Dili's $15M: when review gets cheap, stop sampling

Every business spot-checks something. You sample invoices, audit a handful of timesheets, review a slice of the expense reports — not because sampling is correct, but because checking everything costs more than the errors you'd catch. Dili just raised $21.7M, including a $15M Series A led by Khosla Ventures, on the premise that compliance automation flips that math.

What actually happened

Dili sells to construction, energy, infrastructure, and advanced manufacturing — industries that live under prevailing wage law. The product monitors prevailing wage and apprenticeship compliance, reviews certified payroll, handles Davis-Bacon requirements, and produces audit-ready reporting. The pitch isn't faster review. It's that the system checks 100% of project data in real time rather than sampling it.

The round included Y Combinator's Garry Tan, Allianz, Brick and Mortar Ventures' Darren Bechtel, and Rebel Fund. Dili says it has processed $1.4 billion in wages, 5.2 million labor hours, and nearly 16,000 certified payroll reports across more than 700 projects, and that it cuts weekly review from seven-plus hours to under five minutes per organization. The company also claims it has prevented over $50 million in fines and clawbacks across its customer base — a vendor figure, unaudited, so weight it accordingly. CEO Anand Chaturvedi's framing is the part worth stealing: "We look at everything, every report, every wage determination, every week."

Why it matters for your business

Sampling is a cost-of-review decision that hardened into a policy. Somebody once decided reviewing 10% was affordable and 100% wasn't, and that ratio has quietly survived every change in what review actually costs. When the per-item cost of checking drops far enough, the sample stops being prudent and starts being the reason errors survive — because the 90% you skipped is where they are.

The generalizable move here has nothing to do with construction. Find the review in your business that runs on a sample: vendor invoices against contract terms, timesheets against job codes, sales orders against pricing rules, returns against policy. Then ask a specific question — not "can AI do this," but what does it cost to check one item, and what does it cost to miss one? If the miss is a clawback, a chargeback, or a regulator, the answer usually stops being close.

One structural note, because it's where these projects fail: Dili's numbers come from checking against a determination — a wage rate that exists as a published fact. That's what makes full-population review tractable. Automating a review that has no ground truth to check against just produces confident opinions faster. Before you build, find the authority the check resolves to. If there isn't one, you have a judgment problem, not an automation problem.

Sources: Dili press release via GlobeNewswire, FinSMEs.