Directed intelligence

In build · 2026

For the institutions that run economies

Nothing above this line leaves.

The institution’s data, decisions and memory

Directed at decisions of this kind

  • Which of forty candidate projects is funded this cycle.
  • Where an intervention stalled, and why.
  • How a ten-year capital budget should be sequenced.
  • Who receives credit, and on what terms, in markets where conventional credit data is thin.

Routed and swapped · nothing of the institution’s is held here

OLU is a directed intelligence platform for finance and planning ministries, development finance institutions, sovereign funds and continental enterprises. The computation moves to the institution’s data. The data does not move to the computation.

  • Frontier models Routed across the best available, swapped as the field moves
  • Cloud and compute Sited where the institution’s regulation requires
  • Enterprise systems Read in place, inside the institution’s perimeter

The position

The seat
between
the parties

Almost every intelligence product on the market asks an institution to move its data onto the vendor’s platform. That is presented as a technical arrangement. It is a jurisdictional one. The data leaves the country and comes to rest under another state’s law — reachable by that state’s courts, subject to its export rules, and capable of being switched off by a decision nobody in the institution was consulted about. That is not hypothetical. Access to frontier systems has already been suspended and restored by orders issued in other capitals, on timetables no customer was asked about.

Nobody signs a paper handing over control of national information. They sign a service agreement. The loss is invisible on the day it happens and obvious the first time it matters — which is also the point at which it cannot be reversed.

The value compounds in the wrong place too. Every decision an institution makes on that platform teaches the platform, and the learning is sold back to the institution that produced it. Raw material out, finished product in, margin captured elsewhere. There is a name for that pattern and it is not a new one: data colonialism. Infrastructure that can be revoked, or that carries value away to be realised somewhere else, is not infrastructure. It is dependency by another name.

OLU inverts the requirement. The computation moves to the institution’s data; the data does not move to the computation. Extraction is closed off by how the platform is built rather than by a clause in a contract, which is what allows a procurement conversation to continue past the legal review.

What that protects is not only the records. It is the thinking layer above them — the questions an institution decides are worth asking, the reasoning it accumulates over years, and the value of turning its own information into its own decisions. Data sovereignty keeps the records in the country. Cognitive data sovereignty keeps the judgement.

Not a question of honesty. A question of position.

Nobody who gains from the answer should help you find it.

The decisions OLU is built for have parties on either side of them. Which of forty projects gets funded. Which supplier wins the work. Which programme moves first and which waits a year. Somebody gains from every one of those answers, and everyone in the room knows who.

OLU is not among them. It does not sell the technology an institution runs on. It does not own any of the projects being weighed. It does not compete with the institutions it works for. There is no answer it gains from, and that is the only reason its work on those answers is worth anything.

The seat, and where it comes from

In any deal that matters there is a seat that belongs to neither side. In capital markets it has a name: the trusted independent intermediary. It is the party paid to have no side, and it stops being useful the day it takes one. OLU holds that seat for decisions made with machine intelligence.

It is the seat Blessing Rugara has held for two decades in African corporate finance. Same institutions, bigger decisions — and a new kind of company that wants both to sell an institution its intelligence and to advise it on what to buy.

So the claim is not that OLU can be trusted. Trust has to be earned, it lasts only as long as the good behaviour does, and an institution cannot check it before it needs it. The claim is that OLU is built so it could not betray an institution even if it wanted to. There is no way to take the data out, because no such path was ever built. There is nothing of its own to sell, so there is nothing to push.

Why institutions are receptive now

Data sovereignty has moved out of position papers and into procurement documents across the continent, which designs extraction-dependent products out of tenders before they are ever demonstrated. Institutions here are also building decision infrastructure without forty years of legacy systems to migrate off first, which compresses implementation and removes the internal politics of defending a sunk investment.

The problem is usually already on the institution’s desk. What has been missing is a counterparty able to address it without asking the institution to give up control of its own information.

Who this is for

The institutions
OLU is built for

It is directed at the decisions those institutions find hardest — the ones that are large, consequential and expensive to reverse.

  • Government and public administration

    Ministries of finance, planning, economic development, health and digital affairs. Premier’s and governor’s offices. Statistics agencies, revenue authorities, border and customs administrations, roads agencies.

  • Development and sovereign finance

    Development finance institutions, sovereign wealth funds, national infrastructure funds, pension administrators and multilateral development banks operating on the continent.

