Governments in the AI era

The Government AI Playbook, Part 1: Start With the Queue

Cari · 2026-09-07 · 6 min read

TL;DR: Do not start your government AI modernization with the biggest budget line or the flashiest ministry. Start with the service that has the worst queue. Score every citizen-facing service by volume times friction times the verification calls it generates, fix the winner first, run the digital path alongside the paper path, and measure days-to-decision honestly. One visible win inside months buys you the political capital to do everything else.

The wrong place to start

Most digital government programmes begin where the money is. A large ministry, a multi-year transformation contract, a platform that promises to fix everything at once. Three years later there is a portal that half works, a procurement file thicker than the legislation it serves, and citizens still standing in the same lines.

The problem is not ambition. The problem is sequencing. A modernization programme that starts with the biggest system takes the longest to show a result, and a programme that shows no result loses its political sponsor before it ships.

There is a better opening move, and every minister already knows where it is. It is the office with the line outside the door.

Why the worst queue is the best first project

The worst queue in your government is a gift. It is visible, it is measurable, and every citizen in it is a voter who will notice when it disappears.

Start there and three things happen inside months, not years.

First, you get a citizen win people can feel. Nobody celebrates a data architecture. Everybody celebrates not taking a day off work to renew a licence.

Second, you build political capital. The minister who cleared the queue gets to propose the next project, and the one after that. The minister who spent the same months on an enterprise platform tender gets questions in parliament.

Third, you build trust in the harder things to come. Registry consolidation, credential infrastructure, cross-ministry data sharing: all of it is easier to sell once citizens have seen one service actually improve.

How to pick the queue: volume times friction times verification calls

Do not pick by instinct and do not pick by which department head shouts loudest. Score every citizen-facing service on three factors and multiply them.

Volume. How many applications, renewals, or requests does the service handle each year? A painful service that ten people use is a footnote. A mildly painful service that fifty thousand people use is a national issue.

Friction. How many trips, documents, signatures, and waiting days does one transaction cost the citizen? Count honestly, including the trip to get the document that the first office demanded.

Verification calls. This is the factor most scorecards miss. How many times does another office, bank, employer, or ministry phone or write to confirm that the permit, licence, or certificate is real? Every verification call is hidden friction multiplied across the whole economy, and it is the factor that credential infrastructure eliminates almost entirely.

Multiply the three. The service at the top of the list is your first project. In most governments it is a licence renewal, a business registration, or a permit that expires annually. It is rarely glamorous. That is fine. Glamour is not the goal, a shorter line is.

Ship the digital path beside the paper path

The fastest way to kill a modernization is to force a switchover. The day the old counter closes and the new system stumbles, you have made the news for the wrong reason.

So do not close the counter. Ship the digital path alongside the paper path. Citizens who want to queue can still queue. Citizens who apply online get their decision faster, and word spreads on its own. The paper path becomes the fallback that protects you while adoption climbs, and it retires itself when the queue is gone rather than when a memo says so.

This also changes what the technology has to be. It does not need to replace the registry on day one. It needs to sit beside it, accept applications, route them, and issue results in a form other parties can verify without calling your office.

That last part matters most. On Cari, permits and licences issue as W3C verifiable credentials, which means a bank, an inspector, or another ministry can check one cryptographically instead of phoning your staff. The verification calls in your scoring formula do not just get faster. They stop arriving.

Speed of deployment matters too, because the whole argument rests on showing a result inside months. Cari deploys 13 registry types in minutes, so the constraint is your process decisions, not the software build. And because the platform is standards-native, with 24,789 UN/LOCODE ports of entry and 6,939 Harmonised System commodity codes built in, trade-facing services do not need a data project before they can start.

Measure honestly or do not bother

Vanity metrics will tempt you. Portal visits, accounts created, downloads. Ignore them. Three numbers tell the truth.

Days-to-decision. From submitted application to issued result. This is the number citizens feel, and it should be on a dashboard the minister sees weekly.

Verification calls avoided. Count the confirmation requests your office used to field and watch them fall as verifiable credentials replace phone-and-fax checking.

Citizen hours returned. Trips avoided, days of income not lost, time given back. This is the number for the budget debate and the podium.

One real example of what the citizen-side number looks like: a small-business owner in Trinidad had been losing three days of income to a queue-driven licence renewal. On Cari's platform, the same renewal completed digitally in under 48 hours. That is the kind of before-and-after a government can stand behind, because it is a person's week, not a systems milestone.

Then expand, registry by registry

The first queue is a beachhead, not the strategy. Once it works, take the scoring formula back out and run it again. The next worst queue is your second project, and it ships faster than the first because the credential infrastructure, the sign-in, and the operational habits already exist.

Registry by registry, the platform compounds. Each new service costs less than the one before it, and each issued credential makes the next verification-heavy process cheaper to fix. That compounding is the subject of the rest of this playbook: Part 2 covers what to build once the beachhead holds, and Part 3 covers how to scale it across ministries without losing the plot.

Common questions

What if the worst queue belongs to a ministry that resists change? Then pick the second worst queue that belongs to a willing one. The formula ranks opportunities, it does not override politics. A slightly smaller win delivered beats a perfect win blocked, and the resistant ministry tends to volunteer after it watches a rival collect the credit.

Does running paper and digital in parallel double our costs? Briefly and marginally, yes, and it is the cheapest insurance you will ever buy. The paper path is already staffed and already running. The digital path drains its volume month by month, and you retire counters on evidence instead of on hope.

Is this too small to count as AI-era modernization? The first project is deliberately small. The infrastructure underneath it is not. Verifiable credentials, standards-native registries, and honest measurement are the same foundations every later AI service will stand on. Starting with the queue is how you build them with the public on your side.

If you are mapping your own worst queue and want to see what the first ninety days could look like, start at cari.global/country-onboarding.