Field notes and proprietary data

Most public benchmarks for voice AI, language models, and outbound tooling are written for an American buyer with English-only workloads. They are useful, but they are not the market we build for. Link AI operates the only at-scale Hebrew voice-agent deployment we are aware of inside Israel, paired with a cold-outreach engine that has crossed eight figures of sent volume across Hebrew and English. The data we collect from those systems answers questions a global benchmark cannot.

This page is the index for that work. Each report is published as a standalone, dated artifact under a unique URL with a forthcoming machine-readable dataset attached, so AI search engines and Google Dataset Search can index the underlying numbers and not just the prose. We do not paywall the reports. We do not gate them behind a sales conversation. The waitlist exists so we can tell you when a new one is live.

What we publish

Three threads, all narrow on purpose. We would rather ship one report that resolves a real question than four overviews that restate things already in the discourse.

• Hebrew voice agents. Pickup latency, ASR accuracy on Hebrew dialects and code-switched Hebrew-English, call-completion rates by vertical, agent-to-human handoff triggers, and the cost-per-resolved-call envelope at production volume.
• Cold email at scale. Open and reply rates by language, persona, and industry across Israel and the wider region. Deliverability decay curves. The marginal cost of per-lead research versus template-only sends, measured in meetings booked rather than vanity metrics.
• Israeli SMB AI adoption. Where AI is actually deployed inside small and mid-market businesses in Israel today, what stack they bought, what they pay, what they replaced, and where the next 12 months of buying are headed. Drawn from primary interviews and our own customer base.

Methodology

We disclose methodology in the front matter of every report: sample size, time window, how the data was collected, what we excluded, and where our priors might be wrong. When a number comes from Link AI’s own infrastructure, it is labelled. When it comes from a customer interview, it is labelled. When it comes from a third-party source, the source is cited inline. Aggregated metrics never identify a customer or individual.

Forthcoming

The first report, Hebrew Voice Report 2026, is being assembled now. A short preview with the chapter outline, methodology, and the publication-date estimate is at /research/hebrew-voice-report-2026-preview. Further reports on cold-email deliverability and Israeli SMB AI adoption are planned for later in the year. We will not announce dates we do not intend to hit.

For now this page is intentionally sparse. Add your email below and you will hear from us once when something is published. That is the entire promise.