insora Research Summary: Adaptive AI Assistant for SME Productivity

Publication year: 2025 • Audience: Product, GTM, and prospective customers • Document type: Research summary

1. SME context and productivity challenge

SMEs represent over 99% of European firms and contribute a majority share of private‑sector employment and value‑added. Despite their central role, SMEs exhibit persistent productivity gaps compared to larger enterprises and US peers. Interviews and surveys with European SME stakeholders highlight administrative burden, fragmented tools, and the rising complexity of knowledge work as structural obstacles to productivity.

2. System overview and philosophy

insora is an adaptive, agentic assistant for SMEs combining retrieval‑augmented generation (RAG) with multi‑agent orchestration and collaboration. The system shifts from pre‑configured, industry‑specific software to context‑aware software that learns from organizational behavior to optimize for organization, team, and individual workflows. It integrates with existing tools via standardized APIs, emphasizing rapid time‑to‑value through minimal setup and optional expert controls for customization.

• Adaptive, context‑aware responses grounded in internal data via RAG.
• Behavioral learning from usage patterns (preferred information types, response formats, automation pathways).
• Seamless integration with current systems; preserves prior investments.
• Low‑friction onboarding with automatic context inference and optional expert tuning.
• Strict quality and filtering to mitigate LLM hallucinations and increase trust.

3. Functional architecture

The architecture maintains dynamic, searchable representations of organizational knowledge using vector embeddings and semantic search across multiple data types (documents, email, structured data, images, diagrams). It supports at‑scale retrieval over very large corpora while minimizing runtime context size sent to LLMs through statistical selection.

• Semantic knowledge base with multi‑level retrieval and automated ingestion.
• Multi‑agent orchestration: an orchestrator coordinates task‑specific agents that can be created dynamically to execute tool/API calls across systems.
• Shared collaboration spaces with real‑time co‑working, action logs, and personal workspaces with invite‑based sharing.
• Continuous improvement via implicit and explicit feedback signals at organization, team, and individual levels.

Figure: High‑level data flow - ingest → embed → retrieve → compose → orchestrate agents → act/log → learn.

4. Adoption barriers and drivers

Across industries, several barriers consistently emerged:

• Complexity and expertise: Concerns about implementation, maintenance, and limited internal skill.
• Privacy and security: Data protection, reliability, and safety requirements.
• Value‑perception gap: Decision‑makers struggle to quantify benefits in financial terms.

Key drivers include innovation orientation, demonstrable efficiency gains, social proof/peer adoption, and strong trust signals (security certifications, encryption, auditability, explainability, and human control).

5. Market segmentation and early adopters

Segmentation integrated structural (industry, size, digital maturity) and behavioral factors (time valuation, AI readiness, implementation time). Early adopters are expected to experience acute pain points, possess higher technical/business acumen, tolerate uncertainty, and use the system intensively - enabling faster iteration cycles.

• Optimal early adopters: Small firms (avg. ~24 employees) in information and communication industries.
• Time‑to‑value: Short implementation cycles (~2.2 months for early adopters).
• Scaling path: After product iteration with small tech firms, expand to mid‑sized and large enterprises for revenue maximization.

6. Key quantitative findings

• Value‑perception gap: ~77.20% exhibit underestimation of benefits.
• Trust and security: ~80% value security & transparency features as decisive.
• Early adopters: LTA ≈ 0.34; implementation ≈ 2.2 months; WTP ≈ €39/month.
• Large enterprises: LTA ≈ 0.18; WTP ≈ €113/month; high total segment value.
• Industry view: Information & Communication shows high adoption potential; Manufacturing exhibits strong value potential.

Segment valuation model (ordinal optimization): Value = LTA × WTP × Expected Users, aggregated across companies in segment.

7. Security, privacy, and compliance

• GDPR compliance, encryption, and audit trails as core design requirements.
• Cloud‑first (for speed and scale) with on‑premises option for specific security postures.
• LLM calls to leading providers with encrypted inputs; Data Processing Agreements to restrict training use and geographic transfer.
• Security certifications (e.g., ISO‑27001) targeted to build trust and reduce procurement friction.

Given European privacy priorities and trust barriers, transparent safeguards and third‑party attestations are critical adoption levers.

8. Cost model and efficiency

LLM usage typically dominates operating cost. The system minimizes in‑context tokens by statistically selecting the most relevant information for each task, yielding orders‑of‑magnitude cost reductions compared to naïve full‑context strategies.

• Predictable cost profile via time‑series of expected token consumption × per‑token price.
• Efficient context construction reduces cost by ~10^3–10^4 versus full context retention.

9. Go‑to‑market implications

• Phase 1: Small tech firms (information & communication); fast iteration; prove concrete efficiency wins; collect references.
• Phase 2: Mid‑sized and large enterprises; expand validated use cases; formalize security proof and procurement readiness.
• Positioning: Emphasize trust & transparency alongside measurable outcomes (time saved, turnaround time, output quality).
• Enablement: Provide quickstart integrations, sample workflows, transparent logging, and ROI calculators.

Pricing and packaging should reflect value realization profiles across segments and be anchored to demonstrable time savings and usage intensity.

10. Limitations and next steps

• Survey preference data may diverge from revealed usage; pilots are required to validate WTP and engagement.
• Security emphasis must not crowd out usability; maintain focus on intuitive design and rapid workflows.
• Continue refinement of agent orchestration and retrieval policies to maximize quality and minimize cost.
• Expand case studies and references across industries to accelerate mainstream adoption.

References (public sources)

1. McKinsey Global Institute (2012). The social economy: Unlocking value and productivity through social technologies.
2. IDC (2004). The High Cost of Not Finding Information. Feldman & Sherman.
3. Asana (2023). Anatomy of Work Index.
4. Microsoft (2023). Work Trend Index (digital debt).
5. Qatalog × Cornell (2021). Context switching in knowledge work.
6. Mark, G., Gudith, D., & Klocke, U. (2008). The cost of interrupted work.
7. Panopto (2018). Inefficient knowledge sharing costs large businesses $47M/year.

Disclaimer

This independent research summary paraphrases internal research findings and does not disclose or reproduce any confidential or NDA‑protected documents. All product descriptions and quantitative insights are presented at a high level and may evolve as the product develops. External statistics are indicative and sourced from publicly available reports listed above.