  • Credit and financial services

    Banks and non-bank lenders operating where conventional credit data is incomplete, together with credit guarantee schemes and public lending programmes.

  • Large enterprise

    Groups operating across several African jurisdictions — infrastructure, construction, resources, logistics, agriculture, security — where the same decision must be made consistently under very different regulation.

Where the platform earns its place

  • An institution has been given a mandate it does not yet have the tools to execute.
  • It is under pressure to demonstrate how a decision was reached, rather than merely what was decided.
  • It holds substantial data that regulation or policy prevents from moving offshore.

Where OLU declines the engagement

Where what is wanted is a software product to install. Where the decision in question sits with a technology function rather than with the person accountable for the outcome.

Both produce implementations that no one owns.

The architecture

The Directive

The governing principle is cognitive data sovereignty: the institution owns its data and owns the thinking layer built on top of it. OLU builds no frontier model, routing instead across the best available and swapping them as the field moves — so an institution adopts advanced capability without single-provider dependency, and without being read as having chosen a geopolitical side.

Exhibit 1 The platform and the layers around it

Institutions served

Data and decisions stay above the line

  • Government and public administration
  • Development and sovereign finance
  • Credit and financial services
  • Large enterprise
OLU The directed intelligence platform — The Directive
  • Context Intelligence
  • Institutional Memory
  • Multi-Jurisdiction Routing
  • Sovereignty by Architecture

Built and owned by OLU

  • Frontier models
  • Cloud and compute
  • Enterprise systems

Input layer · routed, and swapped as the field moves

A supplier of the layer below cannot sit here. The seat exists to move freely across that layer and to move away from any part of it as the field changes, and no supplier can do that to its own product.

Exhibit 2 What crosses the boundary, and in which direction

Above the line · the institution’s data, decisions and memory

The boundary

Set by the institution’s regulation, not by contract

Crosses down

  • The question, expressed as a task
  • Method, coefficients and assumptions
  • Non-identifying parameters

A question is not a record. What descends is what the institution has approved for the purpose, and it is logged.

Crosses up

  • The result, and its confidence
  • The reasoning that produced it
  • Method improvements, carrying no content

Everything returned is attributable and versioned, so the decision can be defended after the fact.

Never crosses

  • Source records and registers
  • Institutional identifiers
  • Deliberative memory

There is no export path to close, because none was built. This is the property counterparties test hardest.

Below the line · routed, and held by no one

Frontier modelsCloud and computeEnterprise systems

The distinction that carries the claim is between a question and a record. The first descends under the institution’s own authorisation and is logged; the second does not descend at all.

  • Context Intelligence

    The situation, the jurisdiction, the history

    The platform carries the specific situation, jurisdiction and history behind a decision, rather than treating each request as though it had arrived from nowhere.

  • Institutional Memory

    Prior decisions and the reasoning behind them

    Decisions persist, and so does the reasoning that produced them. The platform is worth materially more to an institution in year three than in month one, and that accrual belongs to the institution.

  • Multi-Jurisdiction Routing

    Several legal regimes at once, as the default

    It operates across several legal and regulatory environments simultaneously. On this continent that is the ordinary condition of cross-border activity, not an exception to be handled later.

  • Sovereignty by Architecture

    Extraction closed off by construction

    There is no export path to close, because none was built. This is the property counterparties test hardest, and the only one that cannot be retrofitted.

Nine verticals

Where the
work stands

OLU advances through direct institutional engagement rather than broad commercial launch, which is the correct sequence for this buyer and produces references that no marketing spend can purchase. Three verticals are in active engagement. The other six open when an institution inside them asks.

  • GovernmentOLU Ministries, premier’s and governor’s offices, national agencies In engagement
  • FinanceOLU Development finance, sovereign funds, capital allocation and project selection In engagement
  • SecurityOLU Duty of care, executive protection, major-event security. Contains AKILI In engagement
  • InfrastructureOLU Capital programmes, roads, ports and utilities agencies Sequenced
  • ConstructionOLU Delivery and contractor performance across jurisdictions Sequenced
  • HealthOLU Ministries of health and national health programmes Sequenced
  • FarmOLU Agricultural production, input finance and land systems Sequenced
  • SecondaryOLU Private equity secondary markets, LP positions and portfolio valuation Sequenced
  • VentureOLU Venture and growth capital on the continent Sequenced
  • GovernmentOLU

    In engagement at ministerial level in West, Central and Southern Africa, and at premier’s-office level in a South African province, where it has been applied as a single instrument across health, education and economic development inside one administration.

