A vapor recovery unit that runs at 78 percent capture instead of 94 percent does not announce itself. There is no alarm, no trip, no work order — just a flare tip that stays lit a few more hours each day and a gas sales meter that quietly reads lower than the reservoir engineer's model says it should. Across a 40-battery field, that gap is worth real money every single month, and most operators only discover it during an annual reconciliation. AI-driven VRU optimization closes that gap by continuously tuning compressor loading, suction back-pressure, and gas routing against live tank and header conditions instead of a static setpoint someone chose during commissioning. You can book a demo to see the capture-rate model running against your own tank battery data.
Flaring Is Rising Again — and Most of It Is Not a Technology Problem
The 2026 Global Gas Flaring Tracker published by the World Bank found that global gas flaring rose for the third consecutive year in 2025, reaching 167 billion cubic meters — the highest recorded level since 2019, and gas worth roughly 54 billion dollars burned without producing a single unit of useful energy. The report is blunt about the cause: what holds back progress is not technical but structural. The equipment to capture this gas already exists, is commercially proven, and is installed at thousands of sites today. The problem is that a large share of it does not run at the capture rate it was designed and permitted to deliver.
That distinction matters enormously for anyone who already owns vapor recovery assets. If flaring were purely a question of missing infrastructure, the only answer would be capital expenditure — more compression, more gathering, more pipeline. But when a VRU that was specified for 95 percent capture is actually delivering 75 or 80 percent because of nuisance trips, oversized compression cycling on and off, a back-pressure setpoint left at its commissioning value, or a routing valve that defaults to flare whenever the header wobbles, the gap is an operations problem wearing a capital-expenditure costume. Those points of capture are recoverable without buying a single new skid.
The economics have also shifted in a way that rewards attention here. The US Energy Information Administration's most recent Short-Term Energy Outlook places the Henry Hub spot price near $3.70 per MMBtu for 2026, and tank vapor is not lean pipeline gas — it is a rich, high-Btu stream carrying meaningful natural gas liquids content, which is precisely why recovered vapor typically prices well above residue gas on a per-Mcf basis. Every hour a VRU spends bypassed, tripped, or throttled is an hour that the richest gas on the lease is being converted into carbon dioxide and heat. Meanwhile the US Department of Energy has identified tank flaring as the second largest source of routine flaring in the upstream sector, which tells you exactly where the recoverable volume is concentrated.
Mapping the Split — Every Vapor Stream Ends Up in One of Three Places
Before optimization can mean anything, you need an honest picture of where low-pressure gas actually goes on a given lease. Vapor arrives at the suction header from several sources with very different flow characteristics, and the routing decision between sales and flare is made continuously, often by a simple pressure switch that has no awareness of what the rest of the facility is doing. The map below shows the typical structure of that decision on a producing tank battery. The percentages are illustrative of a mid-performing site rather than a measured field average, but the shape is consistent nearly everywhere.
The important insight in this map is that the 17 percent going to flare is not one failure. It is the accumulated residue of hundreds of small routing decisions made under conditions the original control scheme was never tuned for — a hot afternoon that raises tank vapor generation by a third, a truck loadout that spikes header flow for twenty minutes, a downstream gathering system that raises discharge pressure and pushes the compressor toward its trip point, an oxygen reading that drifts above the pipeline specification and forces a divert. Each of those events is individually reasonable. Collectively they are the difference between a good year and a mediocre one, and they are exactly the class of problem that continuous optimization handles better than a fixed setpoint ever will.
Six Reasons Vapor Recovery Units Underdeliver in the Field
When a VRU is specified, the vendor quotes a capture efficiency in the 90 to 95 percent range, and that number is genuinely achievable — under the flow, pressure, and gas composition conditions assumed in the sizing calculation. Field conditions drift away from those assumptions almost immediately. Understanding which of the following six failure patterns dominates on a specific site is the first diagnostic step, because they call for different responses, and the wrong response is expensive. Replacing a compressor that was only ever mis-tuned solves nothing while consuming a year of capital budget.
Notice that only one of these six is a hardware sizing problem, and even that one is often addressable through better load management and recycle control rather than replacement. The rest are control, visibility, and configuration issues — precisely the domain where a model that watches every tag continuously outperforms a periodic engineering review. An engineer reviewing a site quarterly sees a snapshot. A model watching the same site sees every temperature swing, every loadout, every gathering pressure excursion, and can correlate the routing decisions against the conditions that triggered them.
