Skip to content
GDPVision
A decision paper

The business case for GDPVision

On instrumenting sovereign economic decision-making — prepared for heads of government, Cabinet Secretaries, ministries of finance, and the officials who must satisfy themselves that this is the right class of system before it is procured.

Adam Anderson · OPEN Interactive
Executive summary

The decision in one page.

The decision. Whether to instrument national economic decision-making as a governed, sovereign system of record — or to continue carrying it on arrangements built for statistical record-keeping and, increasingly, on uncontrolled personal use of consumer AI inside the ministries.

Why now. Three pressures now arrive at once: Citizenship by Investment at half of government revenue faces a scheduled end with no priced replacement; the tariff regime is being rewritten in capitals the region does not control; and the hurricane corridor is re-arming after a single storm cost 226 per cent of GDP. The response must be engineered against debt at ninety per cent of GDP, with a quarter of revenue already committed to interest, and using authoritative data that arrives roughly eighteen months after the period it describes.

The central argument. A government's capacity to act on its economy is bounded by the quality of its instrumentation. The region's instrumentation was built to keep a record, not to govern from. It is now being asked to carry the largest economic transition in its modern history, and it cannot.

On the obvious cheap answer. A capable official with a frontier language model can draft a note, summarise a report and compare two options. That is true and worth conceding. But a language model is a component, not a system. It holds no shared corpus, cannot reproduce its own answer, has one actor and no roles, produces text rather than a record, and cannot lawfully receive Cabinet material. The choice is not whether a government uses AI — that is settled, and it is already happening on personal accounts. The choice is whether that use is governed.

What is recommended. A confidential Cabinet briefing, then a time-boxed pilot against one live decision with agreed conversion criteria, then national deployment: one isolated instance per nation, owned outright by the government, with contracted export, escrow and exit.

What is not claimed

No measured outcome is published for GDPVision. No percentage improvement, no time saved, no revenue attributed. The case below argues from mechanism and from the size of the decisions involved, not from results we cannot yet evidence.

01 · What is at stake

Six figures, each carrying its source and its grade.

That is a habit rather than a flourish: the argument of this paper is that provenance survives scrutiny, and a document making that argument should meet the standard it sets.

0%
Bof government revenue, upper band

IMF Article IV consultations, 2022–2024

Citizenship by Investment at the upper band across the five OECS states operating a programme. That money is in hospitals, in schools, and in the reserve a small island reaches for in the weeks after a storm.

0%
Adebt-to-GDP, upper band

IMF WEO, 2024

In high-debt cases interest consumes around a quarter of revenue before Cabinet takes its first discretionary decision of the year. Remove up to half of what remains, and no version of the arithmetic resolves through prudence.

0%
Aof GDP lost in a single night

Government of Dominica Post-Disaster Needs Assessment, 2017

Hurricane Maria. The storm is the most predictable feature of the regional fiscal environment; only its timing and landfall are uncertain.

0
AOECS votes on the Inclusive Framework steering committee

OECD Inclusive Framework governance roster, 2024

Rules are written elsewhere. External decisions arrive as facts, on timetables the region does not set.

0%
Aof tertiary-educated citizens have emigrated

World Bank / OECD DIOC, 2020

Institutional capability is being lost faster than it is being replaced.

0mo
Blag before authoritative sector data reaches Cabinet

ECCB & NSO release cadence review, 2024

A decision taken today rests on a settled picture of the year before last.

Set those figures together and the problem becomes unmistakable. Three sovereign pressures are now running on the same deadline: the revenue model that pays for half the state is ending; the trade rules that shape competitiveness are being written elsewhere; and the hurricane corridor is no longer a seasonal worry — it is a structural fiscal risk. A government facing all three at once has no fiscal slack, a shrinking pool of people who can do the work, and an evidence base that describes the country as it was two years ago. The challenge is not a lack of effort. It is a lack of instrumentation fit for the transition ahead.

02 · The problem

It is instrumentation, not effort.

It would be easy to attribute the gap to capacity or to will. In my experience that is both wrong and unfair.

Officials in these ministries are capable and overworked. The analysis has not been done because the underlying evidence does not exist in a usable form. Ministries hold fragments in incompatible formats. The dependency relationships between sectors — what happens to construction when tourism moves, what happens to public revenue when construction moves — live in the heads of a small number of people, some of them close to retirement.

