When Hurricane Sandy put New York data centers underwater, David Tyler's disaster recovery plan did its job. Roughly 1,200 database-licensed processor cores were flooded, yet business continuity largely held for the multinational financial services company where he worked at the time.
The expensive surprise came later, at the negotiating table. During talks on an upcoming enterprise licensing agreement, the database software vendor said only active, countable cores could be included.
Excluding the drowned infrastructure created potential exposure "running into tens of millions" once capacity came back online, said Tyler, who now leads Outlier Technology, a consultancy focused on improving enterprise technology systems. His background includes developing complex data and trading systems for financial firms.
"Operational resilience did not protect us from commercial exposure created by the same incident," he added. The better protection may have been "contractual clarity over disaster-related unavailability rather than a second database stack."
Tyler's experience raises a question CIOs face at every renewal and architecture review: How much should you pay today to preserve a technology choice you may never need?
Second clouds, standby vendors and parallel platforms are sold as insurance. But the premiums don't stop at the license fee.
Start with the clock, not the vendor count
There's no consensus on a universal percentage of IT spend to set aside. The better practice is to price options against time instead.
Kevin M. Coles, a fractional CIO and founder of Coles Technical Group, says an alternative is worth paying for only if you can answer three concrete questions. "How long would a switch actually take today, in weeks, not intentions? What does one day without this system cost the business? And is the backup truly independent? Most fail the third."
Michele Grant, an M&A integration leader whose work includes a $30 billion media merger, uses a formal method: measure how long the business can survive an outage against how long a switch would take.
Business process owners set each process's maximum tolerable downtime. The CIO then compares that limit with how long it would take to switch to an alternative, including contract negotiation, data migration, testing and training.
If switching would take longer than the business can tolerate, the CIO must weigh the cost of keeping an alternative ready against the disruption it could prevent. If the switch could be completed within that limit, "an exit plan tested once a year protects the service," she said.
Coles sees optionality earning its cost in identity, email and data backups, "where a lockout stops the whole business." It rarely does for line-of-business apps "with a clean export path and a contract you can exit in 30 days."
Karmendra Pandey, practice architect in TEKsystems' AI and machine learning practice, offers a cost test. CIOs should pay to preserve options where switching would be slow and hard to reverse, such as "the data layer, the identity layer, long contracts," but "accept lock-in cheerfully where switching is a configuration change," Pandey said.
Even then, the cost matters. If maintaining an alternative each year would cost a substantial share of a forced migration, "you're not buying protection, you're buying a second estate to maintain," Pandey noted.
The carrying cost nobody budgets
"The license for a second platform is the smallest cost on the page," said Allan Dabre, a senior manager in technology compliance and AI risk at PwC, speaking for himself. "The real cost is governance: two access review cycles, two vulnerability programs, two sets of audit evidence."
Worse, the fallback decays. "Patching slips. Runbooks decay. Reviews get deferred because the system is barely used," Dabre said. "Funding an option you will not maintain is worse than not having one. It is unmanaged risk with a second invoice."
"A warm standby that's never tested isn't an option. It's the feeling of one."
Kevin M. Coles
fractional CIO, Coles Technical Group
Pandey tells CIOs to budget three numbers: standby cost, exercise cost and switch cost. "Most budgets only contain the first, which is how companies discover their 'alternative' takes nine months to stand up." In AI systems, every prompt, policy and model change must be evaluated against both paths, or "the redundancy is decorative."
Three costs of keeping an alternative available
Keeping a backup vendor or technology platform available costs more than its licensing fee. CIOs need to account for three expenses when deciding whether an alternative is worth maintaining.
Standby cost: Licenses, infrastructure and staff required to keep the alternative ready.
Exercise cost: Testing, data synchronization, security checks and practice runs to keep the alternative usable.
Switch cost: Migration, integration, testing, training and possible business disruption when switching.
The question isn't just how much an alternative costs to maintain, but whether the business can actually use it when needed — and whether that flexibility justifies the expense.
