By Saad Iqbal
Picture a mature field with four hundred wells and three production engineers. Nodal analysis — the technique of finding where a well’s inflow performance relationship crosses its tubing performance curve — has been the right way to diagnose a underperforming well since the 1980s. The problem has never been the method. It’s the math: by the time an engineer has hand-built and tuned a model for well number thirty, the data behind well number one is already three weeks stale. The well that’s quietly losing 15 barrels a day to a partially plugged gas-lift valve sits in the queue behind forty other wells, waiting its turn for attention that may not come for a month.
That queue is exactly what AI-assisted nodal analysis and production-system-optimization tools are built to eliminate. Instead of an engineer manually rebuilding a model well by well when time allows, these platforms keep an inflow-outflow model continuously tuned against live surface data across an entire field, and surface only the wells where the gap between theoretical and actual rate is both real and fixable. The nodal analysis itself hasn’t changed. What’s changed is how often it gets run, and on how many wells at once.
Why manual nodal analysis stalls at scale
Nodal analysis works by plotting the inflow performance relationship (IPR) — what the reservoir can deliver — against the tubing performance curve (TPC) — what the wellbore and surface equipment will allow — and reading the operating point off their intersection. Move the intersection with a bigger choke, a deeper gas-lift injection point, or a smaller tubing string, and you’ve found your production uplift. The physics is well understood and has been for decades.
What breaks down is the refresh cycle. A few recurring problems show up across most upstream teams running this manually:
- Models get built once at well startup and rarely revisited until production drops enough to trigger a manual review — by which point days or weeks of recoverable rate are already gone.
- Engineers default to whichever wells are easiest to model (good instrumentation, simple completions) rather than the wells with the biggest actual opportunity.
- Multi-zone or commingled wells get modeled as a single lumped system because reconciling zone-level inflow by hand for every well in a portfolio isn’t practical on top of everything else on an engineer’s plate.
None of this is a modeling-skill problem. It’s a throughput problem, and throughput is exactly what automation and AI are good at fixing.
What separates a real AI nodal analysis tool from a dashboard
- Physics-based, not black-box. The credible tools in this space still run multiphase-flow correlations and IPR/VLP physics under the hood — AI is used to keep the models tuned and to flag which wells need a human look, not to replace the underlying engineering with a pattern-matching guess.
- Field-wide, continuous refresh. The entire value proposition is running the comparison across every well, routinely, rather than well-by-well on demand.
- Works from the instrumentation you actually have. A platform that requires permanent downhole gauges on every well to function is a non-starter for most mature onshore assets; look for tools built to estimate flowing bottomhole pressure from surface data and periodic well tests.
- Automation hooks. Can it be scripted (Python toolkit, API) so a sensitivity sweep across hundreds of wells runs overnight instead of one case at a time in a GUI?
Five tools worth putting in front of your team

1. SLB Pipesim
Pipesim is SLB’s steady-state multiphase flow simulator and remains the most widely deployed nodal analysis engine in the industry, covering well performance, artificial lift design, and network/facility modeling in one package. Since the 2017.1 release, its Python toolkit has supported scripted, batch nodal-analysis runs across many wells at once — the feature that turns Pipesim from a per-well desktop tool into something that can realistically cover a large portfolio on a schedule rather than on request.
Best for: teams that already have Pipesim models built and want to automate the refresh cycle with Python rather than switch platforms.
2. Petroleum Experts IPM Suite (PROSPER + GAP)
Petroleum Experts’ PROSPER handles well-level nodal analysis — IPR and vertical lift performance — while GAP extends the same modeling into full multiphase network optimization across a gathering system. The combination is well-centric at the PROSPER layer but field-wide once linked into GAP, which is useful for teams that need both the detailed single-well diagnosis and the network-level view of how fixing one well’s choke setting shifts backpressure on its neighbors.
Best for: reservoir and production engineering teams who need an integrated well-to-network model, not just isolated well diagnostics.