So when a Cabinet asks what happens if half the revenue goes, the honest answer from the ministry is: give us some months, and we will commission something. The commissioned study arrives after the decision window has closed, describing a situation that has already moved.

The evidence is scattered. No single place a Cabinet can look, and no single number everyone accepts. A great deal of the most expensive time in the country is consumed reconciling figures that were each built on a different basis at a different time, and all of which are defensible.

The evidence is ungraded. A precisely measured figure and a well-intentioned estimate appear in identical type in the same Cabinet paper. Nothing tells a minister which is which — which is why leaders hesitate before saying a number in public, and why a dispute about a national figure can run for days.

And nothing is rehearsed. Policy is committed to and its consequences discovered afterwards. The question that actually matters — what happens if we do this, and what happens if we do the other thing — goes unanswered at the moment it is asked.

None of this is a failure of will. It is a failure of instrumentation, and unlike the external deadline it is entirely within the region's power to fix.

03 · The obvious cheap answer, taken seriously

A language model is a component. It is not a system.

Frontier language models are extraordinary. A competent permanent secretary with a consumer subscription can summarise an Article IV consultation, draft a Cabinet note, compare two policy options, sketch an investment case and produce a press line — in an afternoon, for the price of a monthly fee. Why procure anything?

Most of that is true, and any vendor who denies it should be treated with suspicion. An official who is not using these tools is at a disadvantage to one who is.

But the objection contains a category error worth naming precisely. A language model is a component. It is not a system. GDPVision itself uses language models — several, behind a gateway, with a documented fallback order and provenance recorded on every fact they return. The question was never model versus system. It is ungoverned model use against governed model use.

What a chat session cannot do — not “does less well”, but cannot, by construction

  1. 01
    It cannot hold what your government knows

    It has a context window you must refill from scratch every time, which vanishes when the tab closes. Nothing accumulates, nothing is shared, and fifteen officials each holding a private conversation produce fifteen unreconciled views faster than before.

  2. 02
    It cannot lawfully receive your Cabinet material

    To get a useful answer an official must paste in the thing that makes it useful — the draft budget, the term sheet, the memorandum — into a service hosted elsewhere, under terms no ministry has reviewed, with retention the government cannot audit.

  3. 03
    It cannot reproduce its own answer

    Ask the same fiscal question twice and you get two answers. GDPVision separates the two things a chat session conflates: the model proposes, and a deterministic engine computes — pure, versioned, containing no randomness, with the engine version pinned to every scenario artefact. A projection made in March re-runs identically in September.

  4. 04
    It has one actor

    Government work is irreducibly multi-actor and differently permissioned: a steward maintains a series, a minister reads a portfolio, a principal decides, a secretary records. A session has no concept of role, ministry or country access. GDPVision enforces access at the database layer through row-level security, not in the interface where a hidden button is not access control.

  5. 05
    It produces text, not a record

    A session's output is a message. The instrument's output is a row, in a system of record every chamber reads. A scenario becomes an artefact with pinned assumptions; a Cabinet decision becomes a commitment with a named owner; the commitment is scored against the mandate at quarter end. Work moves between chambers rather than being retyped — and that carrying by hand is where government work is actually lost.

  6. 06
    It waits to be asked

    Coverage builds overnight; a source goes unreachable; a research stage fails at three in the morning. The instrument runs continuously — press discovery and clustering across entity feeds, watchlists refreshed on schedule, source health retried, stale locks reclaimed on an eight-minute heartbeat, failed stages redriven with escalating fallback.

  7. 07
    And it cannot be governed

    It has no doctrine, no approval gate, no register discipline, and no record of what it advised. GDPVision holds a doctrine enforced in code — the chamber drafts, principals decide, nothing releases autonomously — with no path from a detected signal to a public statement that does not pass through a named, accountable human being.

04 · The alternative is not free

Because it is already running.

This is the part most often missing from the comparison, and it is the one that should concern a Cabinet Secretary most.

Personal use of consumer AI is not a hypothetical future state. It is the current state in ministries across the region, undertaken by conscientious people trying to do difficult work with inadequate tools. The government therefore already carries four liabilities, none of which appears in any budget line.

Disclosure exposure

Sovereign material outside the jurisdiction with retention nobody can audit. This is discoverable, and it will eventually be discovered by someone unfriendly.

