Saudi Arabia just set an 8% efficiency minimum for every AI deployment

The Cabinet has declared 2026 the Year of AI for Saudi Arabia. Inside the KSA AI Adoption Framework 2025, issued by the Saudi Data and Artificial Intelligence Authority (the kingdom's national authority for AI policy and regulation), sits a single number that will set the minimum for every AI procurement in the Kingdom for years. 8%. That is the operational efficiency improvement an AI deployment has to deliver to count as a success under the framework's Outcomes pillar. Below that number, the deployment is a failure in the formal sense, regardless of how the demo went or how thick the steering committee deck became.

The number does the work because of how the framework requires it to be measured. The improvement has to be observed in production, mapped to enterprise key performance indicators the institution already runs, traced through lifecycle documentation, and produced inside an availability standard of fewer than thirty minutes of downtime per year. There is no place to put a pilot result. There is no place to put a model evaluation score. There is no place to put a slide that says peer benchmarks suggest. SDAIA will look at the dashboard, on the bank's own systems, after handover. That is the audit.

The reassuring reading of this minimum is that it is calibrated about right. Industry benchmarks from the major hyperscalers and the global advisory firms routinely report double-digit gains in throughput, processing time, or unit cost across back-office, contact-centre, and underwriting workflows. The conclusion, on this reading, is that 8% is high enough to discipline the market and exclude vanity pilots, and low enough that competent delivery teams using current foundation models should comfortably clear it. The framework formalises good practice. Vendors with strong AI credentials will continue to win. Those without will lose. The buying institutions get sharper procurement language. The Big 4 and global SI category, on this reading, has nothing structural to fear: their analytics groups already publish benchmarks at or above the line, their pricing can absorb a performance clause, and their offshore delivery teams know how to hit a productivity baseline.

The reading is comfortable, and it misses where the 8% is measured. Audit-defensible production efficiency, on the bank's own dashboard, after handover, is a measurement the legacy consulting model is not priced to deliver. A standard global SI engagement is shaped around discovery, design, build, and rollout phases, each priced and measured by the firm's own milestone deliverables. Efficiency is asserted in the business case at the beginning and rarely re-measured against that baseline at the end. The framework changes the order of operations. The 8% has to be defensible after handover, on the client's systems, with SDAIA reading the dashboard. The integrator's milestones are tied to the project plan. The procurement clause is tied to the institution's KPI. The two systems point at different things, and the integrator's incentive structure does not point at the institution's KPI dashboard after handover. Nobody on the vendor side is paid for the audit. Nobody attends the meeting. The framework asks the institution to discover this gap on its own time. By that point the integrator has billed and rotated. The strategy firm that wrote the business case left twelve months earlier. Nobody in the room owns the number.

The practical move for a procurement lead inside the Kingdom is narrower than it sounds. Read every signed AI engagement against two questions. Can it produce a defensible 8% on the institution's own KPIs at the next audit cycle. If not, what does it take to re-baseline. Then look at the next engagement on the desk. Is it being signed with a vendor whose delivery model can underwrite the minimum, or with a vendor whose delivery model is structured to escape it. The contractual artefacts that matter are the baseline measurement protocol, the post-handover audit schedule, and the named individuals who remain on the ground after go-live. A senior practitioner named on a single engagement is not a senior practitioner billed across six clients with the obligation to attend a weekly check-in. Time commitments should be expressed in days per month, not in percentages of allocation. The vendor that accepts this clause is signalling a different operating model. A clause that names a vendor's senior practitioner as personally accountable for the production efficiency case, with consequences for missing the minimum, is the simplest test of whether the delivery model is real. If the vendor cannot accept that clause, the model has answered the question for you.