Common Questions.
Straight Answers.

IIA agents are designed as decision-support tools, not autonomous actuators. They analyze data, surface anomalies, generate recommendations, and document findings — but they don't turn valves, override control systems, or execute physical actions. Every action that matters goes through your engineers.

The reliability bar we set is: does this agent perform at or above the level of a competent domain expert working from the same data? In our deployments, agents consistently surface issues that manual review misses — not because the AI is smarter, but because it processes all available data without fatigue or cognitive shortcuts.

If you want to see it in action, run our Well Integrity simulator or the HSE & Permit to Work demo — both walk through safety-critical scenarios with full reasoning traces.

Agents are trained to escalate rather than guess when a situation falls outside their confidence range. When uncertainty is high, they surface the relevant data and flag it for engineer review — they don't hallucinate a recommendation to fill a gap.

IIA agents are also purpose-built for specific O&G workflows, not general-purpose AI retrofitted for the field. Each agent embeds 46 years of upstream and downstream expertise as its prior — so "edge case" for a generic AI is often baseline domain knowledge for ours.

The SEE Protocol simulator shows exactly how we extract and encode expert judgment, including how agents handle the ambiguous situations experts debate.

Every recommendation includes a full reasoning trace — the data sources used, the logic applied, and the confidence level. Your engineers can audit any output before acting on it, which means errors are catchable before they become incidents.

When errors do occur (and they do — this is real AI, not marketing), we track them as incidents, retrace the failure mode, and update the agent's behavior. IIA is engaged as a long-term operational partner, not a software vendor who disappears post-deployment. Our team has skin in the game when something goes wrong.

We also recommend a structured parallel-run period during deployment — your existing process runs alongside the agent so you can calibrate confidence before relying on it as your primary signal.

Standard deployment runs 2–4 weeks from signed contract to agents operating in your environment. The variation is almost entirely driven by data access: if you can give us a live feed from your systems in week one, we're live in week two. If data access requires IT tickets and procurement cycles, add time accordingly.

Contrast this with enterprise AI competitors who quote 6–18 month "transformation programs." Those timelines exist because they're building general-purpose infrastructure. IIA agents are pre-built for specific O&G workflows — we're configuring, not constructing. See the full breakdown at pricing.

Four phases, all within the 2–4 week window:

• Week 1 — Data Integration: Connect to your existing data sources (SCADA, historian, EDMS, spreadsheets). No rip-and-replace. We work with what you have.
• Week 1–2 — Calibration: Configure agent parameters to your specific assets, thresholds, and operating philosophy. If you're running Knowledge Capture in parallel, this is where captured expert judgment gets encoded.
• Week 2–3 — Parallel Run: Agents run alongside your existing process. Engineers compare outputs. Discrepancies get reviewed and agent behavior is tuned.
• Week 3–4 — Go Live: Primary handoff. Your team operates with agents as standard workflow. IIA stays engaged for the first 30 days of live operations.

No wholesale changes. IIA agents plug into your current tools — they surface outputs through interfaces your team already uses (email, Slack, your existing dashboards, or a lightweight web UI). We don't require you to adopt a new platform as a condition of deployment.

Over time, some workflows naturally evolve because the agents eliminate tasks that currently consume engineer time. That's the point — but it happens organically as your team builds confidence in the outputs, not as a forced change management project on day one.

Read-only access to the data sources relevant to each agent's function. For a Reservoir Management agent: production data, pressure/temperature readings, and historical decline curves. For Well Integrity: inspection records, completion data, and real-time sensor feeds. Agents never write to operational systems.

We work with whatever formats you have — structured databases, SCADA exports, flat-file historians, PDFs of hand-entered reports. Messy data is the norm, not the exception. We've never encountered a brownfield operator with clean data, and we don't require it as a precondition for deployment.

Most operators see positive ROI within 60–90 days of go-live. The mechanics are straightforward: IIA costs $60K–$120K per year per agent. A single NPT reduction event, a single well integrity catch before a workover, or a single well intervention avoided using better reservoir data typically returns that in one incident.

Use the ROI Calculator to model your specific asset base — it runs the math on your fleet size, average intervention cost, and current NPT rate. The GoM case study shows a documented example: $2.3M in avoided costs against a $90K annual contract.

