Research Positioning

Problem framing

Evidence-backed synthesis

  • SME decision-making is shaped by uncertainty, resource constraints, concentrated owner-manager judgment, weakly structured choices, and uneven information quality.
  • The literature reviewed so far suggests that SME decision problems are not just about missing data in the abstract; they are often about making time-sensitive choices with incomplete information, limited analytical capacity, and limited organizational slack.
  • The SME AI literature is stronger on adoption barriers and functional applications than on the architecture of decision support systems for owner-manager decisions.

Conceptual framing

  • A narrow way to frame the problem is:
    • SME decision support requires more than retrieval, prediction, or generic memory because many real decisions depend on reconstructing the relevant business situation across goals, constraints, triggers, actions, outcomes, and evidence.
  • This framing should remain provisional until supported by stronger task-level evidence.

Why SME decision support is hard

Evidence-backed synthesis

  • SME decisions are often made under uncertainty rather than under stable planning assumptions.
  • SMEs face stronger short-term liquidity pressure and lower human-resource slack than larger firms.
  • Decision authority is frequently concentrated in owners or small management teams.
  • Tactical and strategic decisions are often weakly structured, which makes information requirements harder to specify in advance.
  • Information use may be irregular, incomplete, or only partly formalized.

Conceptual framing

  • Together, these conditions make SME decision support hard because the relevant decision state is distributed across:
    • partial records
    • current constraints
    • tacit manager knowledge
    • prior episodes
    • shifting external conditions

Why current AI support is insufficient

Evidence-backed synthesis

  • Current AI systems appear strongest at:
    • document retrieval
    • summarization
    • question answering
    • prediction in bounded tasks
    • limited conversational continuity
    • workflow guidance in repeated tasks
  • Adjacent literature shows that fragment retrieval is not the same as sufficient context.
  • Structured decision-support work suggests that ordinary RAG is weak when decisions require weighting, traceable reasoning, or explicit comparison of alternatives.
  • Workflow-memory and episodic-memory papers suggest useful directions, but they are not SME-specific evidence that business decision reconstruction is already solved.

Conceptual framing

  • A conservative interpretation is that current systems are often optimized for helpfulness over fragments, not for explicit reconstruction of decision episodes.
  • The likely insufficiency is not that AI lacks all memory, but that the available memory forms may be misaligned with what decision-grade SME reasoning requires.

Candidate research gap

Evidence-backed synthesis

  • There is no strong SME-specific literature base yet showing that current AI systems adequately reconstruct decision-grade business context for owner-manager decisions.
  • There is adjacent evidence that:
    • retrieval can be insufficient
    • conversational memory is limited
    • workflow memory helps with procedure
    • process-intelligence literature supports traceability and event-aware reasoning in structured settings

Conceptual framing

  • Candidate gap:
    • a missing integration between SME decision-support needs, memory or context design for AI systems, and traceable reconstruction of business decision situations.
  • More narrowly:
    • the gap may be not “AI for SMEs” in general, but AI support for weakly structured SME decisions that require cross-source, time-aware, traceable context reconstruction.

Competing framings of the gap

  1. Readiness framing
    • The real gap is SME capability, not AI architecture.
    • On this view, data quality, budgets, infrastructure, and skills are the primary bottlenecks.
  2. Traceability framing
    • The real gap is not memory but explanation and auditability.
    • On this view, decision traceability and structured rationale matter more than richer memory representations.
  3. Task-design framing
    • Current systems seem insufficient because they are built for retrieval, drafting, and productivity, not for decision support.
    • On this view, the key problem is evaluation target mismatch.
  4. Process-intelligence framing
    • The relevant precedent is process mining or decision mining rather than LLM memory.
    • On this view, the solution space may already exist for structured workflows and only needs adaptation.
  5. Scenario-memory framing
    • Current systems lack a sufficiently rich representation of the business decision episode itself.
    • On this view, fragment retrieval, user memory, and conversation history do not yet reconstruct the situation well enough for decision-grade reasoning.

Strongest version of the scenario-memory thesis

Working Hypothesis

  • Many SME decisions are weakly structured, cross-document, time-dependent, and constraint-heavy.
  • Current AI systems appear better at surfacing relevant fragments than at reconstructing the full decision situation in a traceable and reusable form.
  • Therefore, a useful next step may be an AI architecture for SMEs that combines:
    • document memory
    • persistent factual memory
    • event or episode-level memory
    • workflow or process context
    • decision traceability
  • Under this formulation, scenario memory is not a standalone memory primitive already established in the literature, but a design shorthand for combining these capabilities around a business decision episode.

Weakest / most defensible version of the thesis

Working Hypothesis

  • The current literature suggests a mismatch between common AI support mechanisms and the context requirements of some SME decisions.
  • In particular, retrieval and summarization alone may be insufficient when decisions require explicit handling of:
    • constraints
    • chronology
    • alternatives
    • outcomes
    • evidentiary support
  • A defensible paper does not need to prove a brand-new memory category.
  • It may be enough to argue that SME decision support needs a more explicit treatment of episode-level context and traceability than current systems typically provide.

What would count as evidence

  • Empirical studies showing that SME decision tasks fail when systems only retrieve documents or summarize fragments.
  • Comparative evaluations where richer contextual reconstruction outperforms plain retrieval or plain conversational memory on SME-like decision tasks.
  • Evidence that owners or managers rely on cross-episode context such as prior decisions, constraints, outcomes, and rationales when making repeat or comparable decisions.
  • Studies showing that process-aware or event-aware context improves decision quality, explanation quality, or user trust in business settings.
  • Evidence that decision-support users need explicit representation of assumptions, constraints, and evidence provenance to act confidently.

What would count as non-evidence

  • Generic claims that AI is useful for business productivity.
  • Evidence that AI improves forecasting, search, drafting, or general analytics without testing decision reconstruction.
  • Memory benchmark gains on unrelated tasks without a credible bridge to SME decision support.
  • Adoption-barrier studies alone.
  • Claims based only on product marketing or anecdotal enterprise-copilot usage.

3 candidate paper contribution statements

  1. This paper synthesizes SME decision-making literature and adjacent AI-memory literature to argue that current support systems are under-specified for weakly structured SME decisions that require cross-source and time-aware context reconstruction.

  2. This paper proposes a conceptual framework for SME decision support that distinguishes document retrieval, persistent factual memory, workflow context, and episode-level decision context, and clarifies where current systems remain limited.

  3. This paper reframes the AI-for-SME research gap from general adoption or functional automation toward traceable reconstruction of decision situations, identifying a narrower and more defensible target for future system design and evaluation.

3 candidate paper titles

  1. From Fragment Retrieval to Decision Context: Positioning a Research Gap in AI Support for SMEs

  2. Toward Decision-Grade AI for SMEs: Context Reconstruction, Traceability, and the Limits of Current Support Systems

  3. AI Support for Weakly Structured SME Decisions: A Research Positioning Note on Context, Memory, and Traceability