ELTIF reporting in five languages · Igor Sabol

The hard part of multilingual investor reporting is not translation: it is the chain that has to finish before any translation can start. The setup is straightforward: an Excel workbook, a Word template, and a macro that builds every PDF off a language switch, on a quarterly calendar that does not move.

Three layers sit on top of each other: the data layer with the numbers signed off by accounting; the translation layer with every label, header, and structural caption that has to exist in five languages; and the narrative layer with the prose that someone has to write each quarter because it changes each quarter. Most readers assume the translation layer is the bottleneck, which is understandable but also wrong.

The translation layer was solved in two pieces. The first was a dictionary: a Languages sheet in the workbook with one row per label and one column per language. Flip the language switch and every chart legend, table header, section title, and structural caption swapped via lookup. Populating it the first time was the work; after that, maintenance was mechanical, a row at a time when a new label entered the report. Boring infrastructure, paid for once, used every quarter for years.

The second piece was the narrative. Machine translation handled a first pass on the executive summary, the commentary, and the per-deal blurbs, and native-speaker colleagues then read each language version and closed the gap, usually within days. This is a recent-decade operating model: the version of this essay set in 2012 would describe a two-week translator turnaround, but by the late 2010s the machine pass was good enough that the reviewer's job was to correct rather than translate. Outsourcing the commentary work to third-party firms was tried twice, with two different firms; neither worked, more on that below.

What gates the calendar is the chain. The numbers from accounting land some time after quarter-end and the PDFs ship a few weeks after that, but the window is always short. The English first draft comes off the numbers, and the executive summary and market commentary follow in English, written by someone close enough to the portfolio to comment on them. Then comes the multi-language chain: machine translation across each of the other languages, native-speaker review on each, integration of the reviewed text back into the workbook, rebuild the PDFs, send. Each link gates the next, and the slowest native-reviewer leg sets the publication date.

Per-deal write-ups were the worst part of this in the early cycles, because they came from a blank page each quarter for each new investment. The fix, which took a while to land on, was reusing the investment notifications the deal team produced at closing, already prepared for investor sign-off. When the notification existed, it was adapted; when it did not, the fallback was fresh prose plus machine translation. That cut a meaningful slice off the cycle, and it was the cleanest fix-upstream move in the whole pipeline, but it did not change the fact that the chain still had to serialize.

The chain is the real cost, and the chain is what gets multiplied by language count when something upstream slips. Late changes from accounting come from several places: commitment changes, late valuation adjustments, a missing investment that should have been booked, a mis-recorded investor redemption, a new commitment that landed in the wrong place. Sometimes a native-reviewer reply comes back late, adding a softer second source on the back end. Either way, the change does not arrive on time: it arrives mid-cycle, after the chain has already started moving.

In a single-language report, one upstream correction costs one rebuild: the number changes in the workbook, the table refreshes, the PDF re-renders, you check the few places the number flows to, you send. In a five-language report the same correction costs five rebuilds, plus a re-translate of any narrative line that mentioned the number, plus a re-review by the native speakers on any of those lines, plus a re-integrate, plus a re-PDF, plus a cross-language diff to catch any version where something moved that should not have moved: one upstream weakness, five downstream rebuilds, all on the same publication deadline.

The honest framing is not "multilingual reporting is hard", rather that multilingual reporting is a multiplier on whatever upstream weakness the operation has: a clean upstream produces the language versions without drama, a noisy upstream produces five times the rework, and the difference is entirely upstream, which a single-language report lets you ignore.

Two attempts were made to escape this lineage. The first was to outsource the commentary and the translations together: third-party firms would produce the English commentary off the numbers and run the multi-language pass alongside, with AI translation as the long-term direction. Neither firm was good enough:...