3. S&P Global Harmony Enterprise (Harmony Optimize)
Harmony Enterprise bundles decline-curve analysis, rate-transient analysis, and nodal analysis into a single platform, which matters because production engineers rarely use nodal analysis in isolation — they’re usually cross-checking a nodal diagnosis against a decline trend or an RTA-derived permeability estimate for the same well. Its nodal module, Harmony Optimize, is built specifically to let engineers compare multiple flow-path configurations and lift-equipment scenarios for a well over time, rather than a single static snapshot.
Best for: reservoir engineering teams who want decline, RTA, and nodal analysis cross-referenced in one project rather than reconciling outputs from three separate tools.
4. Cognite Data Fusion (the platform behind ADNOC/SLB’s AiPSO)
Cognite Data Fusion is the industrial data platform underpinning AiPSO, the AI-powered Production System Optimization system ADNOC and SLB deployed across eight fields, aiming toward full-field deployment by 2027. AiPSO layers SLB’s AI and machine-learning models on top of Cognite’s contextualized data fusion to proactively monitor and optimize an entire production system end-to-end, rather than one well’s nodal diagnostic at a time. It’s a glimpse of where field-wide nodal and production-system optimization is heading at supermajor scale — continuous, not periodic.
Best for: larger operators with the data infrastructure and in-house analytics capability to build toward continuous, field-wide production system optimization rather than well-by-well review.

5. petropt (Python)
For teams that want to build their own lightweight screening layer before committing to a commercial platform, petropt is an open-source Python library covering PVT correlations, IPR construction, decline-curve analysis, and multiphase flow/friction calculations. It won’t replace a full commercial nodal-analysis package for detailed design work, but it’s a genuinely useful way to script a first-pass IPR/VLP screen across a well list in Jupyter before deciding which wells deserve a full PROSPER or Pipesim model.
Best for: engineers who want a free, scriptable first-pass triage tool to prioritize which wells get full nodal-analysis treatment in a commercial package.
Pairing this with the fundamentals
Every tool above is automating the same intersection you’d find by hand. If nodal analysis itself is still a bit fuzzy, Nodal Analysis: Find Your Well’s Operating Point walks through the IPR/VLP intersection manually before you hand the repetition to software. And because gas lift is one of the most common levers nodal analysis identifies for uplift, it’s worth reading Best AI Plunger Lift Optimization Tools for Gas Wells alongside this piece if liquid-loaded gas wells are part of your portfolio — plunger and gas-lift optimization are usually the next step after a nodal diagnosis flags the well.
If your production system also runs on artificial lift at scale, it’s worth cross-checking candidate wells against the ESP side of the house too — see Best AI Tools for ESP Failure Prediction & Run-Life Optimization for the lift-equipment half of the same production-optimization picture.
The bottom line
The honest pitch for AI in nodal analysis isn’t that it discovers new physics — IPR and VLP curves work exactly the way they did forty years ago. The pitch is throughput: a platform that keeps every well’s model current, instead of the handful an engineer has time to revisit this month, turns nodal analysis from a diagnostic you run when a well is already visibly underperforming into a screening tool that catches the loss while it’s still small. Whether that means scripting Pipesim’s Python toolkit, linking PROSPER into GAP for a network view, or going all the way to a continuously-tuned platform like AiPSO depends entirely on how many wells you’re responsible for and how much engineering time you actually have per well per month.
Common pitfalls to watch for
- Trusting a continuously-updated model on a well with no recent well test to anchor it — AI-refreshed models still drift without periodic ground-truth data.
- Treating a flagged “underperforming” well as an automatic workover candidate — nodal analysis tells you where the bottleneck is, not whether fixing it is economic.
- Skipping the zone-level reconciliation on commingled wells because the field-wide tool defaults to a lumped model — check whether your platform actually supports multi-zone IPR before trusting its allocation numbers.
