Deepki is the most trusted sustainability solution for real estate. Its comprehensive platform enables real estate owners, investors, and occupants to manage risk through data-driven insights, improve the sustainability and financial performance of their assets, and meet growing investor expectations and regulatory requirements.
Deepki's customers don't care about portal mechanics; they care about actionable, trustworthy, audit-ready data to make strategic decisions. By pairing Deepki's in-house strengths with Deck's AI Agents for the long tail, the team scales coverage, maintains quality, and brings to market an unmistakably AI-enabled product experience.
The result is faster time-to-live on complex sources, stronger coverage in difficult regions, and a product story that feels unmistakably AI-enabled — all without pulling engineers off the roadmap.
Challenge
Deepki monitors the sustainability performance of over €4 trillion in real estate assets under management (AUM) through its platform across 80+ countries. Historically, their teams built and maintained their own scrapers to protect quality and control. As the company accelerates expansion into new markets, three pressures have intensified:
- Breadth vs. bandwidth: New sources span regions and provider types; onboarding internally could take months and required recurring maintenance whenever portals changed.
- Hard access problems: Many portals deploy CAPTCHAs or anti-bot defenses that stall traditional approaches.
- Decision-grade bar: Any solution had to preserve auditability and reliability because the data informs high-stakes portfolio decisions.
"Keeping up would have meant doubling headcount every year — not viable — so we began evaluating partners with proven ways to fetch bills from difficult portals."
Tony Sarkis, Senior Operations Manager, Deepki
Build vs. Partner: A Data-Driven Vetting
Deepki compared build vs. partner on key factors:
- Man-days per connector (by complexity tier), plus recurring maintenance windows.
- Annual document volumes and per-document economics to identify a long-tail threshold where partnering wins.
- Proof-of-concept outcomes on difficult portals (including CAPTCHA-heavy regions) and measured time-to-live from credentials to first data.
Conclusion: Keep important, high-volume pipelines — where Deepki's internal expertise shines — in-house; outsource the long tail, where breadth, speed, and portal hostility make Deck the smartest choice.
Solution
Deck's AI Agents take on the messy, portal-level work so Deepki can focus on product and analytics. In practice:
- Long-tail acceleration: New, low-to-medium-volume sources moved from months to days to go live — meeting customer timelines without continuously scaling headcount.
- CAPTCHA capability: Agents reliably navigate CAPTCHA-protected portals, unlocking previously blocked sources.
- Hybrid by design: Deepki integrates once and routes target sources to Deck where fit is strongest, while core, high-volume connectors remain internal.
Results
Speed where it matters
The time-to-live for complex, niche sources dropped to days, helping Deepki deliver coverage fast without derailing its roadmap — or having to grow its headcount.
Competitive in difficult markets
Consistent access to CAPTCHA-heavy portals turned formerly blocked sources into standard coverage, strengthening win rates where "can you actually fetch this data?" is the deciding question.
Economic clarity
A clear outsourcing threshold emerged. Deck is the most efficient choice for the long tail (~300 sources), while Deepki keeps very high-volume pipes in-house — aligning per-document economics with the reality of ~6M invoices per year.
Collaboration
The accessibility and hands-on support of Deck's team stood out for Deepki. The relationship started with a learning curve on both sides — new APIs, evolving documentation, and teams getting used to working together. Once a weekly cadence and a shared channel were in place, momentum shifted: integration complexities were surfaced more quickly, the right experts jumped in fast, and source activations became predictable.