7 Ways Buy-Side Firms Are Sourcing Non-English Primary Research in 2026

Cross-border primary research is a persistent workflow problem. An Asia-Pacific consumer thesis, a European industrials name, or a Latin American fintech often requires expert calls, filings, and management commentary that never originated in English. The rise of MCP-connected research agents has sharpened the problem: agents ingest transcripts as text, and the language of that text determines what gets retrieved, weighted, and cited.

This guide walks through seven sourcing patterns buy-side firms use in 2026 to get non-English primary research into the stack without losing provenance or compliance discipline. Each item covers the mechanics, a named example of who operates in the category, and the honest tradeoff on cost, latency, or provenance risk.

1. Native-language expert calls with human consecutive interpretation

This is the traditional expert-network default. Networks including Dialectica, Third Bridge, and Coleman routinely staff Mandarin, Japanese, Korean, and European-language experts and pair them with a professional consecutive interpreter. The analyst asks a question in English, the interpreter renders it into the expert's language, the expert answers, and the interpreter renders the answer back.

The mechanics are well understood and the compliance surface is familiar. Interpreters typically sign the same NDAs as the expert, calls are recorded in both language streams, and the moderator layer sits inside the network's existing MNPI screening process. The tradeoff is cost and latency. Consecutive interpretation roughly doubles a call's runtime, adding 40 to 60 minutes to a 60-minute session, and unit cost roughly doubles as a result. For a deep-dive on a single expert this is defensible. For a 12-expert scan across four APAC markets it starts to hurt.

2. Native-language calls with AI-assisted real-time translation

An emerging pattern layers AI-assisted real-time translation over the recording rather than a human interpreter over the live call. The analyst hears English through a translation feed while the recording captures both language streams. The transcript then carries the source-language utterance and the English rendering side by side.

The attraction is speed and cost: the call runs in roughly its native duration, the analyst can react in near real time, and the recorded transcript preserves the original-language ground truth for compliance review. The honest tradeoff is in the transcript. Real-time machine translation still trips on industry-specific vocabulary, code-mixed speech (Mandarin technical terms embedded in English, common in Taiwan semiconductor calls), and named entities. Buy-side teams using this pattern generally treat the live English feed as a directional aid and the post-call bilingual transcript, cleaned by a human moderator or a translation reviewer, as the citable record.

3. Japan-specialist networks

Japan is a market where domicile and language sit tightly together. Local experts often prefer to be sourced through a Japanese-domiciled network operating under Japanese compliance norms, and the pool of relevant experts inside Japanese corporates is not fully addressable through the New York and London tier-one desks.

VisasQ is the anchor example. The Tokyo-based network acquired Coleman in 2021, which gave buy-side firms outside Japan a direct route into a Japanese-domiciled expert base with local screening and local contracting. The tradeoff here is workflow: buy-side ops teams have to manage a separate MSA, a separate compliance channel, and a separate billing relationship for the Japan pipeline, and the interpretation question from item 1 or item 2 still applies on top. What the specialist gives you is depth of coverage inside the Japanese enterprise stack that a generalist tier-one desk cannot fully match.

4. China-focused networks under the post-2023 regime

China is the most compliance-heavy leg of any cross-border research program. Capvision went through the 2023 mainland regulatory tightening and now operates under stricter cross-border data rules. Lynk is the other named China-focused network buy-side firms use. Both sit inside a regulatory frame that includes PIPL, revisions to the counter-espionage law, and evolving rules on cross-border transfer of information that a Chinese authority may treat as sensitive.

The mechanics have shifted meaningfully. Expert screening on China topics is tighter, some topic areas are effectively off-limits regardless of the expert's private willingness to speak, and cross-border data handling now has explicit process steps rather than implicit ones. The tradeoff for buy-side firms is that the compliance overhead is real and non-negotiable, but the alternative, doing China work without a network that operates inside the local regime, carries a materially worse provenance and compliance posture. Firms running China theses in 2026 typically pair a local-network expert leg with an in-house translation and review step (see item 7).

5. Local-language earnings and management call transcription

Earnings calls, capital markets days, and management commentary in non-English markets used to be a documentation gap for global buy-side desks. That gap has narrowed. Vendors including Aiera and AlphaSense have expanded non-English earnings coverage and now distribute local-language calls as an AI-grade dataset, meaning transcripts that are timestamped, speaker-tagged, and clean enough to feed into a research agent's retrieval layer.

The use case is different from an expert call. This is publicly disclosed management commentary, so the compliance question is not MNPI but accuracy: does the transcript faithfully render what the CFO said in Japanese or Portuguese, and does the English rendering preserve the numbers and the hedges. The tradeoff is that transcript quality varies by language and by vendor, and buy-side teams that cite non-English management commentary in an investment committee memo typically anchor the citation to a timestamped recording rather than to the English rendering alone.

6. Multilingual survey and synthetic-panel work

The survey layer sits next to the expert-call layer and has its own multilingual pattern. Traditional multi-country survey overlays from ProSapient and Atheneum let buy-side teams field a channel check or a demand-signal survey across five or six markets in local languages, with the vendor handling translation of the instrument and the responses.

The newer layer is synthetic. NewtonX launched B2B Synthetic Personas in 2026, an approach that lets analysts probe a modeled respondent grounded in a real underlying B2B panel. For non-English contexts the appeal is coverage: a synthetic layer can be queried in-language without waiting for a field. The honest tradeoff is provenance. A survey response from a screened, live B2B respondent carries a different evidentiary weight than a synthetic response derived from a model over a panel, and buy-side teams that use synthetic overlays generally treat them as a hypothesis-generation step rather than a citable primary source.

7. In-house LLM translation of primary documents

The last pattern is the one most buy-side research ops teams have industrialized in the past 18 months: in-house translation of primary documents through an LLM pipeline, with human review on material claims. The inputs are filings, regulatory releases, local-language press, and sometimes the raw recording of an expert call or a management event. The outputs are English renderings that get loaded into the same document store the research agent already queries.

The mechanics are straightforward but the discipline matters. A responsible pipeline preserves the source-language document alongside the English rendering, timestamps the translation, records the model version used, and routes anything that will be cited in an IC memo through a human reviewer who reads the source language. The tradeoff sits on the review step. Skipping it saves time and creates provenance risk: an LLM rendering of a Chinese regulatory filing that misreads a number or softens a hedge is not defensible up the chain. Buy-side firms that get this right treat in-house translation as an accelerator on the collection step, not a substitute for the analysis step.