A board meeting reviewing printed performance reports
Assurance & capabilities

The trust layer the next ten years will demand.
Verified. Independent.

Every AI strategy rests on a data foundation. Most organisations build the AI first and discover the foundation was never verified. This is why decisions fail, AI stalls, and regulators find what you couldn't.

Take the diagnostic Book a Briefing →
5 min
Diagnostic to gap analysis
4
Engagement tiers
10
Core sectors
Why organisations engage us

The barrier to AI value is rarely the model. It is the data underneath it.

And the cost of getting it wrong lands on risk, compliance and the bottom line — quietly, quarter after quarter, before anyone realises the foundation was never verified.

01
AI programmes stall because data quality is unknown.
Pilots cannot scale when no one can vouch for the inputs. We make data provably ready for production.
02
Regulators increasingly require explainability and traceability.
Decisions must be defensible on demand. We provide lineage and audit evidence that holds up.
03
Board numbers are trusted because nothing has visibly broken yet.
Country teams manage their own numbers before they reach the centre. We surface what insiders cannot.
04
Data costs keep rising while utilisation stays low.
Duplicated feeds and unused licences drain the budget quietly. We find and remove the waste.
05
Leadership lacks confidence in the data behind decisions.
Executives quietly re-check the numbers, slowing everything down. We give them an independent read they can act on.
Scrabble tiles spelling DATA BREACH
What that costs

Every quarter on unverified data carries a price.

Stalled programmes, regulatory exposure and wasted spend — accumulating silently until something breaks in public.

Risk

Stalled AI programmes

Investment sits idle when pilots cannot be trusted into production.

Risk

Regulatory findings

Decisions you cannot explain or trace become compliance exposure.

Risk

Failed model deployments

Models built on flawed inputs underperform — or fail in production.

Cost

Duplicated data spend

Overlapping feeds and unused licences quietly inflate every budget.

Cost

Decisions on unverified data

Strategic calls made on numbers no one independently checked.

Cost

Hidden from the board

Country teams tidy the numbers before they reach the centre. The board sees what they are allowed to see.

What we actually verify

The unglamorous specifics. The ones that determine whether your AI holds up.

Defensibility is not a philosophy. It is a set of checkable facts about your data, your vendors, your models and your reporting chain — each of them verifiable, none of them self-assessed.

These twelve are not a framework we invented. They are the twelve questions that actually get asked when something goes wrong: by a regulator, by an auditor, by an acquirer's diligence team, or by a board member who has stopped believing the dashboard. Each one is answerable with evidence or it is not.

They are also the twelve dimensions the diagnostic scores — so the assessment and this page describe the same thing, in the same order.

Source code on a dark screen
01

Vendor dependency & liability

What you depend on but do not control — and who carries the risk when it changes.

02

Data quality & reliability

Whether it is complete and current — and whether you would know if it stopped being.

03

Lineage & provenance

Every number, through every transformation, back to a source you can name.

04

Governance & ownership

A named owner who can resolve an incident, not a RACI slide that says one exists.

05

Audit & regulatory exposure

Evidence generated by running, or assembled by hand under deadline.

06

AI / model & eval integrity

Training sets, evaluation sets and live inputs, verified before they shape an output.

07

Data & licence spend

What every pound on data, licences and compute is doing — and what is dormant.

08

Team & operating model

How much capacity goes to manual validation that should be automated.

09

Decision trust

Whether decision-makers act on the number, or quietly re-check it first.

10

Transparency & candour

What people say when their manager is not collecting the answer.

11

Reproducibility

Whether an AI result holds the second time, and whether you could explain any difference.

12

Pilot-to-production rate

Whether pilots reach production, and whether data readiness is what stops them.

ISO 27001 information security
SOC 2 aligned controls
GDPR & data-protection compliant
Independent · vendor-neutral
Vendor & third-party liability

You have outsourced the work. You have not outsourced the liability.

The most common misconception in enterprise AI is that buying from a large vendor transfers the risk. It does not.

01

You are the deployer, not the bystander

Under the EU AI Act the obligation to evidence data governance sits with you. Your vendor's conformity assessment discharges their duty, not yours. DORA requires a register of critical providers with tested exit paths; NIS2 names supply-chain security directly.

02

Licence terms are enforced retroactively

Third-party data frequently prohibits model training, derived-data reuse or redistribution. Exchange and market-data audits routinely produce fee claims backdated years.

03

Vendors change without telling you

A model updates, a schema shifts, a sub-processor is swapped — and outputs move underneath decisions you have already taken. Nobody is monitoring the vendor's side of the boundary.

A hand drawing a block out of a stacked tower
Every block you did not place yourself is a dependency you did not test.
Why it cannot be done in-house

Nobody can objectively audit a vendor they chose.

The person who ran the procurement has sunk cost and a reputation attached to the decision. Internal audit reports to the executive who signed the contract. We sell no software, no systems and no implementation — which makes us the only party in the room with no reason to protect the relationship.

“If your business model depends on the relationship surviving the answer, the answer was never independent.”

No audit practice. No software to license. No transformation agenda. No vendor alliance. One interest: an independent answer you can put in front of a board.

What a vendor health check covers
  • Liability mapping — who actually carries the risk across your top vendors, read from the contracts rather than assumed from the logo.
  • Licence terms against real use — whether your agreements permit the model training, derived data and redistribution you are already doing.
  • Change exposure — which models, feeds and schemas can move under you without notice, and what breaks downstream when they do.
  • Exit readiness — whether you could actually leave a critical provider, in the form DORA now expects you to evidence.
Contract and documentation work. No system access, no security review, no data-sharing agreement.
7%
Of global turnover — the EU AI Act ceiling you carry as deployer
£250k+
Dataset risk de-risked before commitment
10–30%
Typical recoverable share of data and licence spend
4–6 wks
From engagement to a defensible evidence pack
Sectors we serve

Different sectors. The same question: can this data be defended?

