Research Methodology

Platform telemetry. Eligoo’s workspaces generate operational records as employees work: tasks, approvals, sequence sends and replies, call outcomes and durations, published posts, CRM changes. For research we use only records from workspaces that have opted in, and only in aggregate: counts, rates and distributions across many workspaces, never one workspace’s numbers.

Controlled experiments. Where a question needs a comparison — does a shorter first message change connect rate; does deeper personalisation change positive reply rate — Eligoo runs the experiment on its own campaigns or with consenting customers who agree to the design in advance. The design, the variable changed and everything held constant are described in the report.

What is not used. Third-party statistics are not restated as our findings. If a report needs to reference an outside source for context it is cited as such and kept separate from what we measured. Individual transcripts, emails and contact records are not quoted, even anonymised, unless the workspace has explicitly agreed to a quoted example.

Before any analysis, workspace identifiers are replaced with random keys that are not stored alongside customer accounts; contact details, names, company names, email content, transcripts and recordings are excluded from the research dataset entirely; and only derived fields — outcome codes, timestamps, durations, counts, categorical settings such as calling window or sequence length — are retained. Aggregates are reported only when enough workspaces contribute that no single one can be inferred from the figure. A workspace that opts out is removed from all future aggregates.

A report is not published until its sample is large enough to be meaningful; the threshold is stated in each report and depends on the metric. Rates on rare outcomes need more observations than rates on common ones, and comparisons between two settings need enough observations in each arm. Where a figure in a report rests on a smaller subset — one industry, one language — that subset’s size is stated next to the figure or the figure is withheld. We would rather publish fewer cuts than publish a cut that a handful of workspaces could move.

The sample is Eligoo customers who opted in, which is not the population of all businesses using AI agents; results describe what happens on this platform, with these employees and these boundaries. Telemetry measures outcomes as recorded by the platform and by people marking meetings and replies, and those markings can be inconsistent. Experiments run on a limited set of industries, languages and list types. Correlations in telemetry are not causal claims; only the controlled experiments support a causal reading, and only for the variable they changed. Every report states which of these apply to it.

If an error is found in a published report — in a definition, a figure or an interpretation — the report is corrected in place, the version number is incremented, and a correction note at the top of the report states what changed and why. Material corrections are also noted on the research index. We do not silently edit figures. Anyone who spots a possible error can write to us through the contact link on the research page.

Cite the report title, Eligoo as publisher, the version and date shown on the report, and the URL. Quote figures with the metric name as defined on this page, and with the sample size and any subset the figure applies to, so that readers can judge it. Do not cite draft figures shared under early access.