Technology investments are meant to make organizations more efficient, lower costs or simplify work. But the business case can look much cleaner on paper than it does in practice when leaders underestimate implementation demands, ongoing maintenance, security risks or how a new tool will fit with existing systems.
Evaluating a technology investment means looking beyond its upfront price and promised gains to the costs and complexity it may create over time. Below, members of Forbes Technology Council share tech investments that can introduce unexpected costs, risks or operational headaches when they aren’t managed carefully, along with ways leaders can keep those trade-offs in check.
Enterprise Software Rollouts
Enterprise software rollouts may look efficient on paper but they backfire when people and processes aren’t managed properly. I believe people, process and technology must move together, or the business absorbs real costs: technical debt, fragmented data and employee disengagement. Technology only works when it moves in harmony with people and processes, like a well-oiled machine. – Gayathri Kolandaisami, Quantumzet Technologies
AI Tools Without Human Oversight
AI tools can look efficient on paper because the entry cost is low and the productivity gains feel immediate. But the real risk shows up when leaders let professionals get too comfortable stepping out of the loop instead of staying engaged in the judgment and review the work still requires. When that happens, we’re not only adding quality and confidentiality risk; we’re also diminishing the role of our professionals and losing our ability to grow and flex in our profession. – Mike Sewell, Gresham Smith
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
Overengineered Connectivity
One investment to be wary of is overengineered connectivity. Using more bandwidth, power and infrastructure than an application actually needs may look future-proof, but it quietly adds hardware, energy, maintenance and lifecycle costs. The smartest architecture is not the most powerful—it’s the one fit for purpose. – Alper Yegin, LoRa Alliance
Disconnected Point Solutions
A tech stack built from separate point solutions can look efficient on paper. You buy the best tech for each job and avoid a platform contract. But the cost moves into integration, identity resolution, data quality and governance. Put AI on top of that fragmented foundation, and things get even more complex. Companies often end up spending as much on their pieced-together solution as they would have spent on a platform—only now they’re also responsible for making all those disparate products work together. – Kimberly Bloomston, 6sense
Customer-Facing Agentic AI
Agentic AI can lower costs and handle interactions at scale in customer-facing environments, but unmanaged AI can quietly erase those savings. AI can fail while appearing to work, confidently giving customers the wrong answer. One hallucination may be recoverable, but thousands? That’s a brand crisis. Leaders need continuous observability and governance to prove AI is behaving as intended. – Sushil Kumar, Cyara
Subsymbolic AI Tools
Tools based on subsymbolic AI (LLMs) can generate additional costs due to the need for human verification of their outputs. Moreover, their accuracy can decline over time, as they increasingly rely on their previous conclusions. A solution is composite AI, where subsymbolic models capture statistical patterns from historical data and, combined with symbolic modeling (constraint satisfaction, planning, knowledge graphs), provide efficient reasoning based on auditable logic. – Filip Dvorak, Filuta AI
Lift-And-Shift Cloud Migration
Lift-and-shift cloud migration is the classic example of a questionable investment. It looks efficient on paper—move workloads as-is, optimize later. But “later” rarely comes; legacy schedulers carry over unchanged, adding no efficiency or visibility gains. Migrations often exceed cost and time projections by 70%, partly due to metered billing. Without orchestration managing dependencies and cost, complexity compounds with every workload wave. – Charles Crouchman, Redwood Software
Custom Software And AI Integrations
Custom software takes years to pay off, while a vendor can go live in days, especially as AI and MCP connectors speed up integration. A firm’s original investment grows exponentially when it has to fix underlying data issues. Building agents to pull in data looks efficient, but sources change, leaving teams stuck patching pipelines—a burden vendors absorb and spread across their client base. – Alex Ford, Encompass Corporation
Low-Cost Data Storage
