Capabilities
What we can address.
Eight capability groups and fifty-plus patterns, drawn from work these platforms already do. This is the layer where AI programmes succeed or stall — read it as range rather than a menu.
How a capability gets built
Read in place, govern it, act within limits.
Every pattern below is built the same way. The order matters: most programmes that stall attempted the third step before the first was finished.
Capability group 01
Foundation and governance
One set of numbers everyone can defend.
- One governed set of numbers
- Definitions agreed once and enforced, so finance, operations and commercial present the same figure with the same meaning.
- Data quality, applied progressively
- Rules in business language applied as data arrives, so records become usable without a cleansing project first.
- Lineage and defensible evidence
- Every figure traceable to the source system, object, row and version, with a full audit of who queried what and when.
- Automated cataloguing
- The meaning of tables and fields proposed automatically across a large estate, then approved by a person.
- Legacy and unreachable systems
- Any external program or legacy application queried as though it were a database table.
- Entity resolution
- Records referring to the same customer, supplier, part or asset resolved into one identity, with the matching rule explicit and reviewable.
- Migration assurance
- Legacy and target read side by side, so reconciliation and parallel running happen on live data rather than sampled extracts.
- A governed layer for existing BI
- The tools you already license connect over standard interfaces and finally have something complete to draw on.
Where it lands: Every sector. This is where engagements typically begin.
Capability group 02
Operations and production
Where margin actually leaks.
- Yield, scrap and material recovery
- Consumption against output reconciled continuously by line, shift, product and cause — not at period end when the material is gone.
- Asset and equipment performance
- Availability, performance and quality per asset with reasons attached, and maintenance spend set against what the asset actually delivered.
- Condition-based maintenance
- Equipment parameters and failure history used to move critical assets off calendar-based intervention.
- Promise integrity
- Order through production, stock and despatch in one live view, with a commitment at risk surfaced while there is still time to act.
- Multi-site comparison
- Sites compared on identical definitions of output, quality, utilisation and cost, without anyone building it by hand.
- Process parameters correlated with outcome
- Machine settings and cycle data correlated against downstream quality, turning process records into prediction.
- Tooling and consumable life
- Actual life achieved against expected, with the quality consequence of wear visible before it produces rejections.
- Capacity and constraint visibility
- Where capacity actually sits, what consumes it, and where the constraint moves as product mix changes.
Where it lands: Manufacturing, energy, logistics, infrastructure, process industries.
Capability group 03
Quality, traceability and compliance
Evidence, not recollection.
- Genealogy and traceability
- The full history of a unit, batch or consignment in one query, and the reverse: which other units share a suspect input.
- Long-horizon retention and retrieval
- Records held for statutory periods in tiered open storage with every tier directly queryable. No restore step.
- Containment and recall scoping
- When a cause is identified, the affected population determined precisely, so remediation covers what is genuinely involved.
- Quality and test analytics
- Inspection and validation results held against the part, batch and revision, so results become trend and prediction.
- Certification and audit evidence
- Documentation assembled from a governed layer instead of reconstructed ahead of each audit.
- Regulatory and statutory submission
- Recurring submissions produced from governed data with provenance, rather than rebuilt each period.
- Responsible sourcing and origin
- Origin, certification and chain-of-custody evidence joined to what was actually received and shipped.
- Field failure feedback into engineering
- Warranty returns traced back to input lot, process conditions and design revision.
Where it lands: Manufacturing, food and pharma, aerospace, automotive, public sector.
Capability group 04
Commercial and revenue
Which business is actually worth having.
- True cost to serve
- Fully loaded cost per unit, order, route or customer — inputs, losses, time, rework and overhead brought together instead of allocated by average.
- Counterparty 360
- One view per customer, supplier or counterparty spanning commercial, operational, quality and financial history.
- Demand and conversion
- Enquiry through quotation to order by channel, market and product, so effort concentrates where it converts.
- Revenue assurance
- Declared or reported revenue reconciled against independent operational evidence, with variance surfaced automatically.
- Position, exposure and margin
- Physical and contractual positions, hedges, movements and finance joined into one live picture.
