Operations and Management → Operational Management Ethics

Operational Management Ethics

1) Why do you need it

Operations are constant speed ↔ risk ↔ cost trade-offs. The ethical framework helps you make decisions under the pressure of data, money and deadlines so as not to deceive users and stakeholders, not to violate privacy and not undermine the long-term sustainability of the platform.

• Set clear red lines and rules of conduct for teams and he-calls.
• Ensure the integrity of SLAs, metrics and communications in incidents.
• Protect the privacy, data and rights of users/partners.
• Make automation and AI manageable, explainable and safe.

2) Basic principles (core)

1. Safety first: solutions should not increase the likelihood of harm to users/data.
2. Measurement honesty: no "cosmetics" metrics, a single SSOT and reproducibility.
3. Transparency of actions: who did what, why, based on what data.
4. Responsibility and accountability: role → authority → audit → implications.
5. Minimizing data: we collect only what we need, limit access and shelf life.
6. Explainable Ops/AI: Automatic solutions are clear, reversible, and disputable.
7. Equity and non-discrimination: "no bias" policies in rules and models.
8. Blameless, but not subjectless: mistakes are a reason to change the system, not hide facts.

3) Ethics of Metrics, SLO/SLA and Reporting

• Unified definitions of metrics (windows, aggregators), formula versioning.
• It is forbidden: to hide incidents in "planned work," to transfer windows/time zones for the sake of "beautiful" SLA, to exclude data without documentary grounds.
• Clear labeling: "estimate," "forecast," "fact," "exception and reason."
• Postmortems are published with facts and actions, not a "PR take."

Anti-patterns: "two versions of p99," manual adjustment of reports, selective periods "without peaks."

4) Privacy and work with PII/payment data

Minimization: by default, the PII does not leave the production loop; masks in logs/dashboards.
Access by role: principle of least privileges; audit each reading of sensitive data.
Retention: clear retention, deletion/anonymization policy.
Data incidents: immediate notification of owners/legal entities according to regulations.

Forbidden: transfer real PII to stage/analytics without anonymization; share with vendors outside the contract.

5) Ethical communications in incidents

Truthfulness and timeliness: ETA statuses, clear language, no omissions.
Don't blame individuals: Focus on facts and systemic causes.
No "quiet" fixes: changes that affect the user must be designated.

Limiting speculation: "We check X, next summary at 8:15 p.m."

What is happening/who is affected/what we are doing/when the next update/where to follow

6) Ethics of automation and AI in operations

Clear perimeter: a list of actions that the AI ​ ​/bot can do without confirmation (only reversible and low-risk).

Explainability: for each recommendation - sources and arguments, prohibition "without references."

HITL (person in the loop): confirmation of sensitive actions (traffic shifting, PSP switching, limit change).
Audit: prompts/actions/decisions log, dry-run reports.
Bias & fairness: regularly checking recommendations for distortions (geo, devices, player type).
Data for AI: PII/secrets "sucking" ban; use of impersonal display cases.

7) Vendor relationships and conflicts of interest

SLA/OLA in SLO: fair map of dependencies; public facts on vendor outages.

Competing interests: Not making architectural decisions due to "personal bonuses/referral schemes."

Ethics of tenders and pilots: comparable tests, documented victory criteria.

Prohibited: hide provider failures as "ours," change comparison metrics "for the winner."

8) Red Lines (disjoint)

Manipulation of data and reports.
Concealment of incidents affecting users/money.
Leverage real-world PII in unprotected environments.
Automation of irreversible actions without HITL and rollback plan.
Pressure on employees to "embellish" metrics or skip the gate.

Violation is a trigger for a formal investigation, up to and including stopping releases.

9) Policies and norms (fragments)

- All metrics are described in the catalog with formula, window and owner.
- Formula change - via RFC and parallel run (old vs new).
- Any exceptions in the SLA are documented and signed by the parties.

- First summary of 15 minutes, then ETA.
- Tone: facts, hypotheses are marked, references to artifacts.
- It is forbidden to promise deadlines without justification (progress/plan/resources).

