AI, Work & Inclusion
Decision Lab
Eleven intervention levers, examined against one country's evidence at a time.
What this will not do. No lever produces a projected outcome, an effect size, a probability or a benefit figure. Where the evidence base supports arithmetic on observed values, you get arithmetic on observed values. Where it does not, you get the qualitative implication the platform is willing to defend, the assumption behind it, who it would miss, and the evidence that would be needed to measure it. A number invented to fill that gap would be the most dangerous thing on this site.
Dependencies between levers
Which interventions assume another is already in place?
| Tier 1 | Electricity reliability Access to finance Public-sector implementation capability Gender, rural and regional inclusion safeguards |
|---|---|
| Tier 2 | Broadband access (assumes Electricity reliability) MSME and informal-enterprise support (assumes Access to finance) Compute and data ownership (assumes Electricity reliability) |
| Tier 3 | Practical AI and digital-skills investment (assumes Broadband access, Electricity reliability) Digital public infrastructure (assumes Broadband access, Access to finance) Local-language technology (assumes Broadband access) Agriculture and creative-enterprise adoption (assumes Electricity reliability, Broadband access) |
Practical AI and digital-skills investment
Training programmes, curriculum change, teacher capability, certification.
The decision. Whether to commit to practical ai and digital-skills investment in Ghana, and in what order relative to other constraints.
Bounded qualitative rule
What the evidence permits. This lever cannot be calibrated from this evidence base. The platform holds no observed series that would quantify it, so no number is produced for it.
What it does not permit. It does not permit any number at all for this lever — no effect, no reach, no cost. A figure here would be invented.
A skills programme can only reach people who are already reachable. Where secondary enrolment and literacy are the binding constraint, a programme aimed at tertiary-level or online learners will select the already-advantaged, and the measured gain will be real and narrow at the same time.
Preconditions
- Broadband access — this intervention's mechanism assumes it.
- Electricity reliability — this intervention's mechanism assumes it.
Likely first reached
- Urban households — 58.9% of the population lived in towns and cities (2025).
- People already online — 72.2% in 2024.
Likely last reached, or missed
- The 27.8% of people not using the internet in 2024.
- Non-literate adults — literacy was 76.5% in 2021 — wherever the channel is text.
- People who left school before the level the programme assumes; those without reliable power or connection to attend online; women and rural learners where enrolment already diverges.
Delivery risk
The reviewed record shows 8 delivered against 2 at intent. Commitments in this country have been converting, on the evidence reviewed.
Evidence needed to measure the effect
Completion and placement figures disaggregated by sex, location and prior education level, published by the programme rather than by its funder.
Signals bearing on this
- Approved Ghana is one of five countries on the Abidjan–Lagos Corridor Highway, with works due to commence in 2026 and completion earmarked for 2030.
- Announced The Creative Arts Agency announced in 2026 that it will require all creative enterprises in Ghana to register and hold a licence, and report
Broadband access
Network build, spectrum, affordability measures, public access points.
The decision. Whether to commit to broadband access in Ghana, and in what order relative to other constraints.
Observed values
What the evidence permits. 72.2% of people used the internet in 2024, against 114 mobile subscriptions per 100 people in 2024. Subscriptions exceeding users is the ordinary pattern — one person can hold several — so this is not a measure of exclusion on its own. What it does mark is that the constraint is not handset ownership.
9.8m people were not using the internet at that observation.
What it does not permit. It does not permit an effect size, a benefit figure, a cost, a return, a date, or a probability. The values above describe the current position, not the result of acting on it.
Coverage is not use, and use is not affordability. The platform holds the share of people using the internet and the count of fixed broadband subscriptions; the gap between mobile subscriptions and internet use is where affordability and device capability sit.
Preconditions
- Electricity reliability — this intervention's mechanism assumes it.
Likely first reached
- Urban households — 58.9% of the population lived in towns and cities (2025).
- People already online — 72.2% in 2024.
Likely last reached, or missed
- The 27.8% of people not using the internet in 2024.
