2026-08-10

Eighteen Portfolio Companies, Eighteen Definitions of ARR

For heads of portfolio operations, fund partners, and whoever spends the first week of every quarter stitching a dozen differently-shaped spreadsheets into one table.

The expensive part of portfolio operations was never the analysis. It's the reconciliation.

Eighteen companies, reporting quarterly. One counts implementation fees as ARR. One books multi-year contracts in full. One treats prepaid class packs as subscription revenue. Every method looks defensible from inside the company that uses it. By the time those numbers land in your master table, you have a total — a total assembled from eighteen different definitions, which is exactly the number nobody wants to put in front of an LP.

We published this as a record you can open. A fictional fund, 18 portfolio companies, 8 quarters. The table has two ARR columns: reported_arr is what each company sent in, unified_arr is the same revenue recomputed under the fund's own definition.

Open the record →

The gap between those two columns is the entire reason the report exists.

First question: who is inflating

Rank all 18 by unified ARR, and identify the three with the largest gap between reported and unified.

Then the observation that actually earns its keep: all 18 companies reported higher than the unified figure. Not one under-reported.

That isn't eighteen isolated accounting errors — it's a systematic tilt. Wherever the boundary is ambiguous, every company leans the way that flatters them. Looking at one company, you'd call it a communication problem. Seeing all eighteen lean the same direction tells you it's structural, and that the fix is the reporting template, not eighteen phone calls.

Xingqiao's gap is close to ¥160M. The agent's read: most likely one-time implementation fees, custom development revenue, or multi-year contracts booked in full. It labelled that as inference, because it is. What it can state is the size of the gap; the cause is a question for the company. Drawing that line explicitly is, we'd argue, considerably more useful than a confident wrong answer.

Second question: who is running out of cash

For 2026 Q2 only — which companies have less than 6 months of runway? Give cash balance and monthly net burn.

Out of 18, exactly one.

Yunma AI: runway 3.7 months, cash balance ¥11.75M, monthly net burn ¥3.17M, projected to run dry around December 2026.

The agent added the context that portfolio operations actually needs: Yunma has the second-highest burn rate of the eighteen but the second-lowest cash reserve. Neither is extreme on its own; together they produce the only red line in the book. The fastest burner is in fact Lingxi AI at ¥4.43M a month — but with ¥45.52M in the bank and 10.3 months of runway, it is not this quarter's problem.

A table sorted on a single metric puts Lingxi at the top and costs you a week on the wrong company.

One more detail in that question is worth naming: it says for 2026 Q2 only. Runway is a point-in-time figure, not something you average across eight quarters — leave the window open and the mean erases the red line. The product doesn't guess here; it tells you explicitly whether your time window was applied.

Third turn: brief the investment committee

Turn this runway warning into an audio briefing for the IC.

A 3 minute 51 second two-host audio briefing, ready to play: Yunma's red line, the financial detail, the contrast with Lingxi, overall portfolio health, and a closing recommendation — a financing plan or a cost-reduction plan within two weeks.

Every figure in that script has to trace back to a query this conversation ran. Anything that can't be traced doesn't get said.

The night before an IC meeting, the value isn't the two hours of writing it saved. It's that everyone hears the same version.

One definition is a piece of infrastructure

Worth being clear about the division of labour: aligning eighteen companies onto one definition is not the AI's contribution. It's yours. You decide what counts as ARR and what doesn't, and that decision lives in the system as a standalone, traceable record.

What the AI does is everything downstream — recompute under that definition, rank, surface the largest deviations, filter to the one red line inside the quarter you specified, and turn it into something an investment committee can listen to. All of it repeats every quarter, and all of it is the part that needs no judgment.

The part that needs judgment — what counts as ARR — stays with you.

Try it with your own portfolio

At the bottom of the page there's Remix with my data. Upload your own portfolio table; the questions and definitions cross over, the data does not. Eighteen companies, fifty, or three — a portfolio company here is a dimension in a table, not eighteen separate workspaces to log into.

Open the record →

Luxiva turns a spreadsheet and a definition you can defend into an agent your whole team can question.

← Back to Blog