- Quotation and pricing support
- The cost, capacity and historical performance data needed to price accurately, available when the quotation is being prepared.
- Channel and secondary sales
- What distributors are actually selling through, not merely what was shipped to them.
- Portfolio and mix profitability
- Contribution by product, segment and customer, including where new lines are displacing legacy ones.
Where it lands: Commodity trading, manufacturing, distribution, professional services.
Capability group 05
Finance and control
Close faster, on numbers you can defend.
- Multi-entity consolidation
- Group, entity, joint-venture and partner views produced from the same governed numbers across currencies and jurisdictions.
- Period close acceleration
- Reconciliation across entities and ledgers produced from a governed layer with lineage, instead of spreadsheets passed between people.
- Working capital release
- Inventory, work in progress and receivables by age, value and owner — including what sits with partners or in transit.
- Inventory ageing and obsolescence
- What is held, how long, what it is worth, and what is tied to declining demand — before it becomes a provision.
- Procurement and supplier spend
- Total spend with each supplier and category across every entity and site, so negotiation happens on the whole picture.
- Capital programme control
- Approved against committed against spent across concurrent projects, visible while a project is live instead of after it closes.
- Provisioning on evidence
- Warranty, credit and obsolescence provisions informed by actual history rather than a standard percentage.
- Cost allocation and recharge
- Shared cost attributed to whatever actually consumed it, with the basis transparent and auditable.
Where it lands: Every sector with a group structure or a demanding close.
Capability group 06
Intelligence and AI
AI that answers from your records.
- Natural-language self-service
- Business users ask in plain language against governed data and receive traceable answers, instead of every question becoming a ticket.
- Proactive insight agents
- Continuous monitoring that raises the exception instead of waiting to be asked — a metric drifting, a commitment at risk.
- Grounded retrieval for AI and agents
- An ontology over live enterprise data, so generative AI and agents answer from governed records inside your environment.
- Document and correspondence intelligence
- Contracts, specifications and correspondence made queryable alongside structured data.
- Forecasting on unified data
- Demand, material and cost forecasting built on the joined picture rather than one system's partial view.
- Synthetic and masked datasets
- Development, test and third-party work supplied with realistic but de-identified data.
- Governed external data sharing
- Customers, partners and regulators given a live view of their own data and nothing else, controlled at row and column level.
- Workforce, skills and cost
- Qualification and certification status against operational need, and labour cost set against the output it produced.
Where it lands: Every sector. Usually the second engagement, once the foundation exists.
Capability group 07
Governed agentic operations
Agents that do real work, inside real limits.
- Threshold-bound autonomous action
- Routine, reversible work done automatically within limits a named person set, and fully recorded.
- Approval routing by value, risk or policy
- The agent prepares the action and its evidence; the right person approves before anything happens.
- Exception-only human queues
- People see what genuinely needs judgement instead of reviewing everything to find the few that matter.
- Case triage and preparation
- Read, classified, routed and drafted by an agent; decided by a person.
- Reconciliation at volume
- The routine majority handled continuously, with genuine exceptions surfaced rather than sampled.
- One control plane over a mixed agent estate
- Inventory, policy, unified audit and cost control across agents from several vendors, not just ours.
- Acting on systems with no interface
- Sandboxed execution and governed screen automation as a bridge to systems that were never given an API.
- Sealed decision record
- Every action signed, with the evidence it relied on, the gate applied, who approved, and what was done.
Where it lands: Regulated operations, shared services, public administration, financial services.
Capability group 08
Sensing, vision and field operations
Seeing what is happening, where it is happening.
- Detection with a ground coordinate
- A detector flags a candidate, a classifier confirms it, a model explains it, and elevation ray-casting places it on your own map.
- Cross-validation against independent sources
- Findings compared against independent data before anyone is asked to act.
- Asset defect classification
- Condition and defect assessment from imagery, against the asset register.
- Safety compliance on existing cameras
- Using the camera estate already installed instead of a new deployment.
- Blind-spot and link-health reporting
- Cameras and links that go dark are reported rather than silently missed.