- Allowed: summaries, tickets, requests for observability, annotations, pre-scale (reversibly).
- Requires confirmation: feilover, changing limits, enabling safe-mode, canary pause.
- Required: activity log, explainability, dry-run before use.

10) Roles and responsibilities

Head of Ops: the owner of ethical policies, the authority of the "stop valve."

Incident manager: quality and honesty of communications, control of post-mortems.

SRE/Observability: SSOT metrics, auditing formulas and alerts, protection against "cosmetics."

DPO/Security: privacy, access, leak investigations.
Legal/PR: compliance with laws/contracts, external communications.
Domain commands: compliance with gates, correct data and artifacts.

11) Dashboards and ethics artifacts

Metrics Integrity - Online↔DWH discrepancies, formula changes, out-of-date panels.
Incident Comms: time to first update, ETA compliance, completeness of summaries.
Privacy & Access: calls to PII, abnormal requests, retention deadlines.
AI Governance: number of auto actions, dry-run share, kickbacks, controversial decisions.
Vendor Truth: Incidents by provider, collating their reports and our SLOs.

12) Checklists

• There are phicheflags and a rollback plan.
• Included SLO alerts and annotations.
• There is no pressure "from above" to bypass the gates.
• Risks/exclusions documented, agreed.

• Timely first update and ETA.
• Facts separated from hypotheses, references to data.
• No attempt to understate scale/impact.
• Postmortem on time, actions scheduled.

• The list of allowed auto-actions is approved.
• Log and explainability enabled.
• PII not used/masked.
• HITL for sensitive operations.

13) Ethics Maturity KPI

Metrics Integrity Score (drift Online↔DWH ≤ 2%, the share of versioned formulas ≥ 95%).
Incident Comms SLA (first summary ≤ 15 min, ETA compliance ≥ 90%).
Privacy Violations = 0, share of access to PII with justification = 100%.
AI Safety: the share of reversible auto-actions = 100%, kickbacks <5%, controversial cases disassembled = 100%.
Whistle Safety Index: anonymous channels work, calls are sorted out ≤ 7 days.

14) Anti-patterns

"Painting the grass": cosmetics in metrics, redefining SLA "retroactively."

"Nightly releases without flags" for deadlines.
Private chats and solutions without logging.
Toxic retro/post-mortem, blame game.
AI without RAG/explainability, black box in operations.

Oversupplied data collection "just in case."

15) Practical language (can be copied into policy)

We tell the truth about the state of the systems.
We do not hide incidents and do not distort metrics.
We protect user data and restrict access.
We automate only reversible and safe actions, the rest is through HITL.
We document decisions and respect the "stop crane."

- SLO/guard rails are active; rollback plan checked.
- Changes of metrics/formulas are formalized by RFC and announced.
- No conflicts of interest, decisions made on data.

16) 30/60/90 - implementation plan

• Approve the "red lines," the code, the policy of incident communications and privacy.
• Assign owners (Head of Ops, DPO, Observability).
• Launch Metrics Integrity and Incident Comms panels.

• Implement RFC for metrics formulas and SSOT; rebuild controversial panels.
• Formalize the perimeter of AI/bots (allowed actions, HITL, log).
• Conduct ethics training for on-call and domain managers.

• Audit compliance, review cases/complaints, update policies.
• Associate ethics KPIs with team OKRs (e.g. Incident Comms SLA, Integrity Score).
• Retro in efficiency and adjustment of "red lines."

17) FAQ

Q: What if a business asks to "tweak" an SLA report?
A: Refuse, citing honest metrics policy and SSOT. Offer an alternative: the "user experience" metric with understandable exceptions drawn up through the contract.

Q: How do you combine release speed and ethics?
A: Small increments, phicheflags, canaries and SLO autogates. Ethics is not a brake, but insurance against expensive mistakes.

Q: When to publicly admit a mistake?
A: Always when the impact is palpable to users/partners. Status template + action plan + deadlines.