Transforming Shell’s tendering process

Tendering ran on Word documents and email. Every submission was rebuilt from scratch, compliance was checked at the end rather than along the way, and the knowledge that made a good decision — past contracts, supplier performance, market timing — lived in people’s heads.

The cost wasn’t only time. Teams were making million-dollar commitments without a clear view of the alternatives.

The opportunity wasn’t automation for its own sake. It was moving the evidence to the moment of decision — so a route is chosen on what the business already knows, not on who happens to know the category best.

That set the rule for where AI sits. It removes effort, not authority: drafting first passes, checking policy continuously, always showing the reasoning behind a recommendation. Every suggestion is the buyer’s to accept, reject or override, and the trail of how a decision was reached survives to Tender Committee.

A north-star prototype made that arguable rather than theoretical — aligning procurement, engineering and legal around one journey, and securing funding for the build.

At Shell the blocker wasn’t effort, it was certainty. People were quietly double-checking with whoever sat nearest, because the system gave them no way to be sure. That reframed the product from faster paperwork to defensible decisions — and set the rule for the AI: it removes effort, never authority.

I design AI into enterprise work by making it legible. Evidence visible, recommendations overridable, and a trail somebody can defend at committee months later. If you’re introducing AI into high-value enterprise workflows, this is what helps teams trust and adopt it rather than bypass it.