- Offline-first field capture
- Capture in the field with interrupted synchronisation resuming where it stopped.
- Voice intake and status answers
- Structured capture over telephony where connectivity, literacy or access is uneven.
- Digital twins and scenario comparison
- A twin over governed state, with deviation alerting and advisory optimisation inside a compliance boundary.
Where it lands: Infrastructure, energy, environment, logistics, public safety.
In practice
The same data, made answerable.
The challenges underneath
What actually blocks enterprise AI.
No system owns the question
ERP is correct. The shop floor system is correct. Lab, CRM, maintenance, HR, the spreadsheets — each correct in isolation. But the questions that matter span all of them, and none of them owns the answer. So the work goes to people: export, reconcile, cross-reference, assemble. Done periodically, because doing it continuously would consume the team. The answer arrives after the decision needed it.
What we do about it
A governed layer that reads every system in place and joins them, without migrating anything. The question is asked once and answered directly, with lineage back to the source record. Each function stops keeping its own quietly different copy, so the management review debates the decision rather than the data.
In plain terms: Ask the question the business actually has, not the one one system can answer.
Confident answers nobody can check
AI produces fluent, plausible prose whether or not it is correct. The failure is not an error message — it is a good answer that happens to be wrong. Annoying in a chat window. Serious in a claim decision, a clinical summary or a regulatory submission.
What we do about it
Grounded retrieval over an ontology built on live data instead of a vector copy. Relationships between entities survive, joins stay correct, and every answer traces back to records. Where the material does not support an answer, the system declines and records why.
In plain terms: It shows its working, or it says it can't.
Agents nobody can hold responsible
Agent programmes stall on governance, not capability. Where would data leave the boundary. What happens when an automated step affects a person. Who can reconstruct afterwards what the system did and why. Without answers, nothing gets approved.
What we do about it
Every action class carries a gate — act automatically, prepare for approval, or advise and let a person decide — enforced on the decision path rather than at the front door. Each action is signed, and can be reconstructed as of any past date, including which model ran with which prompt.
In plain terms: Agents work like employees: a role, an owner, a budget, a record.
Your context living in someone else's system
Sovereignty, residency and classification requirements increasingly rule out architectures that move data into a hosted service. And beyond compliance: the accumulated context of an organisation is among the most valuable things it owns, and most AI adoption quietly transfers it into a platform that cannot be left.
What we do about it
One codebase, five deployment topologies — on premises, private cloud, public cloud, air-gapped, or federated per jurisdiction. Standard Kubernetes instead of a provider-specific managed service. Inference on local models inside the boundary. Open columnar formats throughout, so the data stays portable.
In plain terms: Own your context. Rent your intelligence.
Programmes that cost more than they return
Warehouse programmes require migration and extended remodelling before any value is visible. Point tools reach only what they connect to. Meanwhile inference bills grow faster than benefit and nobody can attribute the spend.
What we do about it
Value delivered as a repeatable cycle in which each investment is tied to one defined question and justified on its own merits. The foundation is established once; every subsequent use case reuses the same connections, definitions and governance, so the cost curve declines. Model routing sends each request to the smallest model that will do the job, with budgets that decline rather than report.
In plain terms: Value early, investment following evidence, and a declining cost per use case.
No usable answer to 'what did it actually do?'
Six months later a regulator, auditor, customer or court asks what happened on a particular day, on whose authority, and on what basis. In most organisations the honest answer is a search through application logs never designed to answer that question.
What we do about it
A sealed decision record per action: what was asked, the evidence relied on, the policy gate applied, who approved, what was done — each entry carrying the hash of the one before it, so deletion or reordering is detectable. Reports produced in the reviewer's own template.
In plain terms: A record you hand to an auditor, not a log you have to explain.
One catalogue, ready to read
Service operations: a capability with its own page.
The deepest single application of these capability groups is service management: named outcomes across the twelve domains of ITSM, each with its gate and its status, running on the platform you already own.
Which of these is closest to your problem?
Name the question you would most want answered and we will tell you what it would take.