Strengthen, Don't Police: How to Get Every Ministry to Say Yes

By Anushka Appala and Dr. Janio Rosales

The hardest problem in government technology is not technical. The models exist. The rules engine is well-understood. The cryptography is decades old. The genuinely difficult part is getting a dozen institutions — each with its own mandate, its own legal counsel, and its own well-earned caution — to agree to work through the same platform. Coordination, not computation, is where projects like this die.

We have come to believe the reason is almost always framing. A platform that arrives as an inspector, promising to watch officials and catch wrongdoing, will be resisted by every office it touches, and it should be. Nobody signs up to be policed. But a platform that arrives as a tool — one that helps each institution do the job it is already legally charged with doing, faster and with a cleaner record — is a different proposition entirely. The task is to give every institution a reason grounded in its own mandate to say yes.

Why the prize is worth the coordination

The status quo is expensive enough to justify the effort. Latin America wastes 4.4% of GDP — roughly US$220 billion a year — across procurement, payroll, and transfer leakage (IDB, Better Spending for Better Lives, 2018). Narrow the lens to contracting and the loss is starker still: an estimated 10 to 25% of every public contract's value is lost to corruption (UNODC). These are not abstract figures. They are ghost beneficiaries on a transfer roll, duplicate payments no one reconciled, and contracts awarded in the dark. Every one of them is a coordination failure as much as a criminal one — information that existed somewhere in the state but never met the decision that needed it.

That is the opportunity. Not to expose people, but to connect the information the state already holds, so that fewer quetzales leak and more citizens are served. The question is how to make that case, institution by institution.

A reason for each institution, in its own terms

MIDES — the Ministry of Social Development — wants to serve families faster and stop paying for people who are not there. NaciluzIA drops the time to complete a social-program application from weeks to minutes, in Spanish and Q'eqchi', while a deterministic rule checks for duplicate and ghost payments before any disbursement is signed. That is not surveillance of MIDES officials; it is MIDES delivering its own program with fewer errors and a defensible record of each one. The ministry gets speed and clean rolls — the two things it is judged on.

RENAP — the national registry — holds the identity data that every benefit check depends on, and its legitimate fear is misuse of that data. The design answer is least-privilege verification: the platform asks only the narrow question it needs answered — is this DPI valid, is this person living — and never bulk-copies the registry. Crucially, every query is recorded in the ledger. RENAP gains something it does not have today: a tamper-evident record of exactly who asked what, when, and under which rule. That strengthens RENAP's control over its own data rather than weakening it.

The Contraloría — the comptroller — already runs the audits. Its constraint is evidence: reconstructing after the fact what happened and why. NaciluzIA hands it a tamper-evident, hash-chained evidence trail, valid under Guatemala's e-signature law (Decreto 47-2008), captured at the moment each decision is made. This does not replace the audits the Contraloría performs; it makes them faster and harder to dispute. We are strengthening a function that exists, not inventing a new watchdog.

SAT and IGSS — the tax authority and the social-security institute — each run their own investigations and each suffer from the same blind spot: patterns that are only visible when records are cross-checked. The platform can surface those cross-checks — an income inconsistency, an enrollment that does not reconcile — and route them to their own investigators for legally authorized human review. The AI surfaces the pattern; it never determines misconduct. It identifies what deserves a look, and leaves judgment where it belongs.

Strengthen the institution, never expose the official

Notice what every one of those reasons has in common. In each case, the platform strengthens the institution's own mandate and hands its own people better tools. It never repositions itself above them as a monitor. This is a deliberate and load-bearing choice, not a matter of tone.

It follows directly from the principle the whole system is built on: AI recommends, policies decide, officials oversee, audits verify. The AI surfaces evidence and drafts the paperwork; a deterministic engine executes the published rule; a named official signs; the ledger records it. At no point does the platform pass judgment on an official. When it flags an anomaly, it is flagging something for legally authorized human review inside the institution that owns the case — not accusing anyone of wrongdoing. "The system catches corrupt officials" is precisely the framing that guarantees no ministry will adopt it. "The system gives your investigators cross-checks they cannot get today" is the framing that gets a yes.

The distinction is not cosmetic. An architecture that could be turned into a surveillance tool against officials would be resisted, and rightly, because trust in Latin American institutions is already thin enough without adding a machine that watches from above. What earns adoption is the opposite promise: this makes your job easier, your record cleaner, and your data more firmly under your own control.

That is the whole strategy. Do not build a police force and ask the ministries to submit to it. Build a tool, give each institution a reason it recognizes as its own, and let coordination follow interest rather than fight it. The technology was never the hard part. Aligning incentives is — and it is solvable, one mandate at a time.

Cada decisión, a la luz.

— Anushka Appala and Dr. Janio Rosales