Your company already knows the answer. It just can't find it.
Small organisations, SMEs and NGOs, are the backbone of the economy, yet they have never had the decision intelligence the big players take for granted. Company Brain turns the knowledge you already have, meetings, emails, spreadsheets, the things in people's heads, into a connected map you can query, reason over and act on. With AI, you can compete.
Figures from the pilot system running at GIS Analytics. The interactive graph on this page is a separate, fictional 52-entity demo.
Institutional memory, with structure.
An eight-month knowledge exchange between UCL's Department of Information Studies and GIS Analytics, building the structure small organisations, SMEs and NGOs, are missing: a property graph of people, clients, projects, decisions, evidence and outcomes, extracted from the sources a business already produces.
The system, built by GIS Analytics on Neo4j and driven by natural-language skills, records not just what was decided but why: who proposed it, what evidence supported or contradicted it, what actions followed. Structural transparency, not bureaucratic overhead.
The result is decision-grade memory an SME or NGO can query in plain language, and an audit trail that assembles itself.
For a small team, scattered knowledge is not just untidy. It is the gap between you and an organisation many times your size. Here is what that gap is made of.
Knowledge fragmentation
Critical context lives in spreadsheets, email threads, meeting notes and individual memory. There is no canonical model, and no path between two facts.
Decision opacity
Choices are made without visibility into prior context, the evidence that informed them, or the outcomes they produced. By the third quarter, nobody remembers what was traded against what.
Governance gap
Transparency and audit expectations are rising: the EU AI Act, procurement reviews, ISO. SMEs lack the infrastructure to meet them without diverting headcount to documentation.
Capture signal.
Meeting transcripts, email threads, accounting exports, hand-written notes. Each source is broken into typed records, and every record keeps a link back to the document it came from: its provenance.
Link structurally.
Nodes enter a Neo4j property graph with typed relationships. The model resembles the business, not a database.
Reason over context.
A question pulls every adjacent node: prior context, supporting and contradicting evidence. The recommendation arrives with its citations built in.
Execute and trace.
Recommendations become actions: drafted emails, scheduled tasks, prioritised lists. Every outcome links back to the decision that produced it.
One connected map of how the business runs.
This is the system's actual viewer, embedded live. It renders a fictional demo organisation from a static snapshot: people, clients, projects, decisions, evidence and the typed relationships between them.
About this data: the organisation shown is fictional and ships as a static snapshot. Real deployments run against the SME's own Neo4j instance, in their own environment and under their own control; nothing about a client business appears on this site.
Specific Q&A
Ask a question, get the answer with its evidence attached: who decided, when, and on what basis. Not a document search: the answer comes from what the company actually knows.
Qualitative advice
Recommendations argued from the firm's own context: prior decisions, commitments, cash position, with supporting and contradicting evidence attached.
Simulations
See the knock-on effects before they happen: which projects, people and revenue are exposed, traced along the firm's real relationships rather than guesswork.
Company check-up
A structured review of the graph itself: stale decisions, contradicted evidence, orphaned actions, concentration risk. The business, audited by its own structure.
A working session with GIS Analytics takes one hour: we map a slice of your company into the graph, and you ask it questions.
In the real world
The graph changed the way we see the whole operation. For the first time we can look at the entire ecosystem in one place, how our kitchens, projects and partners connect, and act on it.
GIS Analytics mapped London's Community Kitchen and its family of ventures, the café, the water brand, the youth academy and the climate hub, into a single knowledge graph, so a fast-moving charity network can finally see itself clearly.
Answers you can take to a review.
Each decision is a first-class node linked to its context, its evidence on both sides, the people involved and the actions that followed. The audit trail is not written after the fact; it is the shape of the data.
Property graph, not RDF.
The data model treats nodes and relationships as first-class objects, each carrying properties of its own. It is close to how a founder describes their business: this client, on this project, with that contract.
The choice was made early, in consultation with the UCL team: property graphs trade some of RDF's semantic rigour for queryability, tooling maturity and a model the team can extend without recompiling an ontology.
Live system on real SME data
433 nodes, 538 relationships, 21 decisions on record, deployed in production at GIS Analytics.
● ShippedOpen-source toolkit
The skills, ingest pipeline and property-graph schema prepared for open-source release. Documentation co-authored with UCL.
● In progressSecond deployment · live
A London community-food charity now runs the system in production: early evidence the model generalises beyond the first pilot. Named case study to follow, with their permission.
● LiveBest-practices guidance
A practitioner-facing best-practices guide, co-authored with UCL, on building decision-grade institutional memory in small businesses.
○ PlannedTalk to the team.
This is a knowledge exchange collaboration with a working system. If you run a small business, ask for the one-hour walkthrough: we map a slice of your company into the graph and you question it. Academic groups with adjacent work, and funders looking at this space: write to the PI or the project inbox.