress may mean resolving an important uncertainty, a required capability now working, a dependency removed, or a process actually changing. Later, the evidence should increasingly concern operating and business outcomes. An investment that continues producing prerequisites deserves different consideration from one that repeatedly produces only another forecast of future value.

That distinction makes dynamic budgeting more disciplined. The enterprise can ask whether the evidence expected at this stage actually exists and decide whether the next investment should maintain the current path, extend it, change the mix of resources, or move elsewhere.

AI spending makes this particularly relevant going into 2027 because the label itself is becoming less useful. What began as spending on models, pilots, and experimentation increasingly turns into data, infrastructure, cybersecurity, applications, process redesign, and talent. AI created some of that expenditure. Some was expanded or accelerated because of it. Some would have happened anyway but now have a different priority.

That was the question behind the Rationalizing AI Spend Research Brief. “How much are we spending on AI?”, “What value are we creating?”, and “What should we fund next?” involve the same investments but answer different management questions. Understanding the difference helps determine where the next investment belongs.

The principle extends beyond AI. A 2027 technology budget should commit enough resources to pursue important outcomes without requiring management to defend the path imagined when the budget was created.

Strategy establishes where the enterprise intends to go. The budget establishes the resources it is prepared to commit. Execution supplies better evidence about how to get there. Decision support preserves and interprets that evidence. Dynamic budgeting gives management the authority to move investment when the evidence justifies it.

Investment should follow the evidence.

Reading Resources

• Deloitte — Finance Trends 2027: Shaping the Next Era of Stakeholder Value — September 2026
  Based on a survey of 1,434 finance leaders across 26 countries, this report helps explain how technology allocation is moving into broader enterprise management. Fifty-four percent of respondents now play a leading role in cross-enterprise AI and technology allocation, while optimizing capital allocation and investment decisions is among the leading priorities for fiscal 2027.
• Deloitte — Fund, Scale, or Stop? AI Is Exposing the Limits of Traditional Approaches to Tech Investment — October 2026
  Useful for the discussion of changing evidence and value. Deloitte argues that AI investment increasingly spans reusable platforms, data, software, consumption, people, and operating changes, while value can appear through several different business outcomes. It supports connecting ongoing investment decisions to evidence rather than relying only on the original business case.
• IDC — 5 Priorities Decide Who Leads the AI Economy — October 26, 2026
  IDC provides timely evidence for the 2027 budgeting problem. Seventy-six percent of surveyed technology leaders report AI-agent spending above forecast, and 63% have at least one-quarter of their agent budget supported by credits or introductory pricing that expires by early 2027. This illustrates how quickly budget assumptions can become inaccurate.

• Erik Brynjolfsson, Daniel Rock, and Chad Syverson — The Productivity J-Curve: How Intangibles Complement General Purpose Technologies

  Useful as the historical basis for the timing discussion. The authors show how general-purpose technologies can require complementary investment in processes, business models, products, and human capital before conventional productivity measures capture the value being created. It supports patience when prerequisites are demonstrably being built, not an indefinite extension for weak investments.

What’s your take? We are fostering a community of executives who want to discuss these issues in depth. This newsletter is but a part of it. We welcome your feedback and look forward to engaging in these conversations.

If you are interested in exploring the full report, discussing the Board’s offering further, or have any additional questions, please contact me at [email protected], and I will be happy to connect with you.

Like getting this weekly in your inbox? Subscribe.