Kyligence.

Somewhere in a glass tower in Pudong, a finance lead types a question into a cell. Not a formula. A sentence. Why did the East region's gross margin slip last week? The cell thinks for a beat, then explains itself - prose, charts, the three customers most responsible. No SQL. No ticket to the data team. No 48-hour wait.

This is what Kyligence sells now. A copilot that sits inside the software the CFO already uses, and answers in the language she already speaks. It is unglamorous, infrastructural, and quietly enormous.

Kyligence is a ten-year-old company built on a fifteen-year-old idea (the OLAP cube), pointed at the very newest problem in enterprise software: how to get an AI to give a useful answer about numbers you actually trust. The wager is that the boring middle of the data stack - the semantic layer, the metrics store, the cube - is where generative AI either earns its keep or doesn't.

For most of the last decade, enterprises did roughly the same thing with data. They piled it into a warehouse. They hired analysts to wrangle it. They built dashboards. Then the dashboards proliferated until nobody knew which dashboard was the right dashboard, and the company quietly developed seven slightly different definitions of "revenue."

The dirty secret of business intelligence is that it has a latency problem and a trust problem at the same time. Queries on real datasets take minutes. The people who need answers don't have minutes. So they make calls on stale numbers, or they don't make the call.

Yang Li saw this from the inside. As senior architect of big data at eBay's Shanghai analytics group, he watched analysts wait for queries to finish the way you wait for a kettle. In 2013, he and a small team started Apache Kylin to do something stubbornly old-fashioned: pre-compute the cube. Push the heavy work to off-hours. Return answers in milliseconds when the human asks.

In 2016, Luke Han and Yang Li bet that an Apache project - even a Top-Level one - was not enough. Open source pays you in stars and praise, not payroll. So they founded Kyligence, set up shop in Shanghai, and almost immediately set up a second office in San Jose. It is the kind of move you make when you believe your buyers are global and your engineers are local, and you refuse to choose.

The early customer list looked, frankly, intimidating for a Series A startup. UBS. China Construction Bank. MetLife. These are not buyers known for taking flyers on small vendors. They bought because the alternative was rebuilding what Kylin already did, and they didn't have the time.

Redpoint Ventures came in. Then Cisco Investments. Then Coatue with $25 million. Then SPDB International led a $70 million Series D in 2021. The total is past $110 million. The valuation has not been disclosed, which is its own kind of statement in a hype-soaked market.

The Kyligence stack reads, from the bottom, like a careful sentence. The OLAP engine - Kyligence Enterprise - sits on your cloud data lake and pre-computes the cubes you need. The metrics platform - Zen - is where humans define what "revenue" or "active customer" actually means, once, so seven dashboards stop arguing. The copilot is the part the CFO talks to.

What you can actually do with it

Ask a question. Get an answer with the math attached. Open Excel and let the copilot fill in the analysis you would have asked an analyst for. Wire the semantic layer into Tableau, Power BI, or your custom app and watch sub-second queries return on petabyte data without anybody re-aggregating anything by hand.

Enterprise software gets validated in roughly one way - by people who do not enjoy being early adopters paying for it anyway. Kyligence's customer roster runs heavy on the conservative end of the buyer spectrum: UBS, MetLife, Ping An, China Merchants Bank, China Construction Bank, Costa Coffee. Manufacturing. Healthcare. Retail. Industries where wrong numbers have consequences measured in regulators and refunds.

The technology stack tells the same story sideways. Microsoft, AWS, Tableau, and Huawei are listed partners. The Apache Kylin project itself - which Kyligence stewards - has been adopted by hundreds of companies independently of any sales motion.

The company writes the mission in modest sentences. Transform the way people interact with data and AI. Make metrics accessible to every business user, not just analysts. The unstated half is more interesting - shrink the distance between a question and an answer until the distance disappears.

The pitch is not "replace your data team." Anyone who has tried to actually replace a data team in a real company knows how that goes. The pitch is closer to: free the data team from being a help desk. Let them build the model once, define the metric once, then watch as the rest of the business stops queuing for definitions and starts asking questions directly.

Every generative AI demo with a chart in it eventually runs into the same wall. The model can hallucinate prose. It cannot hallucinate the company's actual revenue. Somewhere, something has to ground the answer in a real number, computed the same way as last quarter, defined the same way across every team. That somewhere is the semantic layer. Kyligence has been building it since before "semantic layer" became a phrase the venture market said in public.

That is the bet underneath the copilot demo. Not that GPT will get better at numbers - it will. Not that LLMs will replace BI - they won't, not cleanly. The bet is that the infrastructure underneath the answer is the moat, and Kyligence already owns a piece of that infrastructure that pre-dates the hype by ten years.

The CFO in Pudong closes her laptop. The question got answered in eight seconds. She did not file a ticket. She did not wait for Monday's standup. She did not ask anyone to "pull a quick number." She typed a sentence into a cell and the cell talked back, citing real customers and real deltas, computed from a cube that knew the answer before she asked.

That is not magic. That is a cube, a metrics store, an Apache project, ten years of patience, a $70 million Series D, and a quietly opinionated bet that the boring middle of the stack is where the value lives. Kyligence built it. The CFO uses it. The dashboard - finally - tells the truth on time.