Methodology

All data on ExitBase is self-reported by people who received a severance package. Reports are submitted through the submit form, which asks for:

• Context. Company, industry, role / function, level or band, location, separation date, years at the company, total years of experience, and the reason for separation.
• Compensation. Base salary and total comp (base + bonus + equity) for the year preceding the separation.
• Severance terms. Weeks of pay, gross severance amount, equity / RSU treatment, COBRA or healthcare continuation, bonus treatment, PTO payout, outplacement support, non-compete duration, and non-disparagement clause.
• Negotiation. Whether the submitter pushed back on the offer, and if so, what changed.
• Free-text notes. Anything the submitter wants the next person to know.

Submitters may optionally attach an offer letter or separation agreement when submitting a report. Attached documents are not displayed publicly; we use them internally to spot-check individual reports and to flag patterns that don't match documented packages. Reports with verifying documents are treated as higher-confidence inputs to aggregate statistics.

We do not verify every field on every report, and we do not independently confirm employment at a company. The dataset should be read as a directionally useful sample, not an audit.

The submit form defaults to anonymous. When the anonymous toggle is on, no email or other identifier is stored alongside the report. If a submitter opts in to providing a contact email, the email is stored privately and used only for follow-up questions about that report — it is never shown to other users.

Free-text notes go through a light review for content that could re-identify the submitter or other individuals (manager names, internal identifiers, etc.). When we find such content we edit it out or remove the report.

Company and role pages show summary statistics derived from submitted reports — for example, median weeks of severance, median total comp at separation, the share of reports that included extended COBRA, and so on. A few things to know about how these are computed:

• Medians, not means. Severance distributions are long-tailed; medians are more robust to a single very large or very small report.
• Sample-size labels. Each aggregate view shows how many reports it is computed from. With small samples, treat the number as a hint, not a benchmark.
• Outliers are kept, not winsorized. We do not trim or cap unusually high or low reports. We do remove reports that we believe are fabricated.
• Currency. Compensation figures are stored in the currency the submitter reported. Aggregate views currently assume US dollars unless otherwise labeled; we plan to add per-currency views as international coverage grows.

• Self-selection bias. People who received unusual packages — very generous or very thin — are more likely to submit. Treat aggregate numbers as a directional sample of motivated reporters, not the population mean.
• Recency bias. Severance norms move with the labor market. A 2022 package and a 2026 package at the same company are not directly comparable; the separation date is shown on every report for that reason.
• Coverage gaps. Early on, some companies and roles have only a handful of reports. We surface sample size so you can decide how much weight to put on each view.
• Self-reporting. Even with optional verification, most fields rest on the submitter's accuracy. We remove reports we identify as false, but we cannot catch everything.

If you spot an error in a report you submitted — or anything on the site that looks wrong — reach us through our contact page. We will correct or remove the report. See our Privacy Policy for how to request deletion of your own submission.