Dreamdata built the finance team they think is coming. It runs on Light.

The Challenge

The old stack wasn't built for the finance team they wanted

Before Light, Dreamdata ran three separate general ledgers in e-conomic. Each entity had its own setup, and consolidation happened manually every month.

The charts of accounts overlapped in some places and diverged in others, so closing meant moving data between systems and spreadsheets.

The bigger problem was the gap between finance and the rest of the business. Sales would sign a contract, HubSpot would show the ARR, then finance would open another system and recreate the transaction, right down to manually typing the customer's name and address.

The Solution

They weren't looking for another ERP

Dreamdata went to market looking for the finance systems being rebuilt around AI. Peter wasn't trying to shave a few hours off bookkeeping, he wanted infrastructure his team could build on. For him the direction was obvious. The question was whether to move now or wait. They moved.

The migration ran through Sunrise, Light's migration platform. Dreamdata brought across its opening balances from 31 December 2025 and every transaction from 2026, line by line. Peter didn't want January living in one system and July in another, especially in an audited year while the company was raising.

The Story

Before they built on Light, they built alongside it

When Andreas joined, he set out to automate as much repetitive finance work as possible. Six weeks before Dreamdata went live on Light, while the migration was still running, he started building against the API. Using Claude, he built what became Dreamdata's Finance Hub.

It reconciles every account in the ledger. It runs deferred revenue waterfalls. It tracks payroll liabilities and receivables. It accrues expenses that haven't cleared approval by month end. And it doesn't blindly accept what's in the ledger: it reads the underlying data and applies Dreamdata's own accounting logic. When Light coded a Microsoft charge as software, for example, the Finance Hub read the receipt and moved it to paid advertising.

Then they kept building. A bill agent reviews coded transactions, fixes anything it disagrees with, adds deferred schedules, splits transactions when VAT treatment differs, and confirms the result back in Light. A receipt agent chases missing receipts through Slack. Peter built a Mercury bank integration on Light's API in an evening, then asked Light to take it over. The result is a ledger that's already been reviewed before month end starts.

Now they're building on top

Dreamdata initially built many of its agents adjacent to Light, using Light's API and data as the foundation. The next step is bringing more of that work into the platform itself.

Peter made the case to our product team in July. Light has to build for thousands of companies, Dreamdata only has to build for Dreamdata. That means its team can spend more compute, more context and more tokens solving a very specific finance task, because the alternative isn't cheaper software, it's another hire.

It's also the direction we were already taking Light. Custom Agents landed in Dreamdata's account in August, giving the team a way to bring more of the workflows it prototyped outside Light directly into the platform, while keeping Dreamdata-specific business logic where it belongs.

Built with them, not just for them

Dreamdata's Copenhagen office is ten minutes from ours, and that proximity has turned into a pretty direct product loop. Kyle and the product team have run workshops with Peter and Andreas in their office, and Andreas has spent hours stack-ranking card feedback with the team. Features have gone from conversation to product in weeks.

The cash balances endpoint exists because Peter asked for it. The AR aging view came out of the same sessions, replacing analysis Dreamdata used to assemble manually.

The relationship goes both ways. Dreamdata builds things on Light that we learn from, and our product team uses those workflows as input for how Light's own AI layer should work. They're not waiting for us to tell them what AI in finance looks like. They're building their version of it, and pushing us to build faster too.

Finance lives in Slack

For most of Dreamdata, finance doesn't happen in a finance system. It happens in Slack. Managers approve bills there, often within minutes. Light pulls receipts from the inbox, and if one is missing, the cardholder gets asked for exactly what finance needs, in the place they're already working. Agents post their results there too.

What Dreamdata runs on Light

• Three entities, consolidated from day one.
• A one-day month-end close, down from 20.
• Agents handling bill coding, VAT, transaction splits, deferred schedules and receipt collection.
• Dreamdata's Finance Hub, built on Light's API, reconciling every account in the ledger.
• AR and invoicing workflows connecting commercial data to finance.
• Bill approvals, reimbursements and receipt requests through Slack.
• Payments to Jyske Bank over a host-to-host connection.
• App Store integrations, including the live Zenegy payroll app.
• Full 2026 transaction history migrated through Sunrise across 13 bank accounts and four currencies.

The Result

Three people, backed by agents

Dreamdata's finance team is still three people. They're running three entities across two continents, while ARR has more than doubled. Peter estimates the same operation would need eight people without Light.

The outsourced bookkeeping work that once amounted to 1.5 FTE is gone, agents handle it instead, and month-end has fallen from 20 days to one.

But the bigger change is what those three people spend their time doing. With clean, structured data underneath them, finance can answer questions that used to get buried under bookkeeping: where should Dreamdata invest, should it open an office in a new region, how aggressively can it hire, did the last sales cohort ramp well enough to justify the next one. The repetitive work moves to agents. The judgement stays with the team.

That's the finance function Peter and Andreas set out to build. First they built around the limitations of their software, then they built alongside Light. Now they're increasingly building on top of it. And Peter doesn't see that shift as optional.