  • FinanceOLU

    In engagement with major infrastructure funds and with sovereign and development finance institutions, on capital allocation and project selection.

  • GlobalWatch, within AKILI

    The traveller risk and duty-of-care module, built to the ISO 31030 framework, moving into deployment with blue-chip global enterprises.

Several of these engagements have progressed through paid engagement into production scoping. Counterparties are described by type here. A number are subject to confidentiality undertakings, and specific names can be given in conversation where it is useful to do so.

How an engagement begins

Five gates.
The institution
holds each one.

The first conversation is a listening one: nothing is presented, no proposal is made, and no commitment is sought on either side. Where both parties find it worth continuing, the next stage is Directed Works — one named problem, working software within two weeks, paid and narrowly scoped by design.

Exhibit 3 From introduction to engagement
  1. 01

    Introduction

    A name and a short note

    Somebody who knows both parties puts the name forward. No commitment is sought at introduction, and none is offered.

  2. 02

    Conversation

    One meeting to hear the problem

    The first conversation is a listening one. OLU hears the decision the institution is carrying. Nothing is presented and no proposal is made.

  3. 03

    Directed Works

    One named problem, working software within two weeks

    A paid engagement, narrowly scoped by design. It is how OLU demonstrates capability rather than describing it. An institution that goes no further still owns what was built for it.

  4. 04

    Lighthouse Partnership

    The platform against a defined operational scope

    A named programme, a function or a region. Operational fluency is established and the capability is retained inside the institution’s own perimeter.

  5. 05

    Institutional Rollout

    Full operating geography and scope

    The platform extends across the institution’s full operating geography. By this stage every claim has been tested on work the institution commissioned itself.

Open today Reached on evidence from the stage before

What is asked of the institution

  1. Nothing
  2. An hour
  3. A defined scope
  4. A defined programme
  5. A mandate

An institution may stop at any gate, and keeps what has been built to that point.

A worked example

One named
question, in
two languages

A West African ministry of planning carries the human capital pillar of a five-year national development plan: nine programmes across education, health, vocational training and school feeding. The planning directorate arbitrates between them and a monitoring committee reviews them, and the two had been working from separate assemblies of the same material.

OLU ran a paid Directed Works day against one question — where execution is falling behind, and what moving money between programmes does to the outcome. The module was delivered in French and English, on published data only. The levers it exposes are operational decisions, not budget lines: coverage, execution, timing, each with its fiscal cost and its confidence level.

Exhibit 4 The decision surface the module exposes

Schematic · not a screenshot · no client data · structure only

Lever Position and modelled range Fiscal cost Confidence

Coverage

How many are reached, across nine programmes

NarrowUniversal

Rises with reach

Modelled

Execution

The share of an allocation actually spent in cycle

StalledFull

Broadly neutral

Modelled

Timing

Where in the five years a programme is sequenced

EarlyLate

Shifts between cycles

Modelled
Position under the plan as adopted
Range the module models, under declared assumptions

Every value carries its confidence, and the modelled-versus-observed distinction stays in front of the reader. Coefficients are versioned, annexed and open to the directorate’s revision. The module claims no causation.

Exhibit 5 The levers, in your hands

The levers · operational decisions, not budget lines

Coverage 44

How many are reached, across the programmes in scope

NarrowUniversal
Fiscal cost
Execution 29

The share of an allocation actually spent in cycle

StalledFull
Fiscal cost
Timing 66

Where in the five years a programme is sequenced

EarlyLate
Fiscal cost
Critical reading

Turn this on to read the module against itself. The uncertainty band widens to ±60%, the transmission coefficients lose their defaults, and every figure is marked for what it is.

Modelled human capital index, 2030

0.298 ± 0.054 Modelled
0.300.400.500.600.70202620272028202920302026 · OBSERVED

Components · the index is their product

Health 0.620
Education 0.480

The questions this module is directed at

Execution stands at 29 per cent of full. Closing that gap alone carries the 2030 index from 0.298 to 0.478, a gain of 0.180.

Modelled

Almost all of the movement falls on education, because that is where the execution lever transmits. The gap itself is read from published execution rates; the 2030 figure is modelled.

The set is fixed when the module is built. A module answers the questions it was directed at and says so when a question falls outside them, which is what makes an answer defensible after the fact.

