Agentic AI · engagement
How an agentic AI programme with Ajinkya is structured
A programme runs in four phases: scoping, pilot, production rollout, and managed service. Each phase has a defined scope, a timeline in weeks or months, and a named team shape you will work with on your side and ours.
What a pilot covers
One agent, one line or a defined asset set, one plant. The pilot delivers the agent running against live production data, the failure-mode catalogue for that agent, the human-in-the-loop gates, and the operational dashboards. Success criteria are agreed during scoping and measured against the plant\u2019s own production data, not synthetic tests.
What a pilot does not cover: multi-plant rollout, additional agent roles, deep custom integrations beyond one ERP and one MES connection. Those are the production-rollout phase.
What a production rollout covers
Multiple agents, multiple lines or plants, on the same platform with the same guardrails. Rollout covers integration hardening, data-quality remediation across plants, operator training, on-call handover, and the managed-service phase that begins at go-live.
Rollout also covers the honest conversation about which agents should never move out of advisory-only mode for your plant, based on what the pilot exposed.
Timeline per phase
| Phase | Typical duration | Primary deliverable |
|---|---|---|
| Scoping | 2 to 4 weeks | Target architecture, data-gap report, agreed success criteria |
| Pilot | 10 to 16 weeks | One agent live against production data with guardrails |
| Production rollout | 3 to 9 months, depending on plant count | Multi-agent, multi-line hardened deployment |
| Managed service | Ongoing, quarterly review cycle | Uptime, model retraining, policy tuning, on-call |
Team shape you will be working with
- Named plant sponsor with decision authority.
- One domain expert per agent role (reliability engineer, quality manager, planner, EHS officer, buyer).
- IT contact owning ERP, MES, CMMS and IIoT integrations.
- Data-quality owner for master data upkeep.
- Engagement lead accountable end-to-end.
- Solutions architect for target architecture and integrations.
- ML engineer(s) for agent design, guardrails and retraining.
- Data engineer for the shop-floor and ERP data pipeline.
- Site-reliability engineer once the agent moves into managed service.
FAQ
What does the pilot actually deliver?
A single agent, running against one line or a defined asset set, with the failure-modes catalogue, human-in-the-loop gates, and dashboards live. Success criteria are agreed in the scoping phase and measured against production data.
How long is a typical pilot?
The pilot window itself is 10 to 16 weeks. Add 2 to 4 weeks of scoping ahead of that. Faster is possible when the data foundation already exists (working IIoT layer, MES with clean genealogy, CMMS with real feedback loops).
What does the team look like on our side?
A named plant sponsor, one domain expert per agent (reliability engineer for maintenance, quality manager for quality, and so on), and an IT contact who owns the ERP / MES / CMMS integrations. See the "Team shape" section below.
What if the pilot does not clear success criteria?
The agent runs in advisory-only mode until it does, and the discovery-phase deliverables (target architecture, data-quality gap report, integration plan) stay with you.
How is production rollout different from the pilot?
Scope, not method. Multiple agents, multiple lines or plants, sharing the same platform and the same guardrails. Managed-service phase begins at go-live and runs alongside your team.
Talk to us about your programme
A 30-minute discovery call gets you a scoped answer on which agents fit your plant, what data your team would need to prepare, and what a pilot would look like.
Book a 30-min callReviewed by Amey Kadle, Founder, Ajinkya Technologies. Last reviewed: 2026-08-29.