Tank Pressure Control Has Almost No Margin — and That Is the Whole Game
Atmospheric storage tanks operate within a pressure band measured in ounces per square inch, not pounds. The usable window between the point where the vacuum breaker opens and the point where the pressure relief valve lifts is genuinely narrow, and every meaningful loss mechanism in vapor recovery lives at one edge of it or the other. Push suction pressure too low chasing capture and you pull air into the tank, contaminating the gas and forcing a divert on oxygen specification. Let pressure ride too high to stay safely away from vacuum and you vent through the thief hatch or route to flare. Book a demo to see how the optimizer holds this band on live site data.
This is the structural reason that fixed setpoints leave capture on the table. A single setpoint has to be conservative enough to survive the worst condition the site will see — the hottest afternoon, the fastest loadout, the highest gathering pressure. That means for the vast majority of operating hours, the unit is running with margin it does not need, which translates directly into recovery it is not taking. A model that knows the current tank temperature trend, the loadout schedule, the compressor's actual available capacity, and the current discharge pressure can safely operate much closer to the favorable edge, and can pull back before an excursion instead of after one. The difference between reactive and anticipatory pressure control is measured in capture points.
There is a related trap worth naming. Some operators respond to oxygen ingress by adding gas blanketing to hold positive pressure in the tank headspace. It solves the contamination problem, but it does so by artificially raising tank pressure, which increases the volume pushed toward the relief path and consumes purchased or produced gas to do it. Blanketing has a legitimate role, but when it is used as a substitute for active pressure management it can move the loss rather than eliminate it — which is why the optimization target should always be the pressure trajectory itself, not a workaround layered on top of poor pressure control.
Five Levers the Optimizer Pulls, and What Each One Prevents
AI optimization in this context is not a black box making mysterious decisions. It is a model that continuously reads a defined set of process signals, predicts where each is heading over the next several minutes to hours, and adjusts a defined set of controllable variables to keep the site inside its recovery-favorable operating envelope. The table below lays out what the model reads, what it changes, and which specific loss mechanism each lever addresses. All of these are adjustments an experienced operator would make if they were watching every tag on every site continuously — which is exactly the part no human staffing model can deliver across a distributed field.
| Control Lever | Signals Read | What It Adjusts | Loss Mechanism Prevented |
|---|---|---|---|
| Compressor Loading and Speed | Suction pressure, motor current, discharge temperature, vapor generation rate | VFD speed, load step selection, recycle valve position | Short-cycling and capacity mismatch during peak vapor generation |
| Suction Back-Pressure Target | Tank pressure trend, ambient temperature, tank level, oxygen concentration | Suction pressure setpoint within the safe operating band | Both air ingress on the low side and thief hatch venting on the high side |
| Multi-Stream Routing Priority | Stream flow rates, Btu content by source, available compression headroom | Which vapor sources hold priority when capacity is constrained | Diverting the richest, highest-value stream when a leaner one could go instead |
| Discharge and Divert Management | Sales line pressure, compression ratio, gathering system trend, gas analyzer | Divert threshold timing, staged pullback, fuel gas diversion | Full flare diverts triggered by transient downstream pressure excursions |
| Predictive Intervention | Vibration signature, bearing temperature, oil condition, trip history patterns | Maintenance scheduling ahead of failure, pre-emptive load reduction | Unplanned compressor downtime and the flaring that continues until restart |
The fifth lever deserves particular emphasis because it changes the economics of the other four. Reliability analysis across production operations commonly puts the cost of site downtime above a thousand dollars per hour once deferred production is counted alongside compliance exposure, and a VRU that is offline does not merely stop recovering gas — it forces the entire vapor stream to flare for the full duration of the outage. Catching a developing bearing or valve issue days ahead of failure converts an unplanned multi-day flaring event into a scheduled intervention during a low-vapor window. That single capability frequently accounts for a larger share of annual capture improvement than all the setpoint tuning combined, because the losses it prevents are concentrated rather than distributed.
What a Recovered Mcf Is Actually Worth Once Everything Is Counted
The headline gas price is the least useful number in this calculation. Recovered tank vapor is rich gas, carrying substantially more heating value and liquids content per Mcf than dry pipeline gas, which pushes its realized value up. Working against that are regional basis differentials — Permian gas has repeatedly traded at steep discounts to Henry Hub — plus the electrical cost of compression, treating, and the operating and maintenance load of running the package. The stack below walks through a representative build-up for one incremental Mcf moved from flare to sales. The figures are illustrative modeling inputs, not a quoted guarantee, and every site's numbers differ with gas composition, power cost, and contract terms.
