01 · Context

The number that was never taken

The Argentine cohort was not improvised. The call for Escala GovTech opened on 3 August and closed on 31 August 2026; an in-person pre-selection event ran on 28 September; the final eight companies were confirmed on 5 October; the in-person modules ran 19–21 October, with online mentoring through November and a Demo Day in Buenos Aires in the week of 16 November. Two of the eight will receive pilot funding.

The programme is the Argentine chapter of a regional arrangement. GovTech Connect is financed by BID Lab, coordinated by the Red de Innovación Local and supported technically by Gobe Ventures; Universidad Austral, through CESIS and EmprendeIAE, is the national executing agent. ANII reported that more than 40 candidacies were submitted by organisations in Argentina, Brazil, Chile and Uruguay to act as those national agents, and that four were selected — ANII in Uruguay, Universidad Austral in Argentina, Impact Hub Brasil and Start-Up Chile. Uruguay alone has five cycles scheduled between 2026 and 2028.

So the supply side is organised, funded and regional — and the supply side is not where the failure sits. When I asked the eight teams to write what problem they solve, for which citizen, and what changes in that person’s life, almost every team began by describing its own product. I had to ban the words platform, app and artificial intelligence from the sentence. Once banned, the sentences changed completely: the passenger waiting without knowing whether the bus has already gone; the three-person procurement team rewriting the same specification every month. That is the sentence a mayor acts on. The other one is POC theater.

Sergio Mastrogiovanni speaking with a microphone on an open-air wooden stage at Triple Fest beside Marina Santucci, in front of a projected slide headed La ciudad inteligente que queremos listing concrete applications for digital government and efficient mobility, with Universidad Austral and CESIS branding below
Triple Fest, October 2026. Marina Santucci and the author, with the slide the session argued against taking at face value: a column of concrete applications, and no column for the number each one is supposed to move.

The second exercise was the three numbers, and the instruction that mattered was the last one: for each number, state the value it stands at today. That is where the cohort stopped. Nearly every team had planned to measure from its own system, starting on day one of the pilot — which is to say, with no before. Without a before, nothing is demonstrable in either direction: the pilot cannot prove it worked, nor a sceptical comptroller that it did not. The session ended with a five-line email template for requesting the baseline from the government counterpart. It is the least technological artefact imaginable, and it decides whether a pilot can prove anything at all.

This is not an Argentine peculiarity. The World Bank’s GovTech Maturity Index 2025 update, covering 197 economies across 48 indicators, found progress real but uneven: João Ricardo Vasconcelos and Cem Dener, writing for the Bank in January 2026, report that the top maturity group grew from roughly 35% of economies in 2022 to 41% in 2025, while “nearly half of economies still lack fully implemented cloud and interoperability backbones” and about four in ten still have no operational one-stop service portal. Digital citizen engagement remains the least mature component almost everywhere. The visible layer keeps being bought; the layers that make it measurable and durable keep being skipped. BID Lab’s own open-innovation guide, which uses Córdoba’s CorLab and the Fondo Córdoba Ciudad Inteligente as a worked case and draws on a GovTech LATAM programme that accompanied 10 municipalities across 7 countries, ends its five-phase sequence not at test the technology but at scale the learning. Most programmes stop one phase early.

A pilot without a before-number cannot prove anything. It can only impress.

02 · Framework

The Day-91 Contract

The frame below is the one I use with public teams and the companies selling into them, assembled from two literatures that rarely meet. Willem van Winden and Daniel van den Buuse, writing in the Journal of Urban Technology in 2017, separate upscaling into roll-out, expansion and replication, and observe that most smart-city pilots fade after the pilot stage; Tania Prinsloo and Funmi Adebesin, synthesising 25 studies for Springer in 2023, give the condition its clinical name — pilotitis — and sort its causes into four families, none of them technological. Eleonora Sista and Pietro De Giovanni, in Smart Cities in 2021, reach the same place from a single logistics case: of the four factor families that decide scaling, only one is technical.

Three layers, in the order they must be written. The order is the argument: each layer is unfalsifiable without the one below it.

