AI Changes Everything About Application Portfolio Management
Application portfolio management has always been resource-intensive. Gathering data on 200 applications, scoring them against technical and strategic criteria, maintaining that data over time as systems change and strategies evolve — the work is relentless, and it's mostly manual. The result is that most organisations do it once, file it, and revisit it only when forced to by a transformation programme or an audit.
AI is changing that calculus. Not by eliminating the need for human judgement, but by collapsing the data collection and initial analysis phases that consume most of the time.
AI-Assisted Scoring
Scoring an application portfolio manually means asking someone — usually the IT leader or a consultant — to evaluate every application against a consistent set of criteria: technical health, vendor support status, usage levels, business criticality, strategic alignment. For a portfolio of 200 applications, that process takes weeks and is inevitably inconsistent. The scoring of application 47 reflects the energy levels of week three, not the same baseline as application 3.
AI-assisted scoring changes the dynamic. Given vendor contract data, usage metrics, integration dependencies, and strategic context, a language model can propose initial TIME classifications and technical health scores across the full portfolio in minutes. The proposals are not decisions — they require review and contextualisation by someone who understands the business. But they shift the work from blank-slate analysis to informed validation.
The practical effect is significant. An IT leader who would previously spend two months gathering and scoring data can now spend two weeks reviewing and refining AI-generated proposals. The final output has the same quality — arguably better, because systematic analysis doesn't fatigue — and the process is four times faster.
Capability Generation
Building a capability model from scratch is one of the most daunting tasks in enterprise architecture. Start too high and you end up with generic categories that don't map to real systems. Start too low and you build a process map that no one can maintain.
AI can dramatically accelerate the bootstrapping phase. Given an industry vertical and a set of strategic themes, a language model can generate a two-level capability map — 15-25 L1 capabilities, each with 3-6 L2 children — calibrated to the context. This is not a finished product. Industry reference architectures embedded in AI training data are not a substitute for understanding your specific organisation. But they are an excellent starting point.
The alternative — building the capability map in a series of workshops from a blank whiteboard — typically takes 4-6 weeks of facilitated sessions and still produces something that needs significant cleanup. AI-assisted generation compresses that into a first draft that can be reviewed and refined in a single workshop.
The downstream effect is that capability mapping becomes accessible to organisations that previously couldn't justify the time investment. A 300-person company with one IT leader can now have a credible capability model without engaging a team of consultants to build it.
Strategy Parsing
The disconnect between strategic documents and IT decisions is one of the persistent frustrations in enterprise architecture. Strategy lives in presentations and board papers. IT decisions live in backlogs and budget cycles. Getting from one to the other requires someone to read the strategy carefully, extract the themes and priorities, and translate them into capability requirements.
That translation step is something AI handles well. Upload a strategy document — an annual report, a digital strategy, a board presentation — and a well-prompted language model can extract the strategic themes, identify the implied capability requirements, and propose mappings to an existing capability model.
This is not a trivial capability. Strategy documents are rarely structured for machine consumption. They use business language, not technical vocabulary. They contain intent and aspiration rather than explicit requirements. Making the connection between "we will win through customer intimacy" and the specific capabilities that statement implies requires the kind of semantic reasoning that language models are genuinely good at.
What Stays Human
It would be a mistake to overstate what AI can do here. The judgements that matter most in portfolio management — whether a capability is truly strategic, which of two competing applications deserves the investment, what the risk of a decommissioning decision is — require context that AI doesn't have. They require someone who understands the organisation's politics, the vendor relationships, the integration dependencies that aren't in any system, and the direction the business is actually heading.
AI compresses the data collection and initial analysis work. It makes the starting point better and gets there faster. But the decisions still require human ownership.
The organisations that will get the most value from AI-assisted APM are not the ones who use it to automate decisions. They're the ones who use it to spend less time gathering data and more time making decisions.
APM in 2026 is AI-native or it's a spreadsheet. The question is not whether to incorporate AI into portfolio management — it's how quickly you can do it without compromising the quality of the decisions it supports.