Just iterate, with uninterruptible chunks of billable time, and keep getting customer feedback on each concrete visual thing you can show them.
If it's inefficient, and might miss big-picture needs/opportunities/coherence, but you're making revenue on that inefficiency, promptly and recurring.
And the customer's many stakeholders are probably happier than if you'd tried to do project management. They don't have to think hard upfront, nor coordinate too much, nor commit to risky decisions. They just have a meeting every week or two, to approve/reject/bikeshed.
Unfortunately, sometimes this methodology is applied to projects for which it's not suited as well, such as for a company that needs a solid product and business model. (Not VC-growth-oriented startup that hires a ton of people, for whom sprint tasks are a "proof of work" productivity, which is a relief from the organization's hard coordination problems.)
And I get the impression that a lot of people in tech right now have never been on a team that exposed them to project management other than creating and doing individual sprint tasks.