Layer 01 · Baseline

The before-number, computed from the government’s own system of record — the case-management system, the appointment queue, the complaints register, the payroll — and never from the vendor’s instrumentation. It carries a named data custodian inside the administration and a date. If the baseline cannot be obtained, that is the finding, and worth more than the pilot: an administration that cannot produce its own current performance is not ready to buy a system that improves it. KPIs before APIs is not a slogan here; it is the sequence.

Layer 02 · Ownership

Not a person — a post. Which office signs continuity on day ninety-one, against which budget line, under which contractual instrument. Written before day one, together with the stop criterion: the number at which the pilot closes without argument, agreed while nobody is yet invested in the answer. The hardest thing to write is not the technology; it is the title of the official who will sign. The discomfort is diagnostic. If that post does not yet exist, the pilot is born orphaned, and far cheaper to discover on paper than with the work finished and the invoice unpaid.

Layer 03 · The Four Signatures

Scaling is not a money or technology problem. It is a trust problem between parties that do not normally speak, requiring four guarantees no single institution supplies. Risk cover: a multilateral or national backer whose presence lets an official justify the risk. Reach: a network with standing relationships across many municipalities. Method: a technical partner with a procedure already tested elsewhere. Neutrality: an evaluator with no commercial stake, because a result measured by the vendor is worth less than the same result measured by a university. In Escala GovTech those four sit on BID Lab, the Red de Innovación Local, Gobe Ventures and Universidad Austral respectively. That is what an ecosystem is for. It is not a networking layer on top of the work; it is the scaling mechanism itself.

So what: a government does not buy technology; it buys a measurable improvement in a result it is already obliged to produce. Everything that does not survive that sentence is POC theater.

03 · Use Cases

Three loops, three systems, three overrides

The eight companies in the cohort span seven urban verticals — citizen contact, mobility, traffic control, public procurement, data protection, health administration, employment and public safety. Three are set out below as operating patterns rather than company descriptions, each naming the system touched, the decision loop changed, the override path and the metric that would demonstrate value. Quantified ranges are illustrative and drawn from no client engagement.

01

Citizen contact measured as resolution, not as volume. System implicated: the municipal case-management system and electronic file, with a conversational channel in front of it. Buenos Aires’ Boti, running on WhatsApp since 2019, is the public reference point. Decision loop: triage of an inbound request into resolve-in-channel, route to a back-office queue, or escalate. Override path: a human agent takes the thread whenever model confidence falls below the declared band, whenever the citizen asks, and always for any request with a legal deadline. Metric that demonstrates value: the share of procedures closed without an in-person visit, and median time to a substantive answer — never messages processed, which measures the technology rather than the citizen’s time. The gap between used it and it solved my problem is the whole measurement, and a show of hands in a public square reproduces it reliably. Illustrative, once the case-management integration exists rather than a standalone bot: 10–20 percentage points of in-person avoidance.

02

Route data as a decision input, not as an open-data trophy. System implicated: the transport authority’s route and timetable registry, published to a standard that journey planners already consume, plus telemetry from the vehicles themselves — Mendoza’s electric buses are sensors that happen to carry passengers. Decision loop: two of them — the passenger’s departure decision, and the operator’s frequency and routing adjustment. Override path: the control room can suspend automated frequency changes and revert to the published timetable, with the suspension logged and attributed. Metric that demonstrates value: minutes of uncertain wait per passenger-day, baselined by survey before the feed exists. In most Argentine cities routes live in somebody’s head or on paper in the mobility secretariat, not as data — so the technology at issue is a file format, and the outcome is returned life-time. Illustrative: in a city of a hundred thousand, even 8–12 minutes per passenger daily aggregates into thousands of hours. An electric bus whose telemetry changes no decision is simply a cleaner bus.

03

Procurement drafting and bid analysis with the signature unchanged. System implicated: the public purchasing platform and the electronic file that carries the award. Decision loop: drafting the technical specification, and comparing compliant offers. Override path: the model drafts and compares but never scores a bidder without an attributable human reviewer, and the procurement officer’s signature is the only act that awards; every model-generated clause is marked as such in the file so an auditor can separate it later. Metric that demonstrates value: cycle time from identified need to signed award, and the share of tenders declared void for defective specification — a number most administrations already hold and almost none publish. This is the quiet use of the technology with the cleanest business case: a three-person purchasing team rewriting the same specification monthly is a measurable cost with a measurable before. Illustrative cycle-time reduction where the drafting stage is genuinely the constraint: 15–25%.

