One number you can defend, from every system you run.
Built for hotel management companies first, rokda reconciles the property and accounting systems across every hotel, holds each metric to one approved definition, and traces every number to the line it came from. Ask why a number moved, and AI analysts lay out the reasons, the evidence and how sure they are. You make the call.
Talk to usThe problem
Most organizations can report a number. Few can defend it.
A company that runs many sites runs many systems. The same account, supplier or product carries a different code at each site, and the same supplier shows up as several records. Each owner or partner reads the results against their own contract.
So when someone asks why a number moved, the answer gets assembled by hand. The finance team spends days in spreadsheets, and two people can still come back with two different numbers. Dashboards show what happened. The why arrives late, or not at all, and the person who has to answer for it is on their own.
What rokda does
One platform, from your source systems to the decision. Each part is its own component.
Connects.
Pulls from your systems of record and your files, on a schedule you can reason about.
Reconciles.
Maps every site's codes, suppliers and records to one master list. Clear matches are applied; uncertain ones wait for a person to confirm.
Defines.
A definitions registry holds one approved definition for every metric, never redefined page to page. It is built so your exceptions sit on top of the standard, versioned, while the standard stays intact.
Traces.
The dashboards and reports you decide from, every number drillable to the line it came from.
Explains.
Ask why a number moved. Several AI analysts work it through and show you the case. You make the call.
A new industry brings its own codes and definitions to the same platform. It does not need a rebuild.
Reconciliation you can check
Reconciling codes across systems is the hard part, so rokda is built to show its work.
Rules per code set. Ledger accounts, suppliers and units of measure each get their own matching rules, set and checked separately.
A person decides the uncertain matches. Clear matches are applied. Anything in between goes to a review queue. AI can put that queue in order; it does not make the decision.
Nothing is overwritten. Every mapping keeps its history. Each decision records the rules it was made under, and each human decision records who made it and when.
Rules change on evidence. A change to the matching rules takes effect only after it does at least as well as the current rules on the same fixed set of checked answers.
How it thinks
Question the question.
Before any analysis, rokda asks you what you are deciding, what is at stake, what is ambiguous, and what is missing. This framing, not the raw question, drives the work.
Weigh it from several sides.
Independent analytical positions work the problem, each through its own lens, grounded in your reconciled, governed data.
Review blind, then revise.
Each position reviews the others without knowing whose work it is, then revises. Every position is recorded before and after review, so you can see who moved, and on what evidence.
Frame the case for you.
You get the case, including where the positions still disagree. You make the call.
Why this is hard to copy
Runs on Databricks. Databricks is the engine: storage, governance and serving. It doesn't ship your industry's chart of accounts, or which contract defines profit. That layer is ours. When Databricks releases something general, we adopt it rather than rebuild it.
The model isn't the moat. Put AI on top of data that doesn't agree and you get a confident wrong answer, fast. We use several models and swap them as they improve, so better AI makes it better, not obsolete.
The work underneath compounds. Your approved definitions, your mappings and the decisions behind them build up every month you run on rokda. That record is what makes the next answer faster to defend.
Your data: where it lives, where it goes, how it's protected
Where it lives. Your analytics data lives in a dedicated environment in a US region, never pooled with another client's. Support requests are kept in your own project with our ticketing provider and deleted when you leave.
Where it goes. To answer a question, the reasoning layer sends the governed figures it needs to AI models. Every model is classified as running inside your environment or outside it, and what your screen says about where your data went is derived from the models that actually took part. If one runs outside, the statement says so. Anything unclassified counts as outside. Clients who need analysis to stay inside their own environment can hold it to models that qualify; the statement strengthens only when every one does.
How it's protected. Encrypted in transit and at rest, backups included. Passwordless sign-in, so there is no password store to breach. Access by invitation, revocable, checked on every request. Your data comes back to you when you leave. Our security overview lists what we have and, just as plainly, what we don't have yet.
First market: hotels
Our first market is hotel management companies running twenty or more hotels for different owners.
Each hotel runs on the systems its brand requires, and they don't talk to each other. The same room type is "King Suite" at one hotel and "KSTE" at the next. Management agreements set reporting deadlines, and the management company's fees are typically calculated from those numbers, so every owner reads profit against their own contract.
For hotels, the definitions registry starts from USALI, the Uniform System of Accounts for the Lodging Industry. The hotel views include profit against budget by hotel, the P&L statement, brand-fee checks against contract rates, and a close view that flags reopened or restated months. No guest or card data enters the product.
Hotels are the hardest version of the problem: many sites, many systems, many owners, and numbers that move every night. That is why we start here. Hotels are where we prove the platform, not where it ends.
Who builds it
Jiten Pai builds rokda, in Boston. For 24 years he has been the person who has to defend a number, most of it financial. As VP of Data Engineering at a lending platform, he owned month-end financial reporting for hundreds of lenders, where every figure had to tie back to the loan systems before it went out. Across his career he has built and led data-loading, dashboard and master-data teams, and chosen the platforms they ran on, so he has sat on every side of "whose number is right."
He works with informal advisors from hotel finance and operations who pressure-test what rokda does against how the work really gets done.
He designs the system and directs the AI coding agents that write the code. Every change passes independent automated code review and an extensive test suite, including tests proven to catch the failures they guard against, before it merges. Every call that matters is his.
Previously at Viacom, A+E Networks and LendKey.
rokda, say row-k-daa
Why rokda? It's the word for cash on the table: money you can count. We hold every number to that standard.