Annual subscription per agent, priced by scope (number of assets, data volume, and support tier). Starter engagements run $60K–$120K per year for a single agent. Enterprise packages covering multiple agents and domains are available at volume pricing.

No per-query billing. No usage surprises. Your team can run the agents as aggressively as they want — the economics don't change. Full pricing details and what's included at each tier are on the pricing page.

Yes — and we recommend it. A focused 90-day pilot on a single agent and a defined asset set gives you real production data before committing to a broader deployment. Pilot terms are structured to convert cleanly to annual contract if results meet agreed benchmarks.

The pilot approach also works well politically. If you're trying to justify AI investment to a skeptical leadership team, a documented 90-day result — measured against metrics you defined before the pilot started — is far more persuasive than a vendor projection. Talk to us about structuring one.

The IIA Knowledge Capture process has three stages. First, structured elicitation sessions with your domain experts — not open-ended interviews, but a protocol designed to extract the heuristics and pattern-matching rules that experts apply but rarely articulate. Second, structured encoding: that knowledge is formalized into decision logic that an agent can apply consistently. Third, validation — the encoded logic is tested against historical cases where your expert's judgment is the ground truth.

The goal isn't to record what an expert knows in a PDF. It's to operationalize it: if your senior reservoir engineer retired tomorrow, the agent applies the same judgment they would have applied, to every well, every day, without forgetting or getting tired. See the Knowledge Capture overview and try the SEE Protocol simulator for a hands-on walkthrough.

Yes, completely. Knowledge captured from your experts, your data, and your operational history belongs to you. It is not used to train shared models, it is not shared with other operators, and it does not leave your contracted scope. Your competitive IP stays yours.

This is also a structural guarantee, not just a contractual one. The knowledge is encoded in agent configurations specific to your deployment — there's no pooled model layer where your data blends with anyone else's.

Interviews capture what experts think they know. The SEE (Structured Expert Elicitation) protocol captures what they actually do. Most domain knowledge is tacit — experts can't fully articulate their pattern-matching because they apply it automatically, without conscious reasoning. Standard interviews produce narratives about explicit rules. SEE produces the underlying decision logic.

The protocol works through scenario-based questioning — presenting specific cases and asking experts to walk through their reasoning in real-time, rather than describing their general approach in the abstract. The result is decision logic that generalizes: you end up with rules that apply to novel situations, not just the scenarios discussed in the session. Try the interactive SEE demo to see the protocol in action.

Not the way the question is usually framed. IIA agents eliminate the parts of engineering work that shouldn't require engineers: monitoring dashboards for threshold alerts, generating templated reports, cross-referencing compliance checklists, running standard decline curve calculations. Tasks that consume 40–60% of field engineer time and contribute nothing to outcomes that require human judgment.

What happens in practice: companies that were planning to hire 3 engineers to handle increasing portfolio size now need 1. Companies that already had 3 engineers can now manage a portfolio 3× larger. We have not seen a deployment that resulted in layoffs — the constraint isn't "too many engineers," it's "not enough bandwidth to do the high-value work." AI resolves that constraint.

If your concern is specifically about workforce reduction optics for a union environment or public-sector contract, that's a real consideration worth discussing directly. We've navigated it before.

The operational model is agent-surfaces, engineer-decides. Agents continuously monitor and analyze — they generate alerts, draft recommendations, and produce reports. Engineers review, act, override, or escalate based on their judgment. The agent is never the last decision-maker on anything that matters operationally.

In practice, this means your engineers start the day with an agent-generated priority list rather than spending the first two hours manually pulling together data. They spend their time on the problems the agent flagged, not on the data collection required to find those problems.

Minimal. IIA agents are designed to fit into existing workflows — the interface is whatever your team already uses (email notifications, a simple dashboard, or API integration with your existing tooling). There's no new platform to master.

What matters more than training is calibration confidence — engineers learning to know when to trust the agent's output and when to dig deeper. That comes from the parallel-run period during deployment, not from classroom training. By the time the agent goes live as a primary tool, your team has already seen dozens of examples of how it behaves and where it's strong or cautious.

If you want structured onboarding content, the Sales Training curriculum covers how IIA agents work, what they're good at, and how to position them internally.