What changes by sector is the evidence a regulator asks for — not whether you need it.

An operator monitoring a hydroelectric station
01 · Energy & Utilities

Generation and asset operations

Dispatch and maintenance models trained on telemetry nobody independently validated.

  • Sensor lineage
  • EU AI Act high-risk
  • Outage forecasting
Transmission pylon and power lines at sunset
02 · Grid & Networks

Load, balancing and network reporting

Load and balancing models where an error is measured in supply, not accuracy points.

  • SCADA lineage
  • Load forecast integrity
  • Regulatory reporting
A technician inspecting parts on a production line
03 · Manufacturing & Industrial

Production, quality and maintenance

Predictive maintenance models still trusted long after their sensors drifted.

  • Model drift
  • Quality-control inputs
  • Supply-chain data
An electric vehicle connected to a street charge point
04 · Mobility & Automotive

Fleet, charging and connected vehicles

Charging, telematics and pricing built on streams nobody has reconciled end to end.

  • Telematics lineage
  • Charging network data
  • Pricing model inputs
Also serving
  • Telecommunications
  • Water & waste utilities
  • Transport & logistics
  • Financial services
  • Health & life sciences
  • Public sector

The same lineage, control-testing and reporting-reconciliation work applies in each — under a different regulator. The method does not change by sector. The evidence pack does.

Free diagnostic · Find your path

Ten minutes. An honest read on where your data foundation is weakest.

A short set of questions about how your data is sourced, governed and reported. You get a gap analysis on screen — which areas are solid, which are exposed, and what it typically takes to close each one. It is an educational tool first: most people learn something about their own estate before they ever speak to us.

  • Gap analysis across lineage, quality, controls, model inputs and reporting
  • Benchmarked against what your sector's regulator actually asks for
  • Your shortlist of modules and indicative price ranges, emailed to you
  • No system access, no obligation, nothing installed
Start the diagnostic → See how we engage
How we engage

Start where the friction is lowest. Expand as trust builds.

Four tiers, from a low-commitment assessment that needs no system access through to a standing assurance relationship. The diagnostic tells you which modules apply to your gaps, and we send the shortlist with indicative price ranges to your inbox.

Tier 1 · Land

Advisory & enablement

No system access · days to close
Tier 2 · Expand

Facilitated & independent

Conversations only · weeks to close
Tier 3 · Convert

Embedded assurance

Data & systems access · months to close
Tier 4 · Sustain

Continuous assurance

Annual subscription · the durable layer

The engagement flow

Three ways in, then one logical path. You only move to the next stage when the previous one has earned it — each stage is scoped, priced and useful on its own.

Entry · start here

Free Advisory Call

A 30-minute conversation about where your AI and data exposure sits today. No preparation, no system access.

30 minutes · no cost
and / or

AI Diagnostic

A short structured questionnaire that returns a gap analysis across lineage, quality, controls, model inputs and reporting.

5–10 minutes · free
and / or

AI Benchmarking

Where you sit against your sector and against what your regulator is already asking others to evidence.

Sector comparison
Any one is enough to begin. Taken together they give us the picture without touching a system.
  1. 01

    Executive AI Workshop

    Half day · leadership alignment

    We turn the diagnostic and benchmark findings into one shared leadership view: where the ambition is, where the exposure is, and which of the two is currently louder. Output is an agreed shortlist of what is worth assuring.

  2. 02

    Scope of AI Assurance

    Fixed scope · fixed price

    We define exactly what will be verified, against which standards, with what evidence, over what period, and what it costs. Nothing open-ended, nothing discovered halfway through.

  3. 03

    AI Project Execution

    Weeks, not quarters

    The work itself — lineage tracing, control testing, model-input verification and the remediation that follows — delivered alongside your teams rather than around them.

  4. 04

    AI Assurance

    Evidence · independent opinion

    The evidence pack: what was tested, what holds, what does not, and what remains open — written so a director can defend it in front of a board, an auditor or a regulator without translation.

  5. 05

    AI Continuous Assurance

    Annual subscription · the durable layer

    Models drift, vendors change, schemas move. Controls stay tested and lineage stays verified between cycles, and each cycle re-runs the benchmark — so the path loops rather than ends.

What clients use us to achieve

  • ✓ Accelerate AI from pilot into production
  • ✓ Reduce operational and regulatory risk
  • ✓ Improve confidence in strategic decisions
  • ✓ Optimise and recover data spend
  • ✓ Give the board a verified, unfiltered picture
  • ✓ Make AI investment provably defensible
Engagement
Typical outcome
AI-readiness assessment
Clear go / no-go in 4–6 weeks
Data-spend optimisation
10–30% reduction in duplicated spend
Transparency survey
An unfiltered view of what's actually happening
Governance programme
Faster, cleaner audit response
Continuous assurance
Reduced operational and model risk
Indicative outcomes from comparable engagements. Scoped to your environment on the assessment call.

How to buy

Tier 1

Reserve a date

Discretionary or L&D budget. No procurement, no system access.

Tier 2

Request a fixed-price proposal

We return a scoped quote within 24–48 hours.

Tier 3

Request a scope call

We scope, then issue an SOW + MSA. Procurement and security review expected.

Tier 4

Annual subscription

Request a scope, then an annual contract on net terms.

Which modules apply to you?

Take the diagnostic. We'll email your shortlist with indicative price ranges — no catalogue to wade through.

Start the diagnostic →
An executive signing off on a tablet

Before you trust your AI, build the layer beneath it.

A 30-minute board-level briefing on where your defensible layer is missing — and the fastest, lowest-friction way to begin.

Book a board briefing AI Diagnostic →