- Non-literate adults — literacy was 76.5% in 2021 — wherever the channel is text.
- People who are covered but cannot afford data or a device capable of running modern software; areas with coverage but no reliable power to charge.
Delivery risk
The reviewed record shows 8 delivered against 2 at intent. Commitments in this country have been converting, on the evidence reviewed.
Evidence needed to measure the effect
Data-price-to-income series and device capability surveys; coverage maps at sub-national resolution.
Signals bearing on this
- Approved Ghana is one of five countries on the Abidjan–Lagos Corridor Highway, with works due to commence in 2026 and completion earmarked for 2030.
- Announced The Creative Arts Agency announced in 2026 that it will require all creative enterprises in Ghana to register and hold a licence, and report
Electricity reliability
Generation, transmission, distribution, last-mile connection, reliability of supply.
The decision. Whether to commit to electricity reliability in Ghana, and in what order relative to other constraints.
Observed values
What the evidence permits. Household electricity access was 91.9% nationally and 82.6% in rural areas (2024). That is a gap of 9.3 percentage points between the national figure and the rural one. Both measure a connection, not hours of supply.
2.8m people had no household connection at that observation.
What it does not permit. It does not permit an effect size, a benefit figure, a cost, a return, a date, or a probability. The values above describe the current position, not the result of acting on it.
The platform measures household ACCESS — a connection — not hours of supply. A country can raise access without changing whether a workshop can run a machine. The rural/national access gap is measurable here; reliability is not.
Preconditions
- None established from this evidence base.
Likely first reached
- Urban households — 58.9% of the population lived in towns and cities (2025).
- People already online — 72.2% in 2024.
Likely last reached, or missed
- The 27.8% of people not using the internet in 2024.
- Non-literate adults — literacy was 76.5% in 2021 — wherever the channel is text.
- Connected households with unreliable supply; enterprises whose viability depends on hours, not connection.
Delivery risk
The reviewed record shows 8 delivered against 2 at intent. Commitments in this country have been converting, on the evidence reviewed.
Evidence needed to measure the effect
Hours-of-supply or outage-frequency data, which the platform does not hold for any country.
Signals bearing on this
- Approved Ghana is one of five countries on the Abidjan–Lagos Corridor Highway, with works due to commence in 2026 and completion earmarked for 2030.
- Announced The Creative Arts Agency announced in 2026 that it will require all creative enterprises in Ghana to register and hold a licence, and report
Digital public infrastructure
Identity, payments, data exchange, service digitisation.
The decision. Whether to commit to digital public infrastructure in Ghana, and in what order relative to other constraints.
Bounded qualitative rule
What the evidence permits. This lever cannot be calibrated from this evidence base. The platform holds no observed series that would quantify it, so no number is produced for it.
What it does not permit. It does not permit any number at all for this lever — no effect, no reach, no cost. A figure here would be invented.
Digitising a service moves it to the people who can reach it. Where account ownership and internet use are well below universal, digitisation shifts the burden of access onto the citizen unless an assisted or offline channel is retained.
Preconditions
- Broadband access — this intervention's mechanism assumes it.
- Access to finance — this intervention's mechanism assumes it.
Likely first reached
- Urban households — 58.9% of the population lived in towns and cities (2025).
- People already online — 72.2% in 2024.
Likely last reached, or missed
- The 27.8% of people not using the internet in 2024.
- Non-literate adults — literacy was 76.5% in 2021 — wherever the channel is text.
- People without an account, a device or a connection; those whose identity documents do not match registry records.
Delivery risk
The reviewed record shows 8 delivered against 2 at intent. Commitments in this country have been converting, on the evidence reviewed.
Evidence needed to measure the effect
Service-level uptake by channel, and the share of transactions still completed in person.
Signals bearing on this
- Approved Ghana is one of five countries on the Abidjan–Lagos Corridor Highway, with works due to commence in 2026 and completion earmarked for 2030.
