Owner, Decision, Deadline: An Async Operating System for Global Work

Write the action contract at the top

A useful message begins with its operating contract: owner, requested decision, deadline, timezone and default if nobody responds. “Please review soon” forces every reader to infer priority. “Mina owns the launch decision; comment on the two named risks by 16:00 UTC Thursday; absent a blocking issue, option B proceeds” is actionable. The body then supplies context, options, recommendation and evidence. The contract is not bureaucracy for every chat. It belongs on work where delay, disagreement or accountability matters.

Timezones should use explicit dates and offsets, preferably a shared standard such as UTC alongside local display. “Tomorrow morning” can describe two different workdays. Deadlines must also allow a reasonable response window for the people expected to contribute. Sending a request at the end of one region’s day and closing it before that region returns creates nominal inclusion without actual participation. A global team should map normal overlap, holidays and on-call coverage rather than expect constant personal availability.

Decision records preserve rationale, not every conversational turn

The Microsoft result deserves careful boundaries. It studied more than 60,000 US Microsoft employees during a firm-wide shift to remote work and used collaboration data from that setting. Organizational structure, emergency pandemic conditions, existing tools and workforce composition limit universal inference. It can motivate questions about silos and cross-group ties, but it cannot establish that an asynchronous practice will cause the same network change in a smaller multinational, a factory, a public agency or a team designed as remote from inception.

A decision record states the problem, constraints, alternatives, decision, owner, date, evidence and review trigger. It links to discussion without requiring a future employee to read 140 comments. Record what was decided and why, including material dissent or uncertainty. If new evidence reverses the choice, create a new entry linked to the old one instead of rewriting history. This gives onboarding staff and auditors a reliable chain and prevents the same settled question from returning whenever team membership changes.

Documentation has a carrying cost. It needs an owner, location, access model and lifecycle. Duplicate pages can be worse than no page because search returns conflicting instructions. Mark canonical records, archive superseded material and expose the last reviewed date. GitLab’s public handbook demonstrates documentation-first operating practices at unusual scale, but another organization should adapt the pattern to its risk and work rather than copy every convention. The principle is that decisions should be findable and changeable through a known process.

Worked example one: a product handoff across three regions

An illustrative team has design in Toronto, engineering in Warsaw and quality assurance in Singapore. Toronto posts a handoff by 20:00 UTC containing the approved prototype, open questions, owner for each question, acceptance criteria and a 14:00 UTC next-day decision deadline. Warsaw begins with the same artifact, implements two paths and records test results. Before ending its day, Warsaw identifies one unresolved privacy dependency and tags the named owner. Singapore receives a runnable build and the expected results rather than a generic request to test everything.

Singapore finds that one path fails the acceptance criterion and posts reproduction steps, severity and screenshots. The failure triggers the prewritten rule: no release until privacy and quality owners approve. Toronto returns to a decision packet instead of three separate chat histories. The process shortens idle time because each handoff contains state and authority. If Warsaw had discovered a security incident rather than a normal dependency, the escalation rule would have opened a live incident channel immediately. Async is the default, not a prohibition on speaking.

Escalate when text becomes slower or less safe than conversation

Move live when a decision is urgent, emotionally charged, highly ambiguous, rapidly branching or dependent on subtle negotiation. Signals include repeated restatement, more than two rounds without convergence, conflicting interpretations of risk, or harm that grows while the thread waits. The call should have a facilitator and a decision goal. Afterward, publish a concise record so colleagues outside the overlap are not excluded. Synchronous time resolves the knot; documentation reconnects the wider team.

Escalation should not reward poor preparation. A live call without evidence or decision rights can consume every timezone and still produce no action. The requester supplies the current record, disagreement and options beforehand when time permits. Teams can maintain a small overlap window for complex coordination and protect the remainder for local work. If the same issue repeatedly requires emergency calls, inspect architecture, staffing or ownership. Communication style may be revealing an operational dependency rather than causing it.

Worked example two: a customer-risk decision that needs a call

A global account team learns that a promised integration may miss a regulatory deadline. The account owner posts a decision record with contract language, engineering estimate, customer impact and three options. Written comments reveal that legal interprets the obligation differently from sales and engineering. Because delay increases exposure and the disagreement concerns meaning rather than missing data, the owner schedules a 30-minute overlap call with the three decision roles. The group chooses a limited compliant release and names the customer communicator, technical owner and next evidence checkpoint.

The written record is updated with the decision, dissent, customer commitment and review date. Colleagues in other regions receive the outcome at the start of their day. This is not a failure of async practice. The initial record surfaced the conflict and made the live conversation focused; the final record prevented the call from becoming private organizational memory. A fully asynchronous exchange might have consumed another day, while an unprepared meeting might have hidden the legal disagreement until later.

Accessibility, onboarding and security determine who can participate

Security incidents require a handoff pattern of their own. Do not paste secrets, credentials or vulnerable customer data into broad decision threads. Use a restricted incident record, stable severity vocabulary, explicit commander and timed updates that reveal only what each audience needs. Preserve evidence according to response policy. When the incident closes, publish a sanitized learning record and track remediation. Async transparency should not become uncontrolled disclosure, while excessive secrecy should not leave downstream teams unaware of actions they must take.

Accessible asynchronous work uses descriptive headings, meaningful link text, alt text, captions and transcripts, readable formatting and language that does not depend on cultural shorthand. Important decisions should not exist only in audio, screenshots or disappearing chat. New employees need a map of canonical systems, examples of good records and permission to ask when a norm is unclear. Documentation can reduce reliance on insider networks, but only if search and access work for the intended audience.

The same record that improves onboarding can expose sensitive data. Classify documents, minimize personal information, restrict customer and security details, log access where appropriate and define retention. Public-handbook practices do not imply public-by-default for regulated work. Remote endpoints and shared links need identity controls and revocation. A mature async system combines explicit action, durable rationale, inclusive formats and proportionate security. Its success metric is not fewer meetings; it is faster, safer decisions with fewer people waiting for context.

1. State owner, requested action, deadline, timezone and default outcome on consequential work.
2. Keep canonical decision records with evidence, rationale, dissent and review triggers.
3. Design handoffs around runnable state, acceptance criteria and named unresolved dependencies.
4. Escalate urgent, emotional or repeatedly ambiguous threads to a focused live conversation.
5. Make records accessible, onboarding-friendly and protected according to their information risk.

Sources

• Yang et al. (2021), The Effects of Remote Work on Collaboration among Information Workers
• GitLab Handbook: How to Communicate Effectively in a Remote Team
• W3C Web Content Accessibility Guidelines