Now apply that margin to the actual variable AI optimization moves, which is not the price — it is the volume that reaches the meter. Consider a tank battery generating 420 Mcf per day of total vapor. At 78 percent capture, 92 Mcf per day goes to flare. Lift capture to 94 percent and that flared volume drops to roughly 25 Mcf per day, recovering about 67 Mcf per day of additional sales gas. At $1.85 per Mcf of net margin, that single battery generates roughly $124 per day, or about $45,000 per year. The number for one site is modest enough to be believable. Multiplied across a 40-battery field, it approaches $1.8 million annually — from assets already installed, already permitted, and already sitting on the lease.
The Compliance Floor Keeps Shifting — Recovery Economics Do Not
Operators evaluating flare reduction investment in 2026 face a genuinely unsettled regulatory picture in the United States, and that uncertainty has made some teams hesitant to commit. The honest reading is that the direction of travel is inconsistent at the federal level while state and international requirements continue to tighten independently. This is precisely the argument for grounding the investment case in recovered gas value rather than compliance avoidance: the revenue from a captured Mcf does not depend on which way a rulemaking goes, while the compliance benefit is a genuine but secondary bonus that arrives regardless.
There is a further consideration for anyone whose gas ultimately reaches export markets. Methane intensity requirements attached to European import rules are pushing verification of production-side emissions performance further up the value chain, and buyers increasingly ask for measured rather than estimated data. A site that already produces continuous, timestamped capture-rate data as a byproduct of optimization is positioned to answer those questions with evidence. A site relying on engineering estimates and monthly summaries is not. The measurement infrastructure that makes optimization possible turns out to be the same infrastructure that makes verification possible.
From First Data Tie-In to Closed-Loop Optimization in 90 Days
Nothing about this deployment requires taking a site down, replacing a compressor, or rewriting your safety instrumented system. The model reads from the tags your SCADA and historian already collect, adds instrumentation only where a genuine measurement gap exists, and begins in an advisory posture where every recommendation is reviewed by your operations team before anything changes. Autonomy is earned through demonstrated accuracy, not granted on day one. Book a demo to walk through what the tie-in looks like for your specific control architecture.
The phased structure exists for a reason that has nothing to do with technology readiness. Operations teams have been offered optimization platforms before, and healthy skepticism about a model touching production controls is entirely warranted. Running in advisory mode for several weeks produces something more valuable than a vendor claim — a documented record on your own site, with your own gas, showing what the model recommended, what happened when the recommendation was followed, and what happened when it was not. Teams that go through that process arrive at closed-loop operation with confidence rather than compliance, and that difference shows up in how consistently the system stays enabled a year later.
Flare-First Versus Recovery-First on the Same Tank Battery
The two operating philosophies below describe the same physical equipment. What separates them is whether flaring is treated as the safe default that engages whenever conditions get uncertain, or as a genuine last resort that the control system actively works to avoid. Most sites drift toward the first posture without anyone deciding to, because every individual divert decision is defensible in isolation and nobody is measuring the cumulative result. The comparison below is what changes when the cumulative result becomes visible and someone owns it.
| Dimension | Flare-First Operation | Recovery-First With AI Optimization |
|---|---|---|
| Routing decision basis | Fixed pressure switch with a single threshold set at commissioning | Continuous model weighing pressure trend, capacity, gas quality, and stream value |
| Response to peak vapor | Divert excess to flare until conditions normalize on their own | Anticipate the peak from ambient and loadout data, pre-position capacity |
| Compressor trips | Auto-restart after the fact, flaring for the full unplanned duration | Predictive alert days ahead, scheduled intervention in a low-vapor window |
| Capture rate visibility | Reconstructed monthly or annually from allocation and estimates | Continuous measured rate with every flaring event attributed to a cause |
| Oxygen and gas quality | Divert on analyzer alarm, investigate afterward if anyone notices | Pressure trajectory managed to prevent ingress before the alarm point |
| Regulatory reporting | Engineering estimates assembled at reporting deadlines | Timestamped operational record available continuously for any period |
| Capital response to a gap | Specify a larger compressor and wait on the capital cycle | Exhaust control and configuration gains first, size capital to the real residual |
The final row is where the argument usually resolves for capital planners. When a site is flaring more than its permit contemplated, the instinctive response is to specify more compression. Sometimes that is genuinely the answer. But committing capital before establishing how much of the gap is control-related rather than capacity-related risks buying horsepower to solve a setpoint problem — and the new unit will run into the same untuned back-pressure logic, the same undiagnosed trip pattern, and the same oxygen excursions the old one did. Establishing the true attribution first costs a fraction of a compression package and often changes the specification substantially.