- Announced The Creative Arts Agency announced in 2026 that it will require all creative enterprises in Ghana to register and hold a licence, and report
MSME and informal-enterprise support
Registration, tax simplification, market access, digital tooling, procurement.
The decision. Whether to commit to msme and informal-enterprise support in Ghana, and in what order relative to other constraints.
Bounded qualitative rule
What the evidence permits. This lever cannot be calibrated from this evidence base. The platform holds no observed series that would quantify it, so no number is produced for it.
What it does not permit. It does not permit any number at all for this lever — no effect, no reach, no cost. A figure here would be invented.
Most support instruments reach registered businesses. Where the majority of enterprise is unregistered, an instrument that requires registration is not a partial reach — it is a different population, and the unregistered are outside it by design.
Preconditions
- Access to finance — this intervention's mechanism assumes it.
Likely first reached
- Urban households — 58.9% of the population lived in towns and cities (2025).
- People already online — 72.2% in 2024.
Likely last reached, or missed
- The 27.8% of people not using the internet in 2024.
- Non-literate adults — literacy was 76.5% in 2021 — wherever the channel is text.
- Unregistered traders and household enterprises; those for whom registration carries a tax or compliance cost they cannot absorb.
Delivery risk
The reviewed record shows 8 delivered against 2 at intent. Commitments in this country have been converting, on the evidence reviewed.
Evidence needed to measure the effect
Enterprise censuses distinguishing registered from unregistered firms, and programme uptake by registration status.
Signals bearing on this
- Approved Ghana is one of five countries on the Abidjan–Lagos Corridor Highway, with works due to commence in 2026 and completion earmarked for 2030.
- Announced The Creative Arts Agency announced in 2026 that it will require all creative enterprises in Ghana to register and hold a licence, and report
Access to finance
Credit lines, guarantees, credit information, movable-collateral registries.
The decision. Whether to commit to access to finance in Ghana, and in what order relative to other constraints.
Observed values
What the evidence permits. 81.2% of people aged 15 and over held an account at a bank or other financial institution, or personally used a mobile-money service, in 2024. Domestic credit to the private sector was 8.9% of GDP in 2024. Account ownership is a payment rail. Nothing in this platform establishes that account records are used in any lending decision, here or anywhere.
4.3m people aged roughly 15 and over held neither, on a rough application of the adult share.
What it does not permit. It does not permit an effect size, a benefit figure, a cost, a return, a date, or a probability. The values above describe the current position, not the result of acting on it.
Domestic credit to the private sector as a share of GDP describes the aggregate depth of lending, not who is refused. An account is a payment rail, not a credit history — the platform holds no evidence that account records are used in lending decisions anywhere.
Preconditions
- None established from this evidence base.
Likely first reached
- Urban households — 58.9% of the population lived in towns and cities (2025).
- People already online — 72.2% in 2024.
Likely last reached, or missed
- The 27.8% of people not using the internet in 2024.
- Non-literate adults — literacy was 76.5% in 2021 — wherever the channel is text.
- Borrowers with no collateral and no formal credit record; women and rural borrowers where lending data is not disaggregated.
Delivery risk
The reviewed record shows 8 delivered against 2 at intent. Commitments in this country have been converting, on the evidence reviewed.
Evidence needed to measure the effect
Firm-level credit surveys recording refusals and their reasons; credit-bureau coverage figures.
Signals bearing on this
- Approved Ghana is one of five countries on the Abidjan–Lagos Corridor Highway, with works due to commence in 2026 and completion earmarked for 2030.
- Announced The Creative Arts Agency announced in 2026 that it will require all creative enterprises in Ghana to register and hold a licence, and report
Local-language technology
Speech and text datasets, models, interfaces in languages people actually use.
The decision. Whether to commit to local-language technology in Ghana, and in what order relative to other constraints.
Bounded qualitative rule
What the evidence permits. This lever cannot be calibrated from this evidence base. The platform holds no observed series that would quantify it, so no number is produced for it.
