Most pilots stall because the AI can't reach the data and systems where the work happens, nobody can measure the result in money, and governance arrives after the demo instead of before it. Start with one back-office workflow, connect it to real systems, set the rules first, and measure it against today's cost.
- 95% of organisations saw no measurable P&L return from generative AI pilots.MIT Project NANDA, The GenAI Divide: State of AI in Business 2025 (via Computing)
- Back-office automation returned more than the sales and marketing tools that received about half of budgets.MIT Project NANDA, The GenAI Divide: State of AI in Business 2025 (via Computing)
- Only 16% of GCC CEOs say their most-used AI tools can reach all the relevant documents and data.PwC 29th Global CEO Survey, Middle East findings (via Middle East AI News)
- 79% of finance leaders had AI cost overruns in the past 12 months.DoiT survey by Sapio Research, Feb 2026 (via Insentra)
What the research says
MIT's Project NANDA found 95% of organisations saw no measurable profit-and-loss return from their generative AI pilots, and that back-office automation outperformed the sales and marketing tools that took most of the budget. In the Gulf, PwC found only 16% of CEOs say their AI can reach all the relevant data.
The four causes we see
The AI can't reach the data. There is no number to beat. Governance arrives late, so security blocks go-live. And the pilot sits beside the work instead of inside the system where the work happens.
How to pick a first workflow
Choose high-volume, rule-heavy back-office work: vendor onboarding, document checks, invoice queries. Measure today's cost and time. Connect the agent to the real systems. Agree the approvals with security before you build.
What a production path looks like
Readiness to choose the workflow, governance to set the rules, a proof of value on the real systems, then production. That is the sequence of our three services, and an enterprise build across SAP Ariba, SAP ERP and an on-prem ERP is underway in Saudi Arabia on this path.
Questions people also ask
How long should a proof of value take?
Do we need a data lake first?
Where should an Australian enterprise start?
One governed layer across every system you own.
Deploy where you decide, connect to SAP and your local systems, and make every agent action approved and logged. See how the pieces fit.
