Read now
The last wave of AI was built on the internet. Every frontier model was trained on roughly the same scraped, public corpus, and the value of having done so is now largely priced in.
The data that matters next was never online. It sits in the file cabinets, PDFs and inboxes of the physical economy, in industries the internet's crawlers never reached. Construction is the largest of them. It is a $2 trillion market that runs almost entirely on information no model has ever seen, and the software it relies on cannot read the little of that information that has been digitised.
That's why we invested in Cascade's pre-seed round.
We spend a lot of time on where AI value accrues once the foundation models commoditise, which they are doing fast. The durable advantage is moving to whoever can get inside an industry whose proprietary data has never been scraped, structure it, and build a system that compounds with every customer. Construction, fragmented, analogue and enormous, is close to the perfect case. Cascade is building that system.
Cascade is an AI coordination platform for architecture, engineering and construction firms. It helps them do the three things the industry is worst at doing systematically: find the right work, win it, and deliver it. It reads the messy, unstructured signals that sit ahead of a formal bid, scores each opportunity for how well a given firm is placed to win it, and turns a process that has always run on relationships and memory into something repeatable.
Underneath sits a proprietary data layer that learns from every pursuit, every win and every loss, so the system sharpens the more it is used. The pricing is aligned to that outcome: a subscription plus a success fee when a firm actually wins the work. The result is a shadow market of projects most firms never knew existed, made visible and winnable. Early customers are closing bids in one to three weeks against a six-month industry norm, with no churn.
The team. Hannia Zia (CEO) and Joana Ferreira (CTO) have built together for years, out of Google, UnlikelyAI and Freetrade. Hannia is a formidable commercial operator who built a multi-million ARR line at her last company. Joana is one of the strongest technical founders we have met, a view held just as firmly by the others who backed this round. They have already pulled senior talent out of the industry's incumbent. Commercial hustle and deep technical range, pointed at an old and hard market, is a rare thing to find.
The thesis fit. Cascade sits within our economic empowerment theme. Business development in construction has always rewarded the firms with the deepest relationships and the longest institutional memory, which means opportunity compounds for those who already have it. Cascade gives the millions of small and medium AEC firms a systematic way to see and win work on merit rather than on their contact book. And when the right firms win the right projects, the affordable housing, schools, hospitals and public infrastructure those projects represent get built better. That is Ada's whole idea: back the best, not the best-connected.
Where value accrues. In vertical AI, the model is not the moat. The moat is the proprietary data no competitor can buy and the network effects that grow with every firm on the platform. Cascade owns both: a continuously learning data layer built from information that exists nowhere else, and a network that becomes more useful to each firm as more firms, and more teaming relationships, join it. A foundation model can write the code. It cannot see this data, and it cannot rebuild this network.
The macro backdrop is unusually favourable. Record public infrastructure spending, hundreds of billions still to deploy from the IIJA, and a data-centre construction boom are all landing on an industry whose core processes remain analogue. Cascade's ambition runs the full arc of a project, from the first faint sign that work is forming through to delivery. If it works, it becomes the coordination layer for a $2 trillion industry, and one of the clearest examples of AI creating value in precisely the place the foundation models cannot reach.
The internet's data has been picked clean. The next decade of AI value is in the industries it never touched, and few are larger or more overdue than construction. The hard part was never building a model. It was getting inside an industry this analogue and earning the data. That is what Hannia and Joana are doing, and we're thrilled to back them.
"The last generation of category-defining companies in finance were not the ones with the cleverest algorithms. They were the ones that got inside the industry, owned the data everyone else needed, and became impossible to unplug. Construction is a larger, messier and more valuable version of that opportunity, and almost none of its data has ever been online. Hannia and Joana have the commercial and technical range to go and get it, and to build the coordination layer the whole sector ends up running on. That is why we wanted to be early."
— Matt Penneycard, Founding Partner, Ada Ventures