MSc Thesis · Oxford Internet Institute · In progress · Submission Jun 2026

From Pilot to Policy:
how OMB memoranda shifted federal AI adoption.

Procurement memos changed how often agencies signed AI contracts.
They didn't change how much the government spent.

Method
Interrupted time series · agency-month panel · fixed effects
Source
USAspending.gov · 2018–2025
Treatments
OMB M-24-10 (Mar 2024) · M-24-18 (Sept 2024)
n
~7,200 agency-months
Tooling
R · Stata · fixest · ggplot2 · custom ETL
LIVE Federal AI procurement: actions vs. spending, 2018–2025
Period
Hover the chart
AI contract actions
AI obligations ($M)
AI contract actions (count, indexed)
AI obligations ($M, indexed)
OMB intervention dates
FINDING

After the OMB memos, the frequency of AI procurement actions jumped sharply — agencies signed more, smaller contracts. Total dollar obligations did not move correspondingly. The policy produced an extensive-margin response, suggesting administrative AI guidance shifts signalling and process faster than it shifts budget.

01 — Question

What does federal AI procurement guidance actually do?

The OMB memoranda M-24-10 (March 2024) and M-24-18 (September 2024) set out detailed expectations for how federal agencies were to acquire, manage, and govern AI systems. The policy framing treats them as enabling: clarifying procurement pathways, removing friction, encouraging responsible adoption.

But what happens in the contract data? Do agencies sign more AI contracts? Bigger ones? Different vendors? Or does administrative guidance produce signalling without substance — language that changes faster than the spending lines underneath it?

This thesis treats the OMB memos as a quasi-experimental intervention and tests their measurable effect on agency-level AI procurement behaviour.

02 — Method

Interrupted time series, agency-month panel.

The unit of analysis is the agency-month. AI-related contract actions are identified using a combination of NAICS code filtering, PSC code filtering, and keyword matching against the contract description field — calibrated against a hand-coded validation sample.

Specification
ITS with agency & month-of-year fixed effects (fixest)
Outcome 1
AI contract action count (indexed pre-period mean = 100)
Outcome 2
AI obligation $ (indexed pre-period mean = 100)
Treatment
Bundled OMB intervention indicator (M-24-10 + M-24-18)
Controls
log(total agency procurement); time-varying agency size
Falsification
Placebo timing tests at false intervention dates
Robustness
Alternative AI identifiers; alternative bandwidths

The chart above visualises the headline finding: a clean level shift in action frequency at the M-24-10 boundary, with a further bump after M-24-18, and no comparable shift in spending over the same period.

03 — Implication

What this means for how we read AI policy.

If administrative AI guidance produces an extensive-margin response — more contracts, not bigger ones — then procurement memos work differently than they're often described. They function as a coordination signal that lowers the friction of starting AI procurement, without the budget reallocation needed to scale.

That has practical implications for anyone reading federal AI activity from the outside. A surge in AI contract counts after a guidance memo is not the same as a surge in AI investment. The two need to be measured separately. A separate set of policy instruments would be needed to shift dollar volume.

It also has implications for the broader argument that procurement is where governance actually happens. Procurement guidance moves the visible surface of AI activity quickly. Whether that surface activity is converting into capacity is a different question — and one this thesis treats as open.