Agreed definitions
The consulting team sets the criterion; each ministry approves what a completed procedure is and what its baseline is. No approved definition, no published indicator.
AI agents, measured.Citizens, more prosperous.
LeanGov measures which procedures citizens complete with the State’s AI agents, where they get stuck and what each one costs in AI. And it turns that measurement into what matters: time given back to people, businesses that are born and formalized, and a contribution to GDP per capita that can be proven.
Or start from what we hear most often:
Case study · first national deployment
National AI-agents program for citizen proceduresGovernment of El Salvador · Secretariat of Innovation · measured by LeanGov from the very first procedure×500licenses in one quarter18processes in 6 months5Mlicenses at scaleSee the case- Controls aligned to the SOC 2® framework
- Read-only: we never touch the agents
- We never see a citizen’s identity
- Runs in the State’s own cloud
The impact
What changes when the State serves citizens with AI agents, and measures it.
A procedure that used to take weeks is resolved in one conversation. That difference, multiplied by millions of citizens and measured procedure by procedure, is economic growth that can be proven.
- Citizen
Time given back to people
Procedures resolved in minutes, at any hour, with no line and no counter. Days of waiting that go back to work, study and family.
Measured as: time saved per completed procedure against the traditional channel. - Businesses
Businesses that are born and formalized
Opening a business, filing taxes or exporting stops being a barrier. More formal entrepreneurs, faster and at a lower transaction cost.
Measured as: businesses created, days to registration and businesses active after a year. - Inclusion
The same State for everyone
The agent serves the capital and the countryside alike, the person who knows the paperwork and the one who never filed anything. Segments that don’t use it become visible, and get fixed.
Measured as: adoption, usage and abandonment by age, region and segment. - Country
A richer country, with evidence
Every completed procedure connects to its economic effect: avoided transaction costs, formalization, SMEs that grow and export, contribution to GDP per capita.
Measured as: the economic-impact module, with declared assumptions and confidence levels.
And the State learns. It knows which agent performs and which fails, where to put licenses and AI budget, and what to fix first. With one single account, auditable, in its own cloud.
See what we measureWhat we measure
One clear unit, one simple account, friction in plain sight.
A completed procedure is one the citizen starts with the agent and the ministry accepts. Everything else is measured against that.
- Completed procedureStarted with the agent, accepted by the ministry. What counts as “accepted” is agreed with each ministry, in writing.
- FrictionProcedures that started with the agent, consumed tokens and never finished. Reported separately and classified by reason.
- ReturnAI investment ÷ completed procedures: what each procedure a citizen finishes costs in AI.
- Baseline and segmentsEach process against its own history and by channel: agent, web or counter. Broken down by the segments the State defines.
Where each agent-enabled process lands
Each bubble is one processReading it: each bubble is a process · size = procedures started per month · position = what the AI spend is buying · illustrative data
The LeanGov methodology
From the national plan to the procedure, with one single account.
Indicators cascade down from the country’s objectives; evidence flows up from every completed procedure. What a ministry measures is what the Presidency decides on.
Top downPriority flows down: vectors, targets and owners
01National strategic planThe country’s vectors and key actors
02Program objectivesAgent-enabled processes with a target: autonomy and % completed
03Management indicatorsAdoption by segment, completed procedures and progress against the plan
04Technical indicatorsEach agent’s operation: where it gets stuck, rework and cost per process
Bottom upEvidence flows up: completed procedures and return
Fair comparisons
The agent channel against the traditional one, and license-rollout stages against each other, to separate the agent’s effect from other changes.
Declared coverage
Every indicator states which data backs it and what is missing. If a source isn’t there, it is reported; nothing is estimated in silence.
Three levels of reading
Program leadership sees the total impact. Each ministry sees its processes. Technical teams see the detail. Through the dashboard, API and MCP, with the same permissions.
How it works
Context, diagnosis and treatment, every month.
One cycle inside the platform. The measured result becomes the new baseline, and if the failure returns, so does the alert.
Stage 1 · Context
We connect read-only and set each process’s baseline
Agent telemetry, each procedure’s status over API, the protected identifier, and licenses and tokens. With that, every process gets its baseline: flow, adoption, friction and AI consumption.
- Agent telemetry, statuses over API, licenses and tokens: read-only
- Protected national ID or tax ID: we join conversation and procedure without knowing whose it is
- Baseline per process: up to 90 days of history, or the first 30 measured
National AI-agents program
Economy
Treasury
Health
One API format for every ministryOne process connected per month, behind a quality gate
Stage 2 · Diagnosis
The platform finds what drifts from its baseline and groups it by cause
A document that blocks, a step where the agent fails, a segment that barely uses it. Every anomaly is grouped by process, step, segment and agent version, and prioritized by the procedures at risk.