For years, enterprises kept everything because storage cost almost nothing. Now, data is the corpus AI reads from. Stale, redundant data isn’t inert. It’s processed and priced by the token, drags accuracy down, and sits within reach of every agent you deploy. The smartest AI investment is deleting what should have been gone years ago. That starts with knowing what you have. – Asaf Kochan, Sentra
AI Coding Assistants
While AI coding allows for a faster pace of development and more efficiency from each team member, it’s also creating vast amounts of tech debt and one-off applications that have no long-term support. The volume of code created and one-off applications are likely to create a long-term maintenance challenge if not managed well. – Mark Beare, Malwarebytes
Single-Model AI Strategies
Think before standardizing all AI workloads on one large model. On paper, it is the efficient choice: one vendor, one integration, no routing logic to maintain and the best benchmark scores. However, in production, you pay top rates on every request—the large majority of which a much smaller AI model would have answered just as well—and you inherit a latency floor you cannot tune. Skipping the routing layer does not remove the complexity. It moves it to the invoice. – Srijith Ravikumar, Amazon
Outsourced IT Infrastructure
Organizations gain skills, speed, standardization and predictable costs when outsourcing IT infrastructure but give up control and visibility as risk, resilience and flexibility shift to a third party. CIOs and CTOs should go in with their eyes wide open, as they remain accountable for outcomes even without direct control of infrastructure operations. Outsource what makes sense and control the rest. – Susan Odle, StorMagic
Automation Of Inefficient Processes
Automating a process nobody fixes first is not efficient. On paper, you cut the hours in half, but what you really did was speed up a bad process and make it harder to change, because now it is buried in a tool instead of in someone’s head. Spend a week asking why the steps exist. Half of them usually should not. Automate what is left. – Kriti Faujdar, Microsoft
Low-Code Automation
Low-code automation can feel like handing each team a box of power tools: Problems get fixed faster and productivity rises overnight. But without blueprints, safety rules or ownership, those solutions become a maze of duplicate apps, weak controls and fragile workflows. What begins as speed can quietly harden into technical debt, compliance risk and a support bill no one planned for. – Bindu Madhavi Mangalampalli, Cotiviti
Workflow Automation Without Identity Controls
Workflow automation can look efficient on paper, but every agent, bot, API and service account becomes an identity with authority. If leaders don’t assign ownership, time limits and review points, automation creates hidden authority pathways, duplicated logic and brittle dependencies. The cost (and risk) shows up later in audits, outages and incident response. – Craig Davies, Gathid
AI And Cloud Sprawl
One investment that often looks efficient on paper is AI and cloud sprawl. Teams can rapidly deploy models, tools and cloud workloads, creating the illusion of agility. But without strong governance, FinOps, observability and lifecycle management, costs compound, duplicate solutions emerge and operational risk grows. The goal isn’t just innovation; it’s disciplined scale with accountability. – Gaurav Rastogi, Hertz Corporation
Industrial AI Without Clear Outcomes
Industrial AI can look like an easy efficiency win, but deploying it without clear operational outcomes can add cost, risk and complexity. Leaders should start with a measurable problem, such as reducing downtime or improving first-pass yield, and ensure AI strengthens decisions and workflows rather than adding another layer of noise. – John Davagian, L2L
Multiyear Cloud Commitments
I am wary of large, multiyear cloud commitments. The discount looks efficient, but a bad usage forecast can lock cash into idle capacity and discourage better architecture choices. I would commit in stages, review utilization monthly and keep part of the workload flexible. – Kayode Faturoti, Breet
Frontier AI Model Investments
The newest frontier model can look efficient but age badly. In creative work, the best model for a workload changes repeatedly, sometimes overnight. We built around that reality: model-agnostic routing, traces and evals. If your product is welded to one model, you may pay once for access and again to rebuild when its quality or economics change. – Vatsal Bhardwaj, Jabali.ai
Microservices Architecture
Microservices can look like a scalability investment but quietly turn every function call into a network dependency, deployment pipeline and observability problem. Use them where teams truly need independent scale or release cycles; otherwise, a modular monolith can deliver the same business speed with far less operational tax. – Akhilesh Sharma, A3Logics Inc.