04 · Implementation

What to require before the money moves

The operative artefact is one page, written before day one and attached to whatever instrument funds the pilot. It carries: the hypothesis stated as a number (“in three health centres, the wait for an appointment falls from eleven days to six”); the baseline measured from the administration’s own records with a named custodian; the stop criterion; the post that signs continuity on day ninety-one, its budget line and its contractual instrument; and the answer to the two questions a public buyer asks whether or not the vendor has prepared for them — what happens to the data when the contract ends, and what happens to the service if the company does not survive. A business model that depends on retaining data belonging to the citizen has a problem to solve before it grows, not before its next meeting.

So what: governance is the scaling mechanism, not the brake. Write day ninety-one before day one, or the pilot is a long demo.

Governance: the independent measurement clause

One control, written into the funding instrument: the baseline and the post-period measure are computed from the administration’s system of record by a party that is neither the vendor nor the sponsoring directorate. Owner: the programme’s national executing agent — the academic partner where one exists, which is precisely what neutrality buys. Attach the human-review rule in the same clause: the machine proposes, a person decides, the decision is attributed and appealable, and any digital channel keeps a non-digital one at equal priority. Exclusion is an indicator, not a footnote.

KPIs: five numbers, each with a stated before

Baseline coverage — share of funded pilots holding a pre-measured before-number from the administration’s own system; cohort baseline at intake was close to zero, target 100% before funds release. Day-91 continuity — share with post, budget line and instrument written before day one; target 100% at approval, realised continuation measured at day 180. Time to first renewable public contract, in months. Non-digital channel share, measured and published, never suppressed. Stop-criterion exercise rate — the share of pilots closed on their own stop criterion; a portfolio where this is zero is not measuring, it is advocating.

Roadmap: twelve months from sentence to second city

Q1 — The sentence and the custodian. Every team writes one sentence: for whom, what improves, measured how, in a ninety-day pilot with which named post, and how it scales after. Send the five-line baseline request; name the data custodian. Q2 — Instrument before pilot. Legal and procurement in the first meeting, not the last; the contractual route to permanence designed while the pilot is still being scoped. Q3 — Run two, measure independently. Fund the pilots whose day ninety-one is written; the executing agent computes the result. Q4 — Replicate or close. Publish the measured result, failures included, then take the instrument — not the demo — to a second municipality. That is from pilot to policy, and interoperability or it doesn’t scale applies to the institutions as much as to the systems.

Socradata Perspective

A city is an operational system with voters attached

Nothing in the three patterns above is unfamiliar to anyone who has put a decision system into an enterprise. A municipal case-management system is a system of record with a queue in front of it; a route registry published to a standard is master data with a consumer; a purchasing platform is a procurement module with an audit obligation. The loops have the same anatomy as those inside an ERP or a WMS: an input of uncertain quality, a model that proposes, a human who disposes, a record that must survive the disposal. What differs is the override path, which must be exercisable by someone with no account, no application and no connectivity, and the evidentiary standard, which is higher because the counterparty is a comptroller rather than a board.

The failure mode transfers too. Enterprises start where the data is already clean and call the result a roadmap; administrations buy the fourth floor — the chatbot, the camera, the dashboard — without the three beneath it or the operating model above it, and call the result a smart city. The correction is the same in both settings and it is unglamorous: order the records, name the owner, write the number before the system is procured. Buenos Aires’ QuarkID is instructive for its sequence — legitimacy, consent and revocation were settled before the application was built, which is why it became infrastructure other services can stand on rather than a well-reviewed pilot. Do not show me the dashboard; show me the decision you took yesterday because of it.

Socradata transforms ERP, WMS and supply-chain data into predictive intelligence and governed operational decision systems. A city is one more complex operational system, with a harder override requirement and a more demanding standard of proof — which makes it the best teacher for the governance a private operator will eventually face.

Find out whether your pilot has a day ninety-one

Every Wednesday, The Operational AI Dispatch translates one consequential AI signal into an operating model, a KPI set and an action plan for leaders running enterprise operations, ERP, WMS, supply chains and public systems. Published weekly by Socradata.