- Anomalies grouped by common cause, not one alert per symptom
- If an agent fails, leadership gets the alert with the evidence
- Prioritized by procedures at risk, not by alert volume
Missing power of attorney34 %1,240 procedures
Agent fails to validate the tax ID22 %810 procedures
Rural areas barely use it12 %430 procedures
Stage 3 · Treatment
Every cause has a treatment. LeanGov proposes; the State decides and executes.
Fix the process at the ministry, adjust the agent, move usage quota between segments or expand licenses where the agent already performs. Prioritized by expected effect; measured afterwards, to close the case.
- Fix the process · adjust the agent · move quota · expand licenses
- The measured result becomes the new baseline
- With history: demand, license and cost projections, and proactive agents
Who does each step
- measure LeanGov
- propose LeanGov
- decide the State
- execute the State
The return
One simple account, audited in the State’s own cloud.
The return on public AI isn’t assigned licenses or conversations. It’s what the citizen completed, over what it cost in AI.
AI investment
US$ 5,240
tokens attributed to the process / month
Completed procedures
12,480
started with the agent and accepted by the ministry
AI cost per completed procedure
US$ 0.42
per process, segment and channel
Friction · reported separately
2,310
started and never finished, by step and reason
- Completed: agreed and signed off by each ministry
- Investment: tokens attributed to the process
- Friction: explained, never silently subtracted
- Illustrative data
How we measure economic impact
From the measured procedure to the growing business, to the citizen’s wealth.
A platform module carries every completed procedure through to economic impact: time saved, businesses created, businesses that grow and export, and estimated contribution to GDP per capita. Every figure declares its assumptions and its confidence level.
- Citizen
Time saved per completed procedure
Against the traditional channel. Agent adoption by age and region; procedures completed through the platform.
- Entrepreneurship
Businesses created through the measured procedures
Days from start to registered business; businesses still active after a year.
- Business growth
SMEs that move up a size bracket and start exporting
Foreign-trade procedures completed with agents; associated formalization.
- National economy
Estimated contribution to GDP per capita
With declared assumptions: avoided transaction costs and formalization, calibrated quarterly and read against the region.
The baselines are validated with program leadership at launch. Whatever has no source or no measurable process behind it is declared as such.
Case study
A national program growing ×500 in one quarter, measured before the volume arrives.
The first deployment: the Government of El Salvador is launching AI agents for citizen procedures, from 10,000 licenses to 5 million in the first quarter of 2027. LeanGov measures the program from the very first procedure, in the State’s own cloud. The same methodology is ready for the next country.
- ×500licenses in one quarter: from 10,000 to 5 million
- 2initial processes: business registration and income tax return
- 18processes under official measurement by month 6
- +3new processes every month, under the license
- Oct 2026LaunchPlatform deployed in the State’s Google Cloud, API and MCP, two end-to-end processes and a technical map of the ministries.
- Jan 1, 2027Platform at scaleCapacity for 5 million licenses and 60 million usage records a month, before the volume arrives.
- Apr 202718 processes measuredUp to 5 being connected at a time, one month per process, behind a quality gate. First economic-impact report.
- May 2027 →Continuous improvementMonthly cycle across every process, three new processes a month, proactive agents and an annual report.
Implementation
A platform ready to scale in three months. A license that follows the program.
Two implementation phases and a license that adjusts to the program’s own license count. A dedicated team, present in the country from kick-off.
- Phase 1 · 3 months
Launch
- Definition of success and indicators by level
- Platform in the State’s cloud, ready for scale
- API and MCP for the State, with the dashboard’s permissions
- Two initial processes with dashboard and alerts
- Technology survey and scale plan per ministry
- Phase 2 · 3 months
Scale-up
- From 2 to 18 processes, up to 5 being connected at a time
- Access per ministry and an incident protocol
- Friction explained and automated reports
- Rules for assigning licenses and agent quality
- Demand, license and AI-cost projections
- License · from month 7
Efficiency and continuous improvement
- Monthly cycle: anomalies, treatments and re-measurement
- Three new processes a month, over the same API
- Friction prediction by process and segment
- Agents for proactive treatments
- Annual report and annual committee
How we connect each process
API as the rule, built together with each ministry. One month per process, without losing quality.
