AI can help filter, group, compare and explain application facts already recorded in a portfolio: ownership, cost, lifecycle, capability support, feature fit, targets, reviews and application-to-application links.
It cannot discover an unrecorded dependency, decide that an application is safe to retire or turn incomplete portfolio data into a complete impact assessment. The strongest questions ask for evidence, missing data and a review queue, not an automatic disposition.
Disclosure: I’m Nick Malik. I built Useful EA, a commercial set of architecture tools, and the section on Useful EA below describes it. The method in this article works with any repository and any AI assistant.
Start with a data-quality question
PromptDescribe the application-portfolio fields and relationship types you can access. Count populated, missing and stale values for owner, annual cost, lifecycle, fit, capability links and last validation. State the portfolio scope used.
An answer about "all applications" is meaningful only when scope and completeness are visible.
Nine useful portfolio questions
| Question to ask | Data required | Safe interpretation |
|---|---|---|
| Which applications support this capability? | Application-to-capability links | Recorded support, not every operational dependency |
| Which applications appear to overlap? | Capability links, descriptions, owners and usage evidence | Consolidation candidates requiring validation |
| Which high-cost applications have poor recorded feature fit? | Annual cost and a defined fit scale | A triage ranking, not a retirement decision |
| Which applications lack an owner, cost or recent validation? | Those fields and review dates | A data-quality queue |
| Which applications approach lifecycle or support deadlines? | Recorded dates and status | Exceptions within the recorded data |
| What recorded flows would retirement touch? | Directed application links | Recorded neighbors or paths, not a complete impact analysis |
| Which targets lack an owner, rationale or dated work plan? | Target flag, ownership, rationale and work items | An execution-readiness gap |
| How is cost distributed by capability, owner, vendor or lifecycle? | Cost plus a grouping field | Totals, with currency and scope caveats |
| Which findings rely on estimates or missing data? | Provenance or confidence markers, if recorded | A confidence review; otherwise the AI must say the distinction is unavailable |
Prompt 1: build a portfolio quality queue
PromptList active applications missing an accountable owner, annual cost, lifecycle status, capability relationship or recent validation. Group by missing field and show the source record. Do not treat missing values as zero or "none."
This is often a better first use of AI than asking for strategic recommendations from incomplete data.
Prompt 2: screen for possible overlap
PromptFind applications recorded against the same business capability or similar function. For each cluster, compare owner, users, cost, fit, lifecycle and dependencies. Label the result "possible overlap." List the evidence required before consolidation and do not recommend retirement solely from similarity.
Prompt 3: find cost-and-fit review candidates
PromptRank applications with a non-zero recorded fit score by annual cost multiplied by the size of the fit gap. Show cost, fit, formula and missing values. Explain that the ranking identifies review candidates rather than making a decision.
Prompt 4: examine a proposed retirement
PromptFor [application], list its recorded capabilities, direct upstream and downstream application links, owner, lifecycle, cost, fit, target status and open work items. Separate recorded facts from missing information. Create a checklist of people and evidence needed before a retirement decision. Do not claim that the recorded links are a complete impact analysis.
Prompt 5: prepare a portfolio review
PromptProduce a review agenda with: overdue or unvalidated records; high-cost poor-fit candidates; possible overlaps; lifecycle deadlines; targets without dated work; and material missing data. Cite every application record used and put unresolved questions under the responsible owner.
Prompt 6: explain the result to a stakeholder
PromptSummarize the portfolio finding for a [CIO, business owner or application owner]. State the evidence, the limitations and the decision still required. Avoid technical jargon and do not invent benefits, savings or risk levels.
What the portfolio cannot answer alone
The recorded portfolio cannot by itself prove:
- that an application is redundant;
- that retirement is safe;
- that a data-flow record represents every interface or dataset;
- that low feature fit means low business value or poor technical condition;
- that an owner agrees with a recommendation; or
- that estimated savings will be realized.
Those need stakeholder confirmation, technical and business analysis, control and retention checks, and an executable change plan. For the decision method, see application rationalization.
Where Useful EA fits
Claude, ChatGPT, Claude Code, Cursor and VS Code can connect to Useful EA directly at mcp.useful-ea.com. You sign in with your usual Useful EA account, and the assistant reads only what you can already open, with your permissions; it can’t change your models. A company’s administrator can turn assistants off for the company’s members. How to connect an AI assistant.
Connected to Application Portfolio Manager, an assistant can list the applications in a portfolio, with filters, and open one application with its custom fields, its capability links and the applications it depends on or that depend on it.
If you’d rather work from a file, export the portfolio as CSV. Each row is one application, with its disposition, primary and supporting capabilities, owner and contacts, vendor and version dates, annual cost, business value, technical fitness, feature fit, target flag, owner intent and last-validated date from review campaigns, the applications upstream and downstream of it, and your own custom fields.
Two distinctions matter for the prompts. The built-in fit field measures feature fit, from -2 to +2: how well the application’s features match what the business needs. Business value and technical fitness are separate scores, each 1 to 5, and an application not yet scored is “not assessed”, which is not the same as a low score. And upstream and downstream links are flows people have recorded, not discovered interfaces.
AI can also help in the other direction: Add from JSON gives you a template built from your portfolio’s own fields for an assistant to fill from a description of one application. The pasted result opens in the normal form, and nothing is saved until you review it and click Save.
Only give architecture data to an AI assistant your organization has approved for that information. An export can include names, owners, costs and plans that are confidential.
Practical rule
Use AI to find exceptions, assemble evidence and prepare human decisions. Ask the portfolio for what it records; ask accountable owners whether the record reflects reality.
Sources
Vendor documentation is cited as context for how the market describes these patterns, not as evidence about Useful EA.
- Ardoq: Q1 2026 AI roundup — questions such as which applications support a capability, and diagrams converted into candidate data
- SAP LeanIX: MCP server for enterprise architecture — structured AI access to application records, capability maps, decisions and relationships
- Essential AI — natural-language questions grounded in a connected model, with answers linked to model elements