What it does not permit. It does not permit any number at all for this lever — no effect, no reach, no cost. A figure here would be invented.
A service in a language a user does not read is unavailable to them regardless of coverage. Where literacy is well below universal, text interfaces exclude by construction and voice becomes the access question rather than a feature.
Preconditions
- Broadband access — this intervention's mechanism assumes it.
Likely first reached
- Urban households — 58.9% of the population lived in towns and cities (2025).
- People already online — 72.2% in 2024.
Likely last reached, or missed
- The 27.8% of people not using the internet in 2024.
- Non-literate adults — literacy was 76.5% in 2021 — wherever the channel is text.
- Speakers of languages with no dataset; non-literate users of text-only interfaces.
Delivery risk
The reviewed record shows 8 delivered against 2 at intent. Commitments in this country have been converting, on the evidence reviewed.
Evidence needed to measure the effect
Language-coverage inventories for deployed services, and use by language.
Signals bearing on this
- Approved Ghana is one of five countries on the Abidjan–Lagos Corridor Highway, with works due to commence in 2026 and completion earmarked for 2030.
- Announced The Creative Arts Agency announced in 2026 that it will require all creative enterprises in Ghana to register and hold a licence, and report
Agriculture and creative-enterprise adoption
Advisory services, market information, payment and rights infrastructure.
The decision. Whether to commit to agriculture and creative-enterprise adoption in Ghana, and in what order relative to other constraints.
Bounded qualitative rule
What the evidence permits. This lever cannot be calibrated from this evidence base. The platform holds no observed series that would quantify it, so no number is produced for it.
What it does not permit. It does not permit any number at all for this lever — no effect, no reach, no cost. A figure here would be invented.
Where a large share of employment is in agriculture but its share of output is much smaller, the productivity question is the employment question. The platform can state both shares; it cannot establish that a given tool moves either.
Preconditions
- Electricity reliability — this intervention's mechanism assumes it.
- Broadband access — this intervention's mechanism assumes it.
Likely first reached
- Urban households — 58.9% of the population lived in towns and cities (2025).
- People already online — 72.2% in 2024.
Likely last reached, or missed
- The 27.8% of people not using the internet in 2024.
- Non-literate adults — literacy was 76.5% in 2021 — wherever the channel is text.
- Smallholders without connection, power or a payment account; creative workers whose sector is not measured in national accounts at all.
Delivery risk
The reviewed record shows 8 delivered against 2 at intent. Commitments in this country have been converting, on the evidence reviewed.
Evidence needed to measure the effect
Adoption and yield or income panels; creative-sector satellite accounts.
Signals bearing on this
- Approved Ghana is one of five countries on the Abidjan–Lagos Corridor Highway, with works due to commence in 2026 and completion earmarked for 2030.
- Announced The Creative Arts Agency announced in 2026 that it will require all creative enterprises in Ghana to register and hold a licence, and report
Public-sector implementation capability
Civil-service skills, procurement capacity, delivery units, maintenance funding.
The decision. Whether to commit to public-sector implementation capability in Ghana, and in what order relative to other constraints.
Bounded qualitative rule
What the evidence permits. This lever cannot be calibrated from this evidence base. The platform holds no observed series that would quantify it, so no number is produced for it.
What it does not permit. It does not permit any number at all for this lever — no effect, no reach, no cost. A figure here would be invented.
The governance indicators are perception-based composites with wide uncertainty. They are a weak signal about capability and no signal at all about a specific programme. Where a country’s delivery record shows announcements outnumbering deliveries, capability is the more likely constraint than intent — but that is an inference from the record, not a measurement.
Preconditions
- None established from this evidence base.
Likely first reached
- Urban households — 58.9% of the population lived in towns and cities (2025).
- People already online — 72.2% in 2024.
Likely last reached, or missed
- The 27.8% of people not using the internet in 2024.
- Non-literate adults — literacy was 76.5% in 2021 — wherever the channel is text.
- Sub-national administrations, which the national indicators do not describe.
