The target, in plain numbers
Monsha'at counts about 1.3 million active enterprises in Saudi Arabia, against 1.7 million commercial registers, employing roughly 8.9 million people. SMEs contribute 22.9% of GDP, and Vision 2030's stated target is 35%. Consumer spending reached SAR 1.41 trillion in 2024, up 7% on the year. The demand side of this economy is not the problem.
To move a 22.9% share to 35%, existing SMEs have to grow and new ones have to formalise, and both of those things create the same kind of work: invoices, reconciliations, stock counts, VAT filings, supplier follow-ups. Growth in revenue is growth in paperwork, and paperwork is done by a back office.
Where the hours actually go
The back office is the part of an SME that does not scale with ambition. A restaurant group that doubles its locations roughly doubles its invoice volume, its reconciliation load, and its stock movements, but its accounting team rarely doubles with it. The result is familiar to anyone who runs one: month-end closes that slip, costs that drift between one invoice and the next, and decisions made on numbers that are three weeks old.
Meanwhile the data itself has quietly become machine-readable. Cloud accounting and ERP adoption in the Kingdom is no longer niche: Qoyod reports more than 25,000 active businesses, Wafeq more than 10,000 customers processing around 2 million invoices a month, Rewaa more than 7,000 merchants, and Foodics more than 40,000 branches across the region. ZATCA's e-invoicing waves push the same direction for everyone else. The records an AI system would need are, for the first time, actually in systems.
What the research says about AI and this work
Anthropic's Economic Index rates business and finance occupations at 94% theoretical task coverage by current AI, and office and administrative support at 90%, while observed real usage sits far lower. That gap between what is technically possible and what is actually adopted is exactly where Saudi back offices are today. Gartner, for its part, expects 40% of agentic AI projects to be cancelled by the end of 2027, mostly for failing to show measured outcomes. Both numbers matter. The first says the work is addressable. The second says attempts to address it will be judged, correctly, on proof.
Where we stand
RoleAxis builds governed AI agents for exactly this layer: today, three agents, covering inventory and purchasing, payables and tax, cash at the tills, the month-end close and the payroll file, reading the ERP and accounting systems a business already runs. Every action an agent drafts waits for a named person to approve it before anything executes. We publish what is live and what is still in development on every page of this site, because the fastest way to lose the argument above is to overclaim it.
RoleAxis is a private company. We are not affiliated with, or endorsed by, the Vision 2030 program or any government body. The targets are the Kingdom's; the analysis here is ours.
Sources
- Qoyod, Saudi SMEs 2026: numbers and growth (Monsha'at figures)
- Knight Frank, The Saudi Report 2025, retail and F&B
- Euronews on the Anthropic Economic Index, March 2026
- Software Strategies Blog, roundup of agentic AI forecasts 2026 (Gartner figures)
More from Insights: all posts. Related: Not every back-office task deserves an AI agent.