Feature · support

Enterprise AI support: tiers, severities and SLAs

A support plan is not a Slack channel. It is a severity model, a response clock, an escalation path and a named owner. Here is what to put in the contract.

Short answer. Self-serve plans include ticket support. Enterprise adds a named customer success manager, custom SLAs, quarterly business reviews and a direct escalation path. Plugsky's current enterprise page promises those items, and a security article states a 4-hour Enterprise support SLA. No full P1–P4 matrix is published. Agree the severity definitions, coverage hours, targets, named roles and exclusions in the order form.

What enterprise support should include

Support is a product with a service level, not a favour from the sales team. The minimum package has five parts: a severity model both sides accept, a coverage window with time zones, a response clock with escalation, named technical and commercial contacts, and a reporting cadence. Architecture reviews, onboarding help and quarterly business reviews are useful additions, but they do not replace the incident path.

TierWho it fitsChannelsNamed contactTypical commitments
Self-serve (Free–Scale)Developers and small teamsPortal and emailNone publishedBest-effort support; plan limits apply
EnterpriseProduction and regulated workloadsPortal, email, escalation channelCSM plus support engineerCustom SLA, QBR, architecture review
Private deploymentVPC, on-prem or air-gappedDeployment-scoped channelNamed technical leadDeployment SLA, security review, update process

The public enterprise page describes a named CSM, custom SLAs and quarterly business reviews. It also lists dedicated capacity, customer-managed keys, audit-log export and a BAA — commercial and security commitments, not response times. Ask for the support schedule as a separate annex.

P1–P4 definitions and targets

Severity is impact, not urgency. A typo in a dashboard can be urgent to marketing and still be a P4. Write definitions in terms of customer impact, affected users, workaround and data risk, then attach coverage and targets. The response times below are AWS Enterprise Support benchmarks, not a Plugsky commitment; they show what a mature schedule looks like.

SeverityDefinitionExampleCoverageBenchmark first responseResolution target
P1Production down for all users, data exposure or active security incidentEvery chat request returns 5xx; confirmed key leak24×715 minutesContinuous work until mitigation; hourly updates
P2Core function unavailable for a subset, or severe degradationOne region or model family failing; no workaround24×71 hourWorkaround same day; daily updates
P3Limited impact with a workaroundElevated latency on one model; a non-critical integrationBusiness hours4 hoursFix in the normal release cycle
P4Question, configuration help or change requestQuota change, billing question, how-toBusiness hours1 business dayAnswer or a dated plan

Plugsky's public pages do not publish this full matrix. One article states a 4-hour Enterprise support SLA. Ask whether that is first response or resolution, which severities it covers, whether it runs 24×7 and which time zone starts the clock. A 4-hour first response is weak for a P1; a 4-hour resolution target may be unrealistic when an upstream provider is the cause.

Response time is not resolution time

Define the milestones separately. Response is a human acknowledgement with a ticket ID and an owner. Triage confirms severity and an initial hypothesis. Workaround restores function without fixing the cause. Mitigation removes customer impact. Resolution fixes the defect. RCA explains what happened and what changed. A first-response SLA can look excellent during a six-hour outage.

Channels, named engineers and escalation

List channels by severity, not just the channels that exist. A portal is fine for P3 and P4. A P1 needs a live path: phone, a shared incident channel or a bridge, plus a named incident commander. Ask for the escalation matrix with roles and backups, not a generic alias, and ask what happens on weekends and holidays.

A named CSM and a named support engineer are different jobs. The CSM owns the relationship, the QBR and the commercial plan. The support engineer owns technical escalation, knows your architecture and can pull in model, platform or security specialists. For production, ask for both, plus a backup. If the contract only names a CSM, you have a commercial contact, not a technical escalation path.

Architecture review, onboarding and QBRs

An architecture review should produce a written record, not a slide. Cover data flow and residency, identity and SSO, key custody, rate limits and concurrency, fallback and retry behaviour, observability, disaster recovery and cost. Assign an owner and a date to each action. Re-run it when you add a region or a regulated data class.

A 30/60/90 onboarding plan keeps the first quarter honest. Week 1: accounts, SSO, keys and quotas. Week 2: integration, evals and a staging load test. Week 3: security review and runbook. Week 4: production cutover with a rollback. Day 60: review latency, error rate and spend against the plan. Day 90: QBR with incidents, credits, roadmap and open risks. The QBR should use your numbers as well as the vendor's.

What to verify in the contract

  • Severity definitions: impact-based, with examples and who can declare each level.
  • Coverage: 24×7 or business hours, time zone, holidays and follow-the-sun handoffs.
  • Targets: response, workaround, mitigation and resolution, measured separately.
  • Exclusions: customer-side issues, upstream providers, beta features and maintenance.
  • Remedy: credits, refunds or termination rights, and whether support credits are the sole remedy.
  • People: named CSM, support engineer, backup and escalation owner.
  • Reporting: incident reports, RCA timing and QBR metrics.

Put the support schedule in the order form, not only in a sales deck. Then test it during the pilot: open a P3 with a known answer, ask for a P1 game-day escalation, and time the response. The Enterprise overview covers the commercial packaging; the docs cover the API; pricing shows what each self-serve tier includes; about explains the team behind the platform.

FAQ

What support comes with self-serve plans?

Ticket and community support, with no published response-time commitment. Production workloads should confirm coverage before relying on the free or low-cost tiers.

Does Plugsky publish P1–P4 response times?

The public enterprise page promises custom SLAs and a named CSM, and a security article states a 4-hour Enterprise support SLA. A full P1–P4 matrix is not published; agree definitions, response and resolution targets in the order form.

Is a named CSM the same as a named support engineer?

No. A CSM owns the commercial relationship and QBRs; a support engineer owns technical escalation. Ask for both roles, names and backup coverage.

What should an architecture review cover?

Data flow and residency, identity and SSO, key custody, rate limits, fallback behaviour, observability, disaster recovery and cost. It should produce a written record with owners and dates.

How do we escalate a P1?

Use the contracted channel, declare the severity, provide request IDs and impact, and ask for the incident commander. The contract should name the escalation path and update cadence.

Sources

Primary sources checked 20 November 2026. Provider terms, limits and prices change; verify the current version before you commit.

  • AWS Support plan comparison — publishes Enterprise first-response times of under 15 minutes for business-critical outages, under 1 hour for production down, 4 hours for impaired and 12 hours for system impaired.
  • Atlassian incident severity levels — defines severity by business impact, with examples for critical, major and minor incidents, and separates severity from priority.

Reviewed 20 November 2026. Plugsky's public support commitments are described on the enterprise and security pages; the binding schedule is in the order form. Confirm the current version before signing.