AI can help traverse and summarize the items and directed links recorded in a Benefit Dependency Network. It can find disconnected items, show paths, identify converging dependencies, list assumptions and prepare review questions.
It cannot prove that one change will cause a benefit, that the network is complete or that an unrecorded connection does not exist. A BDN expresses contribution logic for stakeholders to test. For the method itself, see how to build a Benefit Dependency Network.
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.
First establish the network’s direction and scope
Different BDN diagrams draw arrows in different directions. Ask:
PromptState what an arrow from A to B means in this model. List the item types or columns and the network scope. Do not begin path analysis until the direction is explicit.
Without that step, an apparently correct path can be read backward.
Eight useful BDN questions
| Question to ask | Data required | Safe interpretation |
|---|---|---|
| Which items have no recorded upstream connection? | Directed links and the first-column rule | A missing recorded link, not proof of no strategic value |
| Show the recorded path from this strategy to this technology | Directed graph and item types | Paths in the model, not proof of causality |
| Which benefits depend on the same business change? | Benefit and change links | A shared recorded dependency requiring review |
| Where do several delivery paths converge? | The full graph | A concentrated dependency in the model |
| Which strategies have no recorded objective, program or project support? | Strategy-to-delivery links | A planning gap in the model |
| Which links are unsupported assertions? | Link notes or validation fields, if the model has them | Evidence gaps, only if those fields exist |
| What changes in the network if this item is removed? | Graph structure | A change to the model’s structure, not operational impact |
| Summarize this network for an executive review | Items, links, owners and status if recorded | A narrative of the model with explicit assumptions |
Prompt 1: find disconnected work
PromptList every item outside the first column that has no recorded upstream connection. Show its type, owner and status if available. Label the result "missing recorded connection"; do not call the work unaligned or unnecessary.
The result identifies questions to investigate. A valid connection may exist outside the model.
Prompt 2: trace a complete recorded path
PromptStarting with [technology or project], show every recorded upstream path to a strategy. Present each path as an ordered list of item names and types. If no complete path exists, show the furthest recorded path and where it stops.
Ask the AI to cite the links, not merely produce a plausible narrative.
Prompt 3: identify shared dependencies
PromptWhich business changes, enabling changes, systems or technologies support more than one benefit or objective? Rank them by the number of distinct downstream paths, then list those paths. Do not equate path count with criticality.
A convergence point can be important, but importance also depends on substitutability, timing, capacity and risk that may not be in the network.
Prompt 4: challenge benefit logic
PromptFor each proposed benefit, list the recorded changes and enablers on which it depends. Generate stakeholder questions about missing owners, measures, timing, evidence and alternative explanations. Do not claim that the dependencies guarantee the benefit.
This turns the AI into a structured reviewer rather than an automatic approver.
Prompt 5: prepare an executive summary
PromptSummarize the network in five sections: strategic objectives, expected benefits, required business changes, enabling investments, and unresolved assumptions. Cite the model items and links used. Put missing owners, measures or connections in a separate section on what limits confidence.
What the network cannot answer alone
The network alone cannot prove:
- that a benefit will occur;
- how much of the benefit one change will cause;
- that an unlinked initiative has no value;
- the operational impact of removing a system;
- that every dependency is recorded; or
- that the planned timing and resources are feasible.
Those need benefits-realization plans, owners, measures, delivery evidence and operational analysis.
Where Useful EA fits
In the Useful EA Benefit Dependency Network, an arrow from A to B means B depends on A, and arrows run from strategy toward delivery. Its alignment indicator flags an item outside the first visible column that has no recorded upstream connection; it doesn’t test whether the full chain reaches a strategy. Keep both rules in the prompts above.
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 the Benefit Dependency Network, an assistant can read a network’s items, the connections between them and its views. Under the current strategy weight set it also sees each program’s and project’s score, rank and whether it is above the funding line, and it can compare two weight sets to show what crossed the line and the decisions recorded.
If you’d rather work from a file, export the model as CSV: each row is one item, with its column and its connections listed by item name.
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 traverse the network, expose omissions and prepare better review questions. Keep benefit ownership, causal judgment and investment decisions with accountable people.
Sources
Vendor documentation is cited as context for how the market describes these patterns, not as evidence about Useful EA.
- BOC Group: ADOIT AI Assistant — natural-language search and limited object and relation operations under repository permissions
- Ardoq: Omnipresent AI Assistant — natural-language repository questions, document extraction and changes staged for authorized review
- Useful EA: Bottom-Up Planning with a Benefit Dependency Network — BDN terminology, arrow direction and the limits of the method