The index is a product, not a sum. Excellent schooling on failing health does not produce average human capital — it is capped by the weaker component, which is why the next unit of effort is worth most where the index is thinnest. Coefficients are versioned, annexed and open to the directorate’s revision. The module claims no causation.

Exhibit 6 The ledger was right. The reading was backwards.

A consignment of medical supplies is being scheduled for District K’s seventeen clinics. Stores are allocated the way stores are allocated everywhere: the next truck goes where stock is lowest. Below is the ledger, the allocation any competent officer would draw from it — and what the same ledger says when read with two other registers.

Stores ledger

Stock on shelf, District K

Six clinics
34 days of stock
Eleven clinics
7 days of stock
Allocation drawn
eleven first · six last

“The six are covered for a month. The consignment goes to the eleven. Correct, on this ledger — and it is how every truck in the country is routed.”

Attendance register

Clinic attendance, District K

District, year on year
−9%
Read as
seasonal
Action
quarterly review

“A soft year across the district. Flagged for the quarterly review.”

Works register

Route N-7, river crossing

Rehabilitation status
suspended
Suspended since
Aug 2027
Priority class
routine

“A delayed rural crossing among forty. Carried forward to next cycle.”

Every reading above is competent. The allocation is the one any storekeeper would draw, and it is about to be actioned. Read the three registers together before the trucks are loaded.

Rules the module runs under

  • No causal claim

    The module models trajectories under declared assumptions. Coefficients are versioned, annexed and open to the directorate’s revision.

  • Two lenses, one platform

    The committee view carries synthesis. The working layer — tables, assumptions, memory — stays with the team that has to defend it.

  • Built to be tested

    A critical-reading mode strips the coefficients of their defaults, widens the uncertainty band, and keeps the modelled-versus-observed distinction in front of the reader throughout.

  • Three-part data regime

    State data stays the state’s, in jurisdiction. Outputs and deliberative memory are the ministry’s. Method improvements carry neither and benefit every deployment.

The people

Who is
in the room

Blessing Rugara

Chief Conductor

Founder and Chairman of Circle Capital Global Holdings, the corporate finance advisory firm he built after seven years practising mergers and acquisitions and structured finance in Minneapolis. Circle Capital has advised on more than $3 billion of transactions across nine African countries, opposite Citi, JP Morgan and the Development Bank of Southern Africa — including the Government of the Democratic Republic of Congo on Ndjili International Airport, and Angola’s Ministry of Tourism on a national hotel programme.

Twenty years as the party both sides could rely on. That is the seat, and OLU is that seat applied to a different kind of decision.

  • Dr Charlie Maere

    Chief Technology Officer

    Doctorate in computer science, specialising in artificial intelligence and distributed systems. Has led national-scale health information deployments across more than 750 hospitals in Malawi under a PEPFAR/CDC programme, and technology delivery across sixteen countries as a global director of digital health and data analytics. Formerly chief technology officer of an engineering firm running teams across the United States, China, South Africa and Ukraine. Published in machine learning, computer vision and medical imaging. Leads the design and engineering of The Directive.

  • Tinashe Jaramba

    Head of Platform Strategy

    An aeronautics engineer working across disciplines, mapping platform capability against institutional requirement.

Advisory board

  • Ibrahima Guimba-Saidou

    Chief Executive of the Partnership for Digital Access in Africa, after a career in senior international technology and telecommunications leadership across the continent.

  • Matsi Modise

    Founder of Furaha Afrika Holdings, with governance roles across fintech, venture capital and financial services.

  • Dr Appolinaire Tiam

    Physician and public health leader, advising on HealthOLU and leading ministry-level health engagement across Southern and Central Africa.

  • Temi Adeniji

    Principal at Porci Capital LLC, bringing international media, rights and investment experience across African and United States markets.

  • Epiphany Vera

    Holds intellectual property in sound, from a former company that produced the audio for Sony and Microsoft gaming platforms. Boston-based.

AKILI is built alongside operators whose practice spans more than three decades of executive protection and major-event security across multiple jurisdictions.

Contact

The first conversation
asks for an hour.

Blessing Rugara takes these conversations directly. Where a suggested institution sits close to work already under way, we say so at once rather than let two approaches arrive from different directions, and we revert on the outcome of any introduction as a matter of course.

hello@oneolu.com
Offices
Mauritius · Dubai · Harare · Johannesburg

oneOLU™ · Alisha Investment and Consulting Ltd · Mauritius, Reg. C13114617

In build · three verticals in active engagement