Unattributable advice

When a figure in a Cabinet paper proves wrong, there is no record of where it came from. “The AI said so” is not a defence in Parliament, and the official who relied on it carries the exposure personally.

Fragmented positions

The reconciliation problem accelerated rather than solved.

Key-person concentration

The capability sits with whichever official is good at prompting — in a region where seven in ten tertiary-educated citizens have already emigrated.

So the choice is not “spend money or spend nothing”. It is: continue carrying an uncontrolled liability at no visible cost, or convert it into a governed capability at a visible one. Finance ministries make that trade in every other domain. It is the same argument as moving from informal borrowing to a documented facility.

05 · The decision

What class of system is this?

Every government runs a small number of tier-one systems: treasury and financial management, revenue collection, the national identity register, settlement infrastructure.

Nobody evaluates a treasury system against a spreadsheet. Not because spreadsheets are bad at arithmetic, but because the two objects are in different classes — and the class is determined by consequence, not capability.

System of record others read as authoritative
Chat subscription
No
GDPVision
Yes
Multi-actor, permissioned, enforced
Chat subscription
No
GDPVision
Yes — row-level, database-enforced
Outputs auditable after the fact
Chat subscription
No
GDPVision
Yes — immutable audit log
Outputs reproducible
Chat subscription
No
GDPVision
Yes — deterministic engine, pinned version
Holds sovereign data lawfully
Chat subscription
No
GDPVision
Yes — isolated instance, chosen region
Survives a change of administration
Chat subscription
No
GDPVision
Yes — the data is the government's
Named supplier accountable
Chat subscription
No
GDPVision
Yes
Defined exit, export and escrow
Chat subscription
No
GDPVision
Yes — contracted

The relevant question is therefore not whether GDPVision beats a subscription. It is whether the evidence base beneath half of government revenue is a tier-one system or not. A government answering “not” has made a decision, whether or not it intended to.

06 · Options appraisal

Three paths, and what the government owns after three years.

Path A

Subscriptions and capable officials

SeatsSessionsProse

Recurring per-seat cost. At the end of three years the government owns nothing: no corpus, no decision record, no scored mandate, nothing that transfers to a successor. Stop paying and the capability stops that afternoon, leaving no residue, because the work lived in individual sessions that were never institutional. It is genuinely cheap, and it is cheap precisely because it accumulates nothing. You are renting cognition; you are not buying an asset.

Path B

Build it internally

CodeStaffingLiability

Significant capital and permanent staffing. What would be commissioned is not a chat interface over a document store — that is a quarter's work. It is a governed schema with grants, row-level security and policies; scores of server-function modules of domain logic; a twelve-sector ontology with a decade of history; a deterministic projection engine; a corpus gateway with deduplication on normalised keys; a twenty-stage country onboarding pipeline including a capital-flows stage that refuses to commit unless the draft balances within ten per cent; an immutable audit log; and cross-chamber promotion paths. Then permanent maintenance, in a labour market already losing skilled people. And the harder part is not the code — it is the judgement encoded in it, which a from-scratch build gets wrong on the first attempt.

Path C

Procure the instrument

CorpusRecordMandate

A deployment the government owns outright, contractually and technically, with full export and verified deletion on termination. At the end of three years it holds a structured corpus of its own economy, a decision record spanning two or three Cabinets, a mandate scored quarter by quarter, and a modelled dependency map — all of which transfer to the next administration.

For a government carrying debt at ninety per cent of GDP, the distinction between recurring expenditure that leaves a residue and recurring expenditure that does not is not philosophical. It is how the estimates are argued.

07 · What the instrument actually is

A live National Ledger beneath eight chambers, with a voice-first Counsel above them.

Chamber 01

The National Ledger

A twelve-sector ontology with a decade of history, a confidence grade on every series, exposure indices drillable to source, and four-layer sector dossiers. The single source of truth every other chamber reads from.

Chamber 02

Portfolio Workspaces

Every minister's contribution to GDP as a standing figure, their dependency web, and their levers ranked by effect with the cost of each attached.

Chamber 03

The Scenario Engine

Rehearsal before commitment. Ripple propagation through the inter-sector web, goal-seek that runs the decision backwards from a target, sensitivity views, and a compensation ledger showing what each gain costs elsewhere.

