Solutions reference

What we have built.

Four working systems. One programme that makes the capability yours. Everything here has been built.

AI is everywhere. Systems an institution can stand behind are not.

BELMOST builds agentic systems that carry real commercial and regulatory weight. They explain their reasoning. They keep people in charge of what matters. They leave an evidence trail behind every decision. This reference describes the ones we have already built: what each does, the workflow it runs, and the rails it runs on.

Two of these systems were born inside an industry and carry its judgment. Two serve functions every institution runs. The fifth is not a product at all. It is the programme by which a client builds this capability in-house.

Client names and engagement detail are not published here. What follows is the work, described at the altitude we publish. The longer story is told in person.
BUILT FOR AN INDUSTRY · VERTICALBUILT FOR A FUNCTION · HORIZONTALSIGNALCustomer lifecycleintelligenceTelecommunications-bornDISTRICTHousing intelligencecommand centreHousing & urban programmesTENDERProcurementintelligenceAny procurement functionLEDGERComplianceintelligenceAny regulated functionthe people, platform andgovernance behind themTHE AI OFFICEThe programme that builds a client's own capability to run systems like these: mandate, platform, playbook, people.P/01
Fig. 0 · The catalogue: four systems on two axes, one programme underneath.

The operating pattern

Every system in this reference is built on the same spine. An orchestration layer decomposes the work and enforces sequence and quality. The reasoning is done by teams of specialist agents, each with a narrow professional persona and an output that can be checked rather than trusted. Deterministic rails do what judgment should never do: rules engines where the answer must be exact, grounded retrieval so every claim traces to a source. Where ambiguity or impact crosses a threshold, the work stops and a person decides. Everything is logged: every claim, every check, every decision. The evidence trail is itself a first-class output.

What we publish here is the pattern. The rosters, prompts, models and thresholds, the thousand decisions that make a system dependable, are the work. They stay with us.

The pattern is public knowledge. The craft is not. A working system is separated from a demo by decisions this page deliberately does not contain.
Client recordsDocuments & policiesMarket signalsORCHESTRATIONdecomposes the task · sequences the work · enforces QAdirects & verifiesSpecialist teamnarrow persona · checkable outputSpecialist teamnarrow persona · checkable outputSpecialist teamnarrow persona · checkable outputevery claim checkedDETERMINISTIC RAILSrules engines · grounded retrieval · structured checkswithin toleranceambiguous or high-impactGOVERNED OUTPUTdecision pack · evidence log · audit trailHUMAN GATEa person decidesapproved
Fig. 1 · The spine every BELMOST system shares. Red marks where people stay in charge.

Customer lifecycle intelligence

Raw customer signals in; executive-grade commercial strategy out.

In production at tier-1 European telecom operators

The console

This is what your commercial team opens on Monday: every segment, the opportunity it carries, and a queue of governed moves ready to run.

Fig. 2 · The SIGNAL console, illustrative and in motion. Composed for this reference, not a client screen.

The situation

In mature, competitive markets, the proposition drifts away from what the customer base actually needs. Somebody senior suspects it, but proving it means weeks of analyst time per hypothesis, so the commercial engine tests a handful of ideas a quarter and scales fewer. The opportunity isn't hidden in exotic data. It is sitting in the systems the business already runs.

The system

SIGNAL reads the customer base the way a senior commercial team would, and does it continuously, across the whole base at once. Under a principal orchestrator, three specialist teams work in sequence. A segmentation team resolves customers into micro-segments, each with a behavioural rationale. A benchmark team maps your proposition against competitors, market trends and what customers have started to expect. A commercial team shapes the right move for every opportunity the first two expose: an offer, a campaign, a retention play, a product adjustment. The output is not a report. It is a decision pack: segments, opportunities, recommended moves with value estimates, and an execution queue ready for the commercial team to run.

Around the commercial engine, SIGNAL industrialises the front of the experiment loop: opportunity detection, root-cause diagnosis, experiment design. The business tests more ideas, sooner, on better evidence.

The workflow

01
Hypothesis & ideation
Frames the commercial questions worth asking.
02
Data sourcing & analysis
Assembles and validates the evidence base.
03
Customer segmentation
Resolves the base into micro-segments, each with a behavioural rationale.
04
Opportunity analysis
Maps customer needs against your proposition and the market to isolate unserved demand.
05
Recommendation & feedback
Proposes the move, estimates its value, and learns from every result.

Feedback loops run from every later stage back into the earlier ones: results sharpen segments, segments sharpen hypotheses.