- 1SpecificationOne procedure-status API format: states, transitions, channel and protected identifier.
- 2Co-construction, includedWorkshop, reference code, review and tests. The ministry installs it in its own systems; we never go in.
- 3Quality gateAPI tested, security approved, reconciliation ≥ 98% and a signed definition of “completed”. If it doesn’t pass, it isn’t published.
- On-site kick-offthe first week, in the country
- Resident success managerdaily coordination with the ministries
- One week a monthmeasurement committee with leadership
- Data scientistanomalies, projections and proactive treatments
Privacy, security and infrastructure
We never receive a citizen’s identity. Data never leaves the State’s cloud.
The platform is deployed in a project inside the State’s own cloud organization, with infrastructure as code and least-privilege access. Privileged access is approved by the State and stays in its audit logs.
Leanmote’s methodology and platform. Controls aligned to the AICPA SOC 2® framework, examined by an independent auditor.
Identity protected at the source
The State hands over the national ID or tax ID already converted into an irreversible code. We never see or store the real number; it only joins the conversation to the procedure without knowing whose it is.
No per-person data
We never show groups smaller than 10 cases or export individual information. Everything is read by process, segment and channel.
We never touch the agents
The agents are operated by the State with its provider. LeanGov reads, proposes treatments and measures their effect. Nothing changes without the State’s decision.
For governments and multilateral organizations
A methodology that replicates in any country. A standard to compare and to fund.
What is measured the same way can be compared across ministries, programs and countries. That is the evidence needed by those who decide, and those who fund, AI adoption in government.
Decide with one single account
Every ministry measured with the same yardstick, alerts when an agent fails, and clear rules for assigning licenses and AI budget.
Return and impact, with declared assumptions
AI cost per completed procedure, return per process and economic impact computed on what was measured. An annual report that reads the country against the region.
Comparable evidence to fund the scale-up
Indicators comparable across countries, measured by an independent third party. The basis for funding and scaling AI-agent programs where they prove a return for citizens.
- Same standard APIfor procedure status, in any ministry
- Same quality gatereconciliation ≥ 98% before any figure is published
- Same indicatorsfrom the national plan down to the procedure, in cascade
- Adapted to each countryto its infrastructure, human capital and institutional maturity
Frequently asked questions
What we get asked before we start.
If your question isn’t here, write to us. We reply within one business day.
Does LeanGov build or configure the AI agents?
No. Citizen-facing agents are built and operated by the State with its provider. LeanGov measures them, proposes agents for proactive treatments and measures their effect. We don’t review the content of each answer either: the provider’s filters do that.
Where does the data run?
In the State’s own cloud: a project inside its cloud organization, deployed with infrastructure as code. LeanGov only accesses that project, with least-privilege permissions approved by the State. The data never leaves it.
Do you see citizens’ personal data?
No. The State hands over the identifier (national ID, tax ID or equivalent) already converted into an irreversible code. With it we join the conversation to the procedure without knowing whose it is. We never show groups smaller than 10 cases or export individual information.
What if a ministry has no API?
We build it together, at no extra cost: we publish a common format, run workshops with their developers, hand over reference code and review the tests. The ministry installs it in its systems. If a system truly cannot have an API, it is connected another way and declared as a risk from the start.
How long does it take to connect a process?
About two weeks if the API already exists or the process is already clear; about four if the ministry builds the API with us. One month at most, survey and security included. A process enters the official dashboard only once it passes the quality gate: API tested, security approved, reconciliation of at least 98% and a signed definition of “completed”.
What does the State need to get started?
An executive sponsor, a business counterpart for the day to day, a technical counterpart who provides the cloud project and access, and an owner per process in each ministry. In the data: the channel flag (agent, web or counter), the protected identifier and access to the agents’ usage data.
Can it be applied in another country or program?
Yes. The methodology, the API format, the quality gate and the cascading indicators are the same for any government AI-agents program; what adapts is the pace, the infrastructure and the order of each country’s processes. The first national deployment is already under way, and the same platform can measure several countries with comparable indicators.
How is LeanGov related to Leanmote?
LeanGov is Leanmote’s government line. It shares its methodology (context, diagnosis, treatment), its platform and its security controls, applied to government AI programs and their impact on citizens.
Before the program scales,the measurement has to be ready.
Tell us which agents you are launching and for which procedures. We come back with a measurement proposal within a week.
- Read-only: we never touch the agents
- Identity protected at the source
- Platform at scale in three months