Chamber 04

The FDI Transition Studio

The Gap priced year by year under the actual glide-path; an investment package builder with capital-to-GDP conversion and honest time-to-impact lags; readiness scored across legal framework, land, workforce, incentives and institutional capacity; and the book sequenced across years.

Chamber 05

The Narrative Chamber

Monitoring, response and syndication in one place. Entity feeds and watchlists refreshed on schedule, a signal desk ordered by economic consequence, structured strategy, channel drafts, an explicit approval workflow, scheduled publication, and a retained archive of what was issued.

Chamber 06

The Cabinet Room

Session Mode running the meeting itself, decisions recorded live with named owners, a commitments cockpit visible between sessions, and a National Scorecard that moves against constant indicators.

Chamber 07

Persona Lab

Rehearsing how a policy, incentive or message lands, privately, before announcement. Explicitly a rehearsal instrument and not a substitute for polling.

Chamber 08

The Mandate Compact

The manifesto decomposed into pillars, pledges and ministry-owned deliverables; quarterly scorecards and a PM Report Card; a signed, versioned compact whose every revision is diffable; and a transformational plan that hands directly to the Narrative Chamber, so what a government announces is the same object as what it decided.

Underneath all of it, one sovereign corpus: public evidence and private Cabinet material held apart and read together, deduplicated, chunked, embedded, with visibility and ownership on every row.

08 · What this is worth to a sovereign economy

Mechanisms, not measured results.

These are mechanisms, not measured results. We publish no outcome figures and will not until one is cleared.

A transition that is priced rather than described. The Gap year by year, a costed replacement book with lags made explicit, and readiness scored so a government can say precisely why a package is not yet investable. Investors do not walk away because a country is small; they walk away because nobody could tell them whether the land title would clear.

A channel, not only a document. OPEN Interactive has convened the Caribbean Investment Summit since 2009 — the room where the packages this instrument produces meet capital. A strategy document dies in a drawer; a package with a channel does not.

Senior time returned. The volume of Cabinet and permanent-secretary time lost to reconciling incompatible numbers is among the largest hidden costs in small-state government, and is almost never measured because nobody has been asked to account for it.

Fiscal leakage closed. Where a quarter of revenue is committed to interest before Cabinet decides anything, a decision taken and then quietly forgotten is money the country did not have to lose. A commitments record makes that visible while it is still recoverable.

Shocks priced before they arrive. Recovery financing negotiated after a storm is negotiated from the weakest position a country ever occupies. Negotiated against a modelled position, it is a different conversation with the same lenders.

A defensible national position in hours rather than days. Investor confidence, currency sentiment and the tone of the next credit review move on narrative before they move on fundamentals.

Institutional memory retained. Against seven-in-ten skilled emigration, a corpus does not resign.

And an asset that survives an election. A decision record is politically neutral by construction: it states what was decided, by whom, and what happened — as useful to an incoming government as to the one that built it.

09 · The five approvals

Sovereign procurement has no single buyer.

Five people must each be satisfied, by different things.

The Principal. Buying defensibility and control of their own record: a number they can state in Parliament and immediately source, and a scorecard against their own manifesto published before a journalist builds one.

The Gatekeeper — Chief of Staff or Cabinet Secretary. Buying time and the absence of embarrassment: a State of the Nation brief generated rather than assembled over days, and a commitments record so the principal is never surprised eleven months later.

The Technical Validator — Ministry of Finance or central bank. Buying method they can interrogate: documented confidence grading, reproducible projections, sensitivity views, provenance to source. This is the person a chat session loses fastest, because they will ask the same question twice and notice the answers differ.

Procurement. Buying a lawful route to award, a named accountable supplier, written exit terms, escrow, and a contract that survives a change of government.

The Sovereignty Gate — national security adviser or data protection commissioner. Buying an answer they can give in public: one isolated deployment per nation; separate database, storage and encryption keys; no cross-instance queries anywhere in the architecture — not disabled, absent; hosting region chosen with the government; no third-party trackers inside the instance.

Note the pattern. The validator and the sovereignty gate do not reject a subscription for being less capable. They reject it for being the wrong class of system — an objection no better model can answer.

10 · Who built it, and why that matters

The hard parts are not the screens.

A reasonable technologist will ask why a competent team could not assemble this in two quarters.