Customer recordsbehaviour · tenure · spendProduct & offer cataloguepricing · bundles · eligibilityPerformance historyoptionalStrategic objectivesoptionalthe evidence basePRINCIPAL ORCHESTRATORSEGMENTATIONa data-science team resolvesthe base into micro-segmentsBENCHMARKmaps your proposition againstcompetitors & market trendsCOMMERCIAL DESIGNshapes the move for everyopportunity the analysis findssegmentsgapsDECISION PACKmicro-segments · opportunity map · recommended moves with value estimates · execution queueEVIDENCE LOGS & DECISIONING AUDIT TRAIL · WRITTEN UNDER EVERY STAGE
Fig. 3 · SIGNAL's mechanism: three specialist teams, one orchestrator, one decision pack.
Initiative creationExperiment launchValue calculationScalingOpportunitydetectionRoot-causediagnosisExperimentplanSIGNAL industrialises the first stage. The rest of the loopstays with your commercial team, fed by the pack.
Fig. 4 · Where SIGNAL sits in the commercial experiment loop.
Deployment
On-premises or private cloud
Scale
Full-base, per-customer scoring
Integration
Standard API integration
Output
Decision pack + execution queue
Oversight
Customer-protection rails

Housing intelligence command centre

City-scale housing intelligence, from a single application to the whole-programme view.

Developed on the real records of a government housing programme

The command centre

The whole programme on one living map. Demand builds across the districts, a stalled case surfaces with its intervention, and a scenario is tested before anyone commits.

Fig. 5 · The DISTRICT command centre, illustrative and in motion. Composed for this reference, not a client screen.

The situation

A housing programme runs on registries that never meet: applications in one system, construction status in another, projects and units in a third. Administrators can see the backlog but not its causes. Families fall out of the pipeline quietly, demand shifts across the city faster than planning cycles, and every strategic question is answered from last year's spreadsheet: where to build, what to build, whom to prioritise.

The system

DISTRICT joins those registries into one geospatial model: every application, plot, project and unit, geocoded, linked and scored. It renders the model as a living, three-dimensional map of the programme. Over that model sit four consoles. Demand forecasts applications across geography and demographics: where need is heading, not just where it has been. Risk radar tracks every application through its journey and surfaces the cases likely to be abandoned, with contributing factors and a recommended intervention. Portfolio puts projects and units in spatial context and matches supply against demand, location by location. Scenario lets planners test policy, budget and growth assumptions against projected waiting times before committing to any of them.

The system reads the citizen lifecycle end to end, from application through allocation, construction, maintenance and engagement, and turns it into the programme's forward view.

The workflow

01
Unify
Registries geocoded and joined into one living model.
02
Forecast
Demand projected across geography and demographics.
03
Watch
Every application tracked; risk scored; interventions proposed.
04
Match
Supply set against demand, project by project, place by place.
05
Simulate
Policy, budget and growth scenarios compared before committing.

The map is the interface. Every number has a place on it, and every place opens into its numbers.

FOUR CONSOLES, ONE MODELDEMANDforecast by placeand demographicRISK RADARat-risk applications,proposed interventionsPORTFOLIOprojects & units in place;supply matched to demandSCENARIOpolicy, budget & growthwhat-ifs, comparedservesUNIFIED GEOSPATIAL MODELevery application, plot, project and unit · geocoded · joined · scored · on a 3D map of the citygeocoded · enriched · risk-scoredApplications& demographicsConstruction status& financialsProject locations& descriptionsUnit specifications& plotsApplyAllocateBuildMaintainEngagethe citizen lifecycle the model reads
Fig. 6 · DISTRICT's strata: registries below, one model in the middle, four consoles above.
Scope
City-scale
Model
Unified geospatial twin
View
3D command centre
Horizon
Multi-year forecasting
Posture
Recommend-only

Procurement intelligence

Structured questioning in; a defensible, policy-aligned tender pack out.

The situation

Sponsors know what they need; they rarely know how to specify it. So requirements arrive half-formed, procurement rewrites them, weeks pass in circulation, and the evaluation criteria are improvised after vendor responses are already in. The result is familiar: packs that vary by author, evaluations that are hard to defend, and cycles measured in months.

The system

TENDER starts where the problem starts: with the sponsor. A guided scoping session asks the structured questions a senior procurement partner would ask, and captures what the business actually needs from a vendor: outcomes, constraints, dependencies, risks. From that captured intent it builds the pack the institution's own policy requires: scope, requirements, evaluation criteria, scoring rubric. It drafts and re-examines until every requirement traces to a stated need and every criterion exists before the first vendor response arrives.

The output is a complete, policy-aligned tender pack: the request for proposals, the statement of work, the evaluation matrix and the scoring sheet. Consistent with each other, and consistent from one procurement to the next.

The workflow

01
Scoping
Structured questioning captures what the sponsor actually needs.
02
Structuring
The pack's template built to the institution's procurement policy.
03
Drafting
Requirements written: complete, consistent, unambiguous.
04
Review & evaluation
Drafts examined and scored until they are defensible.
05
Assembly
The final pack generated: RFP, SOW, matrix, scoring sheet.

The review loop between drafting and evaluation runs until the pack holds, not until a deadline arrives.