Delivery risk
The reviewed record shows 8 delivered against 2 at intent. Commitments in this country have been converting, on the evidence reviewed.
Evidence needed to measure the effect
Programme-level delivery audits; procurement completion rates.
Signals bearing on this
- Approved Ghana is one of five countries on the Abidjan–Lagos Corridor Highway, with works due to commence in 2026 and completion earmarked for 2030.
- Announced The Creative Arts Agency announced in 2026 that it will require all creative enterprises in Ghana to register and hold a licence, and report
Compute and data ownership
Data centres, cloud, network capacity, data-governance rules.
The decision. Whether to commit to compute and data ownership in Ghana, and in what order relative to other constraints.
Bounded qualitative rule
What the evidence permits. This lever cannot be calibrated from this evidence base. The platform holds no observed series that would quantify it, so no number is produced for it.
What it does not permit. It does not permit any number at all for this lever — no effect, no reach, no cost. A figure here would be invented.
Hosting capacity located in a country is not capability owned by it. The platform records where facilities and cable landings are; it holds nothing on who owns them, who can afford them, or whether local firms use them.
Preconditions
- Electricity reliability — this intervention's mechanism assumes it.
Likely first reached
- Urban households — 58.9% of the population lived in towns and cities (2025).
- People already online — 72.2% in 2024.
Likely last reached, or missed
- The 27.8% of people not using the internet in 2024.
- Non-literate adults — literacy was 76.5% in 2021 — wherever the channel is text.
- Local firms priced out of the capacity that exists; researchers with no allocation.
Delivery risk
The reviewed record shows 8 delivered against 2 at intent. Commitments in this country have been converting, on the evidence reviewed.
Evidence needed to measure the effect
Ownership and utilisation data for installed capacity; local-firm pricing.
Signals bearing on this
- Approved Ghana is one of five countries on the Abidjan–Lagos Corridor Highway, with works due to commence in 2026 and completion earmarked for 2030.
- Announced The Creative Arts Agency announced in 2026 that it will require all creative enterprises in Ghana to register and hold a licence, and report
Gender, rural and regional inclusion safeguards
Targets, disaggregated reporting, targeted delivery channels.
The decision. Whether to commit to gender, rural and regional inclusion safeguards in Ghana, and in what order relative to other constraints.
Observed values
What the evidence permits. Where the platform holds a disaggregated series: rural household electricity access sits 9.3 percentage points below the national figure (2024); 55.3% of women aged 15 and over were in the labour force, employed or unemployed (2025) — a participation rate, not women's share of the workforce. These are the only inclusion dimensions this platform can measure for most countries. Disability, internal migration status and sub-national variation are unexamined, so a safeguard aimed at them could not be verified here.
What it does not permit. It does not permit an effect size, a benefit figure, a cost, a return, a date, or a probability. The values above describe the current position, not the result of acting on it.
A safeguard that is not measured is not a safeguard. Where the platform holds a rural and a national figure for the same measure, the gap between them is the size of the problem a safeguard would have to close; where only a national figure exists, the distribution is unexamined and the safeguard is unverifiable.
Preconditions
- None established from this evidence base.
Likely first reached
- Urban households — 58.9% of the population lived in towns and cities (2025).
- People already online — 72.2% in 2024.
Likely last reached, or missed
- The 27.8% of people not using the internet in 2024.
- Non-literate adults — literacy was 76.5% in 2021 — wherever the channel is text.
- Groups for which no disaggregated series exists at all — which, for most of these countries, includes disability and internal migration status.
Delivery risk
The reviewed record shows 8 delivered against 2 at intent. Commitments in this country have been converting, on the evidence reviewed.
Evidence needed to measure the effect
Disaggregated reporting as a condition of programme financing.
Signals bearing on this
- Approved Ghana is one of five countries on the Abidjan–Lagos Corridor Highway, with works due to commence in 2026 and completion earmarked for 2030.
- Announced The Creative Arts Agency announced in 2026 that it will require all creative enterprises in Ghana to register and hold a licence, and report