Grant shows how those costs can be missed. The second contract is one budget line, but testing, data synchronization and training are "spread across the operating budgets of several teams." She prefers them shown as one annual figure, with exit criteria and a review date for every alternative.
Coles puts it more bluntly: "A warm standby that's never tested isn't an option. It's the feeling of one."
Redundancy that isn't
Some paid-for alternatives fail because they aren't truly independent. For example, "a second cloud or backup product signs in through the same identity provider, so when identity goes down, both sides are locked out," Coles said. Backups stored in the same tenant, or "redundant" circuits entering the building at the same point, fail the same way.
Those shared dependencies can be harder to spot in AI systems.
"Redundancy is a property of failure domains, not of vendor logos," Pandey said. He regularly finds two model providers behind the same identity layer and gateway, or two clouds sharing one identity provider and one continuous integration/continuous delivery control plane. Count an option as real "only if the shared layers are the ones you can survive losing."
“An alternative that takes six months to activate may provide useful leverage before an annual renewal. It is worthless as the answer to a two-hour continuity requirement.”
David Tyler
CEO, Outlier Technology
The price of having no way out
"It's not the migration cost, it's the migration cost paid on the vendor's schedule, at the moment of maximum weakness," Pandey explained. In AI, he noted, models are deprecated and repriced "on quarters, not years."
That can weaken a buyer's leverage long before a switch is needed.
Bill Huber, a partner at ISG, said what you pay depends partly on how long it would take to leave. For high-migration-cost platforms such as VMware under Broadcom, enterprises must prepare alternatives well before renewal. "The renewal date is the deadline. The migration timeline tells you when to start," said Huber.
Dabre sees regulatory fragmentation, vendor consolidation and AI-driven repricing making lock-in costlier. "Customers with a credible exit negotiate. Customers without one get a notification." He urges a trigger set in advance, such as moving if a vendor raises prices more than 15% in a year. "No future leadership team pulls a trigger nobody defined."
Grant adds the M&A angle. Savings from retiring a duplicate system start only when its contract permits exit, "so the exit terms in each contract set the date the synergy begins."
"An alternative that takes six months to activate may provide useful leverage before an annual renewal," Tyler said. "It is worthless as the answer to a two-hour continuity requirement."
"If the budget doesn't include periodically running real work down the alternative path, strike the line item and admit you standardized."
Karmendra Pandey
practice architect, TEKsystems' AI and machine learning practice
Cheaper options, and when to consolidate
The best-value optionality often isn't a running duplicate. Tyler says the better protection may be a contractual right, such as clear terms for exit or disaster-related outages, or retaining staff who know how to make a switch. It may also mean keeping an up-to-date record of the business rules that would need to be moved, since recovering that knowledge “can be a larger task than installing the replacement technology.”
Huber says supplier governance, FinOps and software asset management programs pay for themselves by exposing entitlements and renewal dates. An accurate model of the cost of staying, switching and maintaining an alternative doubles as a negotiating tool.
In a recent data center procurement with competing Dell- and Cisco-based bids, credible alternatives won better lead-time commitments "without requiring a commitment to operate two complete environments indefinitely," Huber said.
Some other options also pay their way. Pandey says a routing layer between applications and AI providers makes AI models swappable, letting teams send work to the cheapest model that can handle it. In his benchmarking, prices for the same work vary more than tenfold, so “the option also pays for itself in ordinary times, not just on the bad day.”
When an alternative can't pass the test, consolidation may be the better risk decision. "If the budget doesn't include periodically running real work down the alternative path, strike the line item and admit you standardized," Pandey said. "That's cheaper than pretending."
As Tyler put it, "The objective is not to eliminate dependencies. It is to know which dependencies are worth having, what they expose the organization to, and whether the available options can be exercised in the time the risk allows."
Pam Baker is a freelance journalist and the author of books including ChatGPT For Dummies and Generative AI For Dummies. Baker is also an instructor on AI topics for LinkedIn Learning and a member of the National Press Club, the Society of Professional Journalists and the Internet Press Guild.