Some of it they could. The hard parts are not the screens. They are the ontology that lets figures from different ministries occupy one picture; the confidence-grading method; the deduplication contract that keeps a corpus trustworthy in year three; the reconciliation gate that knows when a capital-flow draft is not yet commit-worthy; the decision that scenarios must be deterministic and the model may only propose; and the judgement that a minister's realistic unit of use is ninety seconds on a phone between engagements, not an afternoon at a dashboard.

Those do not come from software experience. They come from having been in the room.

OPEN Interactive has convened the Caribbean Investment Summit since 2009, delivered digital government infrastructure at national scale under confidential engagement with the Office of the Prime Minister of St. Kitts & Nevis, and maintained head-of-government relationships across the OECS for seventeen years. SEDE — the Saint Lucia prototype, a working sovereign decision engine with a live macro model, voice console, dossier corpus and ingest pipeline — is the interaction-proven core that GDPVision v1 absorbs.

This is not a global product adapted downward. It is an instrument designed against the exposures small island states actually carry.

11 · Risks, in both directions

Stated plainly.

Risks of proceeding

Adoption is the real one: an instrument ministers do not open is worth nothing, which is why the console is three tabs built for ninety seconds on a phone. Second, scope — a single-chamber entry that never instruments the Ledger underneath has nowhere to expand to. Third, ordinary supplier risk, answered with escrow, export and defined exit terms agreed early rather than late.

Risks of not proceeding

The shadow-AI liability continues, uncosted and undocumented. Institutional memory continues to leave with departing officials. The transition continues to be planned against evidence describing the year before last. And the first serious public dispute about a national figure is met with days of ministries telephoning one another while the doubt hardens.

12 · Recommended path

Briefing, pilot, deployment.

  1. 01

    Stage 1 — Confidential Cabinet briefing

    Sixty minutes, in person or over secure video, under NDA on request, nothing recorded. Prepared against the nation's own public data, so what is seen is that economy rather than a generic demonstration.

  2. 02

    Stage 2 — Time-boxed pilot

    Against one live decision, with the Ledger instrumented underneath from day one and agreed criteria for what converts it. A pilot without conversion criteria becomes a free consulting engagement that ends when the sponsor changes job; we would rather name that in advance.

  3. 03

    Stage 3 — National deployment

    One isolated instance per nation, owned outright by the government, with contracted export, escrow and exit, and a hosting region chosen with the government.

13 · The test to run before deciding

Put these to any AI tool a government is considering — including ours.

  1. 01Show me the source document behind this figure, and its confidence grade.
  2. 02Run this projection again and give me the identical numbers.
  3. 03Tell me who in this government has read this analysis, and when.
  4. 04Show me the decision this analysis led to, its named owner, and where it stands.
  5. 05Where is this data held, under whose keys, and in which jurisdiction?
  6. 06If we stop paying you, what do we keep — and in what format?
  7. 07Who is accountable, by name, when this is wrong?

Recommendation. Proceed to a confidential Cabinet briefing, and require the seven-question test above of us and of every alternative considered.

Sources and confidence grades

Every figure, with the grade we assign it.

Every figure in this paper is published with the grade GDPVision assigns it. A grade is not a claim of accuracy — it is a statement of how much weight a number will bear. We publish ours because we ask governments to do the same.

  • B
    CBI share of government revenue, upper band — 50%
    IMF Article IV consultations, 2022–2024.
  • A
    Debt-to-GDP, upper band — 90%
    IMF World Economic Outlook, 2024.
  • B
    Interest as a share of revenue, high-debt cases — c. 25%
    IMF Article IV consultations.
  • A
    Hurricane Maria cost to Dominica — 226% of GDP
    Government of Dominica Post-Disaster Needs Assessment, 2017.
  • B
    Authoritative sector data lag to Cabinet — c. 18 months
    ECCB & national statistical office release cadence review, 2024.
  • A
    Tertiary-educated emigration rate, upper-band Caribbean states — 70%
    World Bank / OECD DIOC skilled-migration database, 2020.
  • A
    OECS votes on the OECD Inclusive Framework steering committee — 0
    OECD Inclusive Framework governance roster, 2024.
Cabinet briefing

Request a confidential briefing.

A short, dignified enquiry from a member of a sitting government or their designated advisor. OPEN Interactive responds within one working day.

Request a Cabinet briefingRead the writing