Due diligence · Corporate, institutional, government
Follow the incentive, then read the contract
Institutions rarely fail children through malice. They fail through procurement cycles that reward announcement over outcome, and governance that disperses accountability until no one holds it.
The structural problem
A superintendent has three to five years in post; an educational effect takes longer than that to establish. A vendor's revenue arrives at signature, and its risk arrives years later, usually at someone else's desk. A foundation is judged on deployment of capital, not on the durability of what the capital built. A legislature funds devices because devices can be counted before the next election.
None of these people is behaving badly. Each is responding rationally to the horizon they are measured on. Diligence in institutional settings is therefore mostly the work of finding where the time horizons diverge and writing contract terms that close the gap — continuity obligations, escrowed data, published outcomes, clawbacks, and an exit that does not depend on the vendor's goodwill.
The pace problem
The same argument is being had inside large firms right now. Consultancies and employers — McKinsey and Starbucks among the names we have heard working through it in public — are describing the hard work of due diligence as the thing that has to be done before AI can be used well. What comes through in those discussions is not certainty. It is that corporations, like individuals, seem genuinely unsure what will work.
That uncertainty is not a failure of intelligence. It is a structural mismatch. Technology does not evolve backwards, and it can arrive very quickly. Human society works slowly: consent, law, professional norms, training, and trust all move at the speed of people changing their minds. The gap between those two rates is where most institutional harm happens — not because anyone chose harm, but because the decision arrived before the judgement did.
An organisation cannot slow the technology down. It can decide, in advance and in writing, which decisions it will not make at the speed of the technology. For an academy working with five- to seven-year-olds, that list is long, and it is the reason this section exists before the first cohort does.
Observation — not a finding
What to demand before money moves
- A named accountable individual, not a committee. If a programme fails, the public should know whose judgement is being reviewed.
- Published success and failure criteria, fixed before the contract is signed and reported against on a schedule.
- Total cost of ownership: training, substitution, support, refresh, and the staff time the programme consumes from other work.
- Independent evaluation funded in the contract but not controlled by either party, with a right to publish unfavourable results.
- Data escrow and continuity obligations that survive vendor insolvency or acquisition.
- Conflict-of-interest disclosure covering board seats, equity, advisory fees, and prior employment on both sides of the table.
- Procurement that is genuinely competitive, including a documented answer to why an existing free or cheaper option was rejected.
- A stated sunset: the date on which the programme ends unless renewed against published evidence.
Named cases
Failure — procurement
LAUSD iPads: the process, not the device
The roughly $1 billion tablet initiative cancelled in 2014 drew scrutiny not only for implementation but for how the contract was arrived at, including contacts between district leadership and vendors before the competitive process. The superintendent subsequently resigned.
The institutional lesson stands independent of any individual's conduct: when the shape of a solution is settled before the procurement opens, the procurement is theatre and the diligence is retrospective.
Failure — legitimacy
inBloom and the missing constituency
A $100 million foundation-funded student-data infrastructure closed in 2014 under parent and advocate opposition. States withdrew one after another. The organisation had contracts with education agencies and the support of major funders, and no relationship at all with the families whose children's records constituted the asset.
Institutional consent is not the same as public consent. Where children are involved, the second is the binding one, and it cannot be obtained retroactively.
Regulation — the floor is rising
COPPA amendments (2025)
The FTC issued final amendments to the Children's Online Privacy Protection Rule in April 2025, tightening obligations around consent, retention, disclosure to third parties, and data security for services directed to children.
Compliance is a floor, not a standard. But an organisation that treats the floor as an aspiration is telling you where its ceiling is.
Regulation — classification matters
EU AI Act, Annex III
The European Union's AI Act classifies certain education and vocational-training uses of AI — including systems that determine access, evaluate learning outcomes, or monitor prohibited behaviour during assessment — as high-risk, carrying obligations for risk management, data governance, documentation, human oversight, and transparency.
We do not operate in the EU. We use Annex III as a checklist anyway, because it is the most explicit public statement anyone has written about which educational AI uses warrant heightened scrutiny.
What we accept in advance
Commitment — binding on us
Commitment — binding on us
Commitment — binding on us
Commitment — binding on us
Related: the sponsor case, transparency, and diligence in technology.
Sources
- 1.Los Angeles Times, “LAUSD abruptly cancels $1-billion iPad program” (25 August 2014)
- 2.Education Week, “inBloom to Shut Down Amid Growing Data-Privacy Concerns” (21 April 2014)
- 3.New York Times, “A Student-Data Collector Drops Out” (26 April 2014)
- 4.Federal Trade Commission, Children’s Online Privacy Protection Rule — final amendments, 90 Fed. Reg. 16918 (22 April 2025)
- 5.EU Artificial Intelligence Act, Annex III (high-risk AI systems)
- 6.Education Week, “Los Angeles Unified’s AI Meltdown: 5 Ways Districts Can Avoid the Same Mistakes” (8 July 2024)