AI Glossary / Graphs and symbolic AI / Intermediate

Case-based planning

Definition

Case-based planning constructs a plan for a new situation by retrieving relevant previous planning experiences and adapting them. A stored case can include the original goals, context, actions, outcome, and lessons from failure or repair. The method uses experience as a starting point while checking whether the adapted plan fits the current problem.

What it means

Instead of always building a plan from scratch, a planner asks whether it has solved something similar before. Similarity must concern features that matter to success, not merely shared words or a familiar title. A past plan can supply useful ordering, resource choices, or warnings about interactions.
Reuse is not blind repetition. New constraints may invalidate an old action, and the old outcome may depend on conditions that are absent today. A useful case therefore records why the plan worked or failed, not just the final action list. Adaptation and validation are central parts of planning from experience.

How it works

A system represents the new goals and context, searches its case memory, and selects one or more relevant experiences. It modifies the retrieved plan to address differences, then evaluates whether the result meets the current goals and constraints. If problems arise, it can repair the plan or seek another case.
The experience from the new attempt may be stored with an outcome and useful indexing features. Failure cases can help anticipate problems rather than merely documenting them afterward. Implementations vary in their retrieval, adaptation, and validation methods; some combine case reuse with a planner that reasons from an explicit action model.

Three practical examples

• A maintenance planner adapts a previously successful service sequence for similar equipment while checking the current machine's access and shutdown requirements.
• An event organizer reuses a prior setup plan but changes staffing and timing for a different venue.
• A workflow assistant retrieves an earlier recovery procedure and checks whether the current failure conditions match before proposing adapted steps.

Think of it this way

Think of an experienced cook adapting a familiar recipe to different ingredients and equipment. The earlier recipe offers structure, while experience warns which substitutions may fail. The analogy is limited because a formal planner must represent the relevant conditions and validate changes; remembering that something worked once is not enough to guarantee that it will work again.

Worked example

Suppose a fictional stored plan prepares two rooms using one team. Inspect room A for 10 minutes, set it up for 20, then inspect room B for 10 and set it up for 20. Total duration is 60 minutes. The new request adds room C, with the same times, and a 75-minute deadline.
Simply appending C produces a 90-minute plan and fails the deadline. If a second qualified team is genuinely available and the rooms have no shared-resource conflict, the adapted plan can assign A and B sequentially to team 1 for 60 minutes and C to team 2 for 30 minutes. Both finish within 75. If both teams require the same single setup tool, that assumption must be revisited. The case supplies an action pattern; current resource and timing checks determine whether its adaptation is feasible.

Illustrative case study

At fictional Cedar Facilities, a planning assistant retrieves prior room-preparation cases. It initially favors a visually similar event, but staff notice that its successful setup relied on overnight access unavailable for the new booking. They add access-window information to the case representation and review the proposed adaptation.
The revised process records the reason for rejecting the old schedule as well as the final plan. This fictional case illustrates learning from a mismatch and improving future retrieval. It does not claim automatic transfer of expertise or a measured reduction in planning effort.

Common misconceptions and mistakes

Case-based planning is not simply retrieving a document or copying a checklist. It requires using the experience to construct a plan appropriate to a new problem. A close semantic match is not necessarily a close planning match: resource availability and action preconditions can matter more than wording.
Past success does not establish present feasibility, and a failed case is not necessarily useless. Its failure conditions may be valuable warnings. Finally, adapting an old plan is not always cheaper than fresh planning. Extensive changes or a poor case library can make reuse difficult, so systems should be able to reject unsuitable cases.

Business use

Businesses can benefit when workflows recur with meaningful variations and past outcomes are recorded well. Useful cases include context, constraints, dependencies, what actually happened, and the reasons for repairs. Governance should distinguish a proposed plan from an approved procedure and prevent outdated cases from silently overriding current requirements. Evaluate retrieval relevance and adapted-plan validity separately: a good search result can still lead to a bad adaptation.

People also ask

Does a case contain only successful plans?

No. Failures and repairs can help a planner anticipate and avoid recurring problems.

Is case-based planning the same as case-based reasoning?

It is a planning-focused use of the broader idea of solving new problems through prior cases.

Can it work with other planning methods?

Yes. Case retrieval and adaptation can be combined with model-based reasoning or fresh search, depending on the system.

Check your understanding

A retrieved plan used two teams, but the new situation has only one. Can the original completion time be reused unchanged?

Answer: Not without justification. The changed resource constraint requires rescheduling or another adaptation, followed by a new feasibility check.

Related terms

• Case-based reasoning
• Episodic memory
• plan adaptation
• plan repair
• retrieval
• action precondition
• planning domain

Learn next

Write a small planning case with separate fields for goals, context, actions, dependencies, resources, and outcome. Change one condition and identify exactly which steps become invalid. This exercise shows what a useful case memory must preserve beyond a narrative description of past success.

More in Graphs and symbolic AI

• CART - CART stands for Classification and Regression Trees, a methodology for learning decision trees from labeled data.
• Cascaded finite-state transducers - Cascaded finite-state transducers are a sequence of finite-state processing stages in which one stage's output becomes input for the next.
• Case agreement - Case agreement is compatibility in grammatical case between words or phrases that a language's grammar requires to agree.
• Cell decomposition - Cell decomposition is a planning technique that divides a space into regions called cells and represents usable connections between them.
• Certainty effect - The certainty effect is a pattern in decision making in which a guaranteed outcome receives disproportionate weight compared with outcomes that are merely...
• Certainty equivalent - A certainty equivalent is a guaranteed outcome that a decision maker values equally to a specified uncertain prospect.

Sources

Sources: Kristian Hammond, original CHEF: A Model of Case-Based Planning: https://cdn.aaai.org/AAAI/1986/AAAI86-044.pdf . Hammond, original The Use of Remindings in Planning: https://escholarship.org/uc/item/77m9448n . Original Model-Lite Case-Based Planning research: https://arxiv.org/abs/1207.6713 .

Previous: Case agreement Next: Case-based reasoning