Why Do Some AI-Assisted Modernization Estimates Come In Too Low?

Two published facts frame the decision.

Amazon’s figure is strong, and it compares AI-assisted upgrades with manual upgrades. It does not measure data migration, integration changes, customer testing, or cutover, and it does not claim to. AI tools can help with some of that work too, such as generating tests or mapping data between schemas. Those savings may be real, but they must be estimated from evidence on the program in question.

Estimates come in too low when that limit is ignored. A saving demonstrated on coding work is easy to assume for the rest of the program. The rest of the program has to be assessed on its own evidence.

A vendor support deadline requires the company to deal with an unsupported component. It does not require a rebuild. Engineering has four practical options, and a program can combine them.

A replacement should be chosen when its business case is stronger than an upgrade’s, after transition costs and risks are included. An upgrade can be technically possible while a replacement is still the better investment because of operating cost, security, maintainability, or platform consolidation. Companies should consider retirement more often than they do.

Some companies choose a fifth course for a limited time: they keep the unsupported version running in an isolated environment with compensating security controls. That can be a reasonable decision. It is a risk decision rather than a modernization plan, so it needs an owner, an end date, and the board’s knowledge.

Two mistakes are common. The first is an unnecessary replacement. A team hears the support date and assumes it means we must rebuild, and a deadline becomes a replacement program. The second is an underpriced replacement. A team estimates the new code with AI assistance, extends the savings to the whole program, and presents an affordable number.

Both mistakes affect the operating plan. An unnecessary replacement takes engineering capacity from growth work. An underpriced replacement can overrun inside the hold or finish after the sale. The program may consume cash and engineering capacity during the hold without delivering the expected operating improvement before exit. Issue 5 described the same risk.

A buyer’s technical diligence team will review the runtime versions, vendor support windows, and modernization plan. A seller who can show the option chosen for each affected system, the reason for it, and a fully costed estimate gives the buyer less uncertainty to price.

The vendor sets the deadline, and engineering has to choose the response.

When I am asked how long a system has left, I don't answer based on its age. I look at four triggers.

The first two triggers are external and dated. The second two are internal, and they are usually known only to engineering.

Amazon’s report shows what AI can do for mechanical, well-bounded work such as a framework version upgrade. That matters, because a system that once looked too expensive to upgrade may now be upgraded rather than replaced.

The same tools make replacements easier to propose. AI can reduce the coding effort for a new system, so the code estimate comes in lower than it once would have. Some of those proposals will be sound. Others will be underpriced, because the estimate covers the code and the other work was never assessed.

In one company I worked in, when I retired a data center and moved our databases to Azure SQL, the code changes were modest: mostly connection settings and deployment scripts. Most of the effort went into three kinds of work. The first was moving data without losing or corrupting it. The second was re-pointing every integration that customers and partners depended on. The third was running the old and new environments in parallel and agreeing cutover dates with customers.

In that program, customers had to agree to cutover dates, and the old and new environments ran side by side until they did. Other migrations are handled differently. Some are invisible to customers, and some use a scheduled outage instead of running in parallel. The lesson that applies to all of them is that the transition method, customer coordination, and validation effort each need their own estimate.

This is where a non-technical operating partner can add the most value. When a modernization estimate arrives, ask whether every line beyond the code was assessed and costed, and what evidence supports each one.

The engineering lead and the business process owner answer this together. A weak answer is: we’ll discover that during migration.

The platform or architecture lead answers this. A weak answer is: the vendor support date means we must rebuild.

The delivery lead answers this. A weak answer is that AI can rewrite the code quickly, offered in place of those estimates.

The CTO and the operating partner answer this together. A weak answer is that it will be the next owner’s problem.

For each one, record the triggering date, the option chosen, the reason, and the owner. Any system on .NET 8 or .NET 9 reaches that point on 10 November 2026. The same check applies to other runtimes, databases, and cloud services.

Ask for data migration, integration changes, testing, cutover, parallel running, and retirement as separate lines, each with its own basis. Do not accept an AI saving on the coding line as a reason to reduce the others.

Where an affected system was capitalized, replacing it earlier than expected can mean a revised remaining life, an impairment, or expensing the unamortized balance when the replacement happens. Those are non-cash accounting effects. Keep them separate from the cash the modernization program needs.

The number of systems whose vendor support ends before the planned exit and that do not yet have an agreed plan with an owner, a budget, and a completion date.

Each of those systems is an unresolved exposure inside the hold. We’ll upgrade it does not count as a plan until someone owns it, it is funded, and it has a date.

Portco Brief is a weekly briefing for PE operating partners and portfolio company executives focused on technology, AI, and value creation. If this was forwarded to you, subscribe at portcobrief.com.

Sources

Microsoft: .NET 8 and .NET 9 will reach End of Support on November 10, 2026, and June 29, 2026.

Amazon Web Services, Amazon Q Developer just reached a $260 million dollar milestone, August 1, 2024.