Sponsor intent"we need a vendor"GUIDED SCOPINGthe questions a seniorprocurement partner askssponsor sign-offPOLICY STRUCTUREthe template your ownprocurement rules requireDRAFT ⇄ REVIEWrequirements written, examined,scored until defensiblereview looprelease approvalTHE TENDER PACKRequest for proposalsStatement of workEvaluation matrixScoring sheetone pack, consistent within itself and with the last one
Fig. 7 · TENDER's flow. Two red gates: a person signs the scope, a person releases the pack.
Intake
Guided sponsor sessions
Alignment
Your procurement policy
Output
RFP · SOW · matrix · scoring
Review
Human gates at scope & release

Compliance intelligence

A tireless compliance analyst: it reads your strategies, policies, plans and reports, and applies the rulebook consistently, every time.

The situation

Compliance review is slow where it should be thorough, and inconsistent where it must be defensible. Evidence arrives in every format an organisation can produce; two reviewers read the same submission differently; and when a regulator asks why a requirement was marked as met, the answer lives in someone's memory. The rules are not the hard part. The hard part is applying them uniformly across thousands of pages, and no team has the stamina.

The system

LEDGER accepts evidence in any format and resolves it into standardised evidence elements, so review addresses substance rather than formatting. Then it adjudicates in three stages: checkers examine every requirement against its evidence, one by one; a validation council cross-checks their findings for consistency; a supervisor consolidates the verdict. Deterministic rules decide whatever is objective; where ambiguity exceeds tolerance, the requirement is flagged for human review rather than guessed at. The same decision policy runs every time, and logs the path it took.

The output is the system's namesake: a line-by-line audit ledger. Every requirement is marked pass, partial or gap, with citations into the evidence, and the remediation plan is prioritised by risk and dependency. Findings feed back into intake guidance, so recurring failure patterns stop recurring.

The workflow

01
Evidence intake
Mixed formats accepted; readiness and metadata checked at the door.
02
Evidence structuring
Documents resolved into standardised evidence elements: substance over formatting.
03
Evaluation
Every requirement examined against its evidence: supported, missing, or dependent.
04
Scoring & justification
One decision policy, applied identically, with the path logged.
05
Recommendation & feedback
Remediation tied to requirements; recurring patterns improve intake.

Feedback loops connect every stage to the ones before it: what evaluation learns, structuring applies next time.

EVIDENCE · ANY FORMATStrategiesPolicies & plansReportsRecordsSTRUCTURINGstandardisedevidence elementsRULEBOOKregulation · standards · policyelementscriteriaCHECKERSrequirement by requirementfindingsVALIDATION COUNCILcross-checks every findingHUMAN REVIEWwhere ambiguity exceeds toleranceconsistent findingsSUPERVISORconsolidates the verdictverdictAUDIT LEDGERReq 01 · PASS · cites §2.1Req 02 · PARTIAL · cites p.14Req 03 · GAP · evidence missinga citation behind every verdictREMEDIATION PLAN · by risk & dependencyrecurring patterns feed back into intake guidance
Fig. 8 · LEDGER's three-stage adjudication. Ambiguity goes to people, never to chance.
Rulebook
Regulation, standards, or internal policy
Privacy
Fully on-premises available
Intake
All document formats
Review
Three-stage adjudication
Output
Dashboard + audit-ready reports
P/01Programme · Governments & enterprise groups

The AI Office

Standing capability, built in-house

Products prove what agentic AI can do. The AI Office makes the capability yours.

The situation

An institution that buys AI systems one at a time never becomes good at AI. Pilots succeed and stall; every new use case starts from zero; governance arrives after the fact, if at all. What separates institutions that compound from institutions that dabble is not any single system. It is a standing capability: a mandate, a platform, a playbook, and people who own them.

The programme

The AI Office is how BELMOST builds that capability inside a client, up to the scale of an entire government. An office is anchored at the centre of the organisation, with executive sponsorship above it, standards and compliance beside it, and a task force that carries agentic systems into every entity. Around it we assemble the platform it runs on, sovereign where required, the governance that keeps it trusted, the partner ecosystem that extends it, and the in-house talent that makes it permanent.

The programme runs two tracks at once. Quick wins put working systems in front of leadership in months, proving value while the structure is still being built. In parallel, the structural track stands the office up: roles hired, platform scaled, playbook and governance embedded, wave after wave of use cases industrialised. The end state is deliberate: a capability that no longer needs us.

The mandate

01
Accelerate
Pilot agentic workflows and prove them through rapid test-and-learn cycles.
02
Enable
Build the long-term capability and ecosystem required to scale.
03
Scale
Roll proven workflows across every entity of the organisation.

Two tracks, one office. Quick wins deliver now; structural capability makes it permanent.

01 · BUILD & PROVEfirst pilots live;the long-term blueprint set02 · STAND UP & SCALEoffice staffed, platform grown;the next wave deployed03 · EMBED & OPTIMISEgovernance across every entity;earlier waves industrialised04 · EXPANDthe operating model embedded;capability outlives the programmeSOVEREIGN PLATFORM & IN-HOUSE TALENT · SCALED IN STAGES · OWNED BY THE CLIENT
Fig. 9 · The capability assembles like the mark: in stages, from the ground up. The void stays open, and the finished mark belongs to the client.

Every system here has a longer story: names, architecture, outcomes. It is told in person.

Start a conversation