Sigma · Production intelligence for waterfloods · Alberta 2026

Every field loses oil it never sees.

Production losses often happen between well tests. By the time they are detected, valuable oil has already been lost. Sigma uses the data you already collect to estimate oil and water production for every well, every day, so you can flag issues earlier, manage your waterflood with better visibility, and increase production.

~2,300 proration facilities ~31,000 producing wells ~38% of monthly filings off by more than 5%

Alberta oil proration batteries, July 2026. Size is oil produced; color is how far that month's test-based estimate sat from the meter. Petrinex public filings.

01Where the oil goes

The losses that hurt are the ones that keep running

A pump that stops is easily detected and gets fixed the same day. Other very significant production losses (several percent of total) are much harder to detect . A pump wearing down, water cut creeping up, gas getting into the pump, a sensor drifting. The well keeps running, nothing trips, and its rate on paper stays at the last test, taken weeks ago. At month end the battery still balances, because the difference is spread across all the wells by proration. The total is right, the well-level picture is wrong, and nothing says which well changed.

Well Alast test 40 m³/dstill running, no alarm Well Blast test 35 m³/d Well Clast test 32 m³/d battery meter 100 m³/d measured separator Tests add up to 107. The meter says 100. So every well is scaled by 100 ÷ 107 = 0.935. 37.430 +7.4 creditedbut never produced 32.738 −5.3 producedbut not credited 29.932 −2.1 allocated (last test × 0.935) actually produced battery total allocated 100 = produced 100 per well off by 7.4, 5.3 and 2.1 Nothing flags it.
How a running well loses oil unnoticed. Well A's pump has worn since its test. It still runs, so nothing alarms, but it now makes 30 m³/d instead of 40. The 10 m³/d difference gets spread across all three wells, the battery balances, and nothing flags Well A’s deficiency.
~2,300
oil and crude-bitumen proration facilities reported production, July 2025 – June 2026
~31,000
producing wells behind those facilities
~670,000
barrels a day passed through them over the year
~38%
of crude-oil monthly oil factors fell outside a ±5% reference band
Monthly correction
0.5%50% water injection
2,164 batteries reporting oil, July 2026. Placed at approximate township centers from Petrinex filings.

It shows in the province's own numbers

Every month, Alberta batteries file how far their well tests missed the meter. In a typical month the miss is several percent: 3.4% for crude-oil batteries and 6.8% for heavy oil. About 38% of filings miss by more than 5%.

Each miss means the well tests no longer added up to the meter. What the filing can't say is which well, or which combination of wells, caused it. Some is normal decline, but some could have been fixed weeks before the monthly report.

And the volume is concentrated: 90% of prorated oil comes from about 600 batteries, so one battery is a fair place to start.

The gap doesn't close on its own

Four years of filings show the same picture. The misses don't settle down, and they swing both ways, so no one-time recalibration fixes them.

Share of monthly oil factors outside the reference band

Crude oil, outside ±5%Crude bitumen, outside ±15%
0%10%20%30%40%50%2022202320242025Aug 2025Crude oilCrude bitumen2022-01: crude oil 40.8% outside ±5%, crude bitumen 27.5% outside ±15%2022-02: crude oil 39.1% outside ±5%, crude bitumen 34.5% outside ±15%2022-03: crude oil 37.6% outside ±5%, crude bitumen 28.9% outside ±15%2022-04: crude oil 37.0% outside ±5%, crude bitumen 28.8% outside ±15%2022-05: crude oil 38.3% outside ±5%, crude bitumen 25.7% outside ±15%2022-06: crude oil 39.8% outside ±5%, crude bitumen 34.5% outside ±15%2022-07: crude oil 38.8% outside ±5%, crude bitumen 34.0% outside ±15%2022-08: crude oil 40.2% outside ±5%, crude bitumen 29.5% outside ±15%2022-09: crude oil 40.4% outside ±5%, crude bitumen 27.0% outside ±15%2022-10: crude oil 41.5% outside ±5%, crude bitumen 30.5% outside ±15%2022-11: crude oil 41.9% outside ±5%, crude bitumen 27.7% outside ±15%2022-12: crude oil 41.9% outside ±5%, crude bitumen 34.4% outside ±15%2023-01: crude oil 40.5% outside ±5%, crude bitumen 26.6% outside ±15%2023-02: crude oil 39.6% outside ±5%, crude bitumen 22.4% outside ±15%2023-03: crude oil 40.1% outside ±5%, crude bitumen 21.3% outside ±15%2023-04: crude oil 38.7% outside ±5%, crude bitumen 21.9% outside ±15%2023-05: crude oil 42.8% outside ±5%, crude bitumen 27.9% outside ±15%2023-06: crude oil 42.8% outside ±5%, crude bitumen 26.9% outside ±15%2023-07: crude oil 39.8% outside ±5%, crude bitumen 27.6% outside ±15%2023-08: crude oil 38.4% outside ±5%, crude bitumen 25.6% outside ±15%2023-09: crude oil 39.5% outside ±5%, crude bitumen 20.7% outside ±15%2023-10: crude oil 38.9% outside ±5%, crude bitumen 25.0% outside ±15%2023-11: crude oil 39.1% outside ±5%, crude bitumen 25.9% outside ±15%2023-12: crude oil 36.6% outside ±5%, crude bitumen 22.9% outside ±15%2024-01: crude oil 40.0% outside ±5%, crude bitumen 24.1% outside ±15%2024-02: crude oil 36.5% outside ±5%, crude bitumen 22.7% outside ±15%2024-03: crude oil 38.1% outside ±5%, crude bitumen 16.4% outside ±15%2024-04: crude oil 37.9% outside ±5%, crude bitumen 16.9% outside ±15%2024-05: crude oil 36.7% outside ±5%, crude bitumen 13.1% outside ±15%2024-06: crude oil 37.7% outside ±5%, crude bitumen 21.2% outside ±15%2024-07: crude oil 37.2% outside ±5%, crude bitumen 23.5% outside ±15%2024-08: crude oil 37.7% outside ±5%, crude bitumen 20.6% outside ±15%2024-09: crude oil 37.6% outside ±5%, crude bitumen 20.3% outside ±15%2024-10: crude oil 39.0% outside ±5%, crude bitumen 20.2% outside ±15%2024-11: crude oil 36.5% outside ±5%, crude bitumen 22.3% outside ±15%2024-12: crude oil 37.2% outside ±5%, crude bitumen 24.5% outside ±15%2025-01: crude oil 37.8% outside ±5%, crude bitumen 22.0% outside ±15%2025-02: crude oil 38.4% outside ±5%, crude bitumen 20.6% outside ±15%2025-03: crude oil 37.3% outside ±5%, crude bitumen 17.5% outside ±15%2025-04: crude oil 36.2% outside ±5%, crude bitumen 26.5% outside ±15%2025-05: crude oil 37.6% outside ±5%, crude bitumen 27.8% outside ±15%2025-06: crude oil 38.6% outside ±5%, crude bitumen 25.6% outside ±15%2025-07: crude oil 36.8% outside ±5%, crude bitumen 24.0% outside ±15%2025-08: crude oil 36.7% outside ±5%, crude bitumen 23.3% outside ±15%

January 2022 to August 2025, every reporting battery. Crude-oil batteries against ±5%: 36 to 43% outside each month. Crude-bitumen batteries against ±15%: 13 to 35%.

250 largest batteries, 44 months. Each cell is one battery-month. Red: the meter read more than the tests predicted; blue: less. The corrections change direction from month to month, so a one-time recalibration can't remove them.

Reference bands are shown for scale, not as violation rates; some heavy-oil batteries have wider limits. A filing is a battery-level number. It says the picture was off, never which well changed, and that per-well picture is what Sigma provides.

02What Sigma does

Every well, every day, and what to do about it

Sigma sits on top of the systems you already run. It takes your SCADA signals, well tests, battery meters and injection data, works out how far each one can be trusted, and updates a daily estimate of every well's oil, water and gas. That one estimate does three jobs. It closes your monthly allocation, it tells you which wells are losing oil and when to act, and it guides injection and pump settings toward more oil for less water.

WHAT YOU ALREADY COLLECT SCADApump speed, run time, pressures Well testsoil, water, gas, every few weeks Battery metersdaily or monthly totals Injection datarates and pressures per injector ONE ENGINE Sigma Every well's oil, water and gas rate A confidence band on every number Each signal's noise and drift, learned Updates as new data arrive Seconds per field on one machine THREE USES Allocation and reportingmonthly volumes, closed to the meter Surveillancewhich wells are losing oil, and when to act Waterflood and lift controlinjection splits, pump set-points
Built on what you have. No new meters, no hardware. Sigma adds a per-well rate and its confidence band on top of your SCADA and production-accounting systems, and reads the noise in each signal rather than trusting the spec sheet.
More data, sharper
The more often your signals are logged, the sharper it gets. Going from weekly to daily pump data cut lost production from 5.6% to 4.9% and false alarms by 27%.
Seconds
Two years of daily data for 60 wells takes about ten seconds on an ordinary computer, so it refreshes as often as your data do.
Self-calibrating
Sigma learns how noisy and how drifted each signal really is, so a degrading sensor becomes a flag instead of a false reading.
03Proof

Tested where the right answer is known

You can't score an estimate against a filing, because the filing never says which well was wrong. So we scored Sigma on four fields where every well's true daily rate is known, two of them built from real field data, and gave it only what a real battery records.

Known truth every well, every dayoil, water and gasfrom a simulator orreal measured rates What a battery records 8-hour tests, weeks apartnoisy, drifting pump signalon-hours and downtimeone meter per battery Each method estimates today's prorationconventional reconciliationtuned filter · ML meterSigma Scored against truth year 1: calibrateyear 2: score, unseenper well, per monthand per day truth goes straight to scoring, hidden from every method
Same test for every method. Nothing sees the true rates. Each method learns from the first year and is scored on the second.
Statistical simulation

Synthetic proration fields

Eight fields of 60 pumped wells in five batteries, two years each. Their proration corrections average 6.6%, in line with Alberta crude-oil batteries (6.4 to 6.7%).

Pictured: one well's true rate, its noisy pump signal and its tests.

Reservoir simulation · OPM Flow

Heavy-oil polymer floods

Two 1,500 cP heavy-oil polymer-flood patterns with 24 producers each, one with a high-permeability channel. Closest of the four to an Alberta heavy-oil waterflood.

Pictured: injected water spreading through pattern 2, July 2019.

Real field model

Norne

The public model of the Norne field, with its real geometry and production schedule, run in OPM Flow. Nine producers in two batteries, 2004 to 2005, with gas present.

Pictured: remaining oil at the top of the reservoir, 2006.

Real measured rates

Volve

Equinor's public Volve data: real daily oil, water and gas for five producers, April 2014 to April 2016, with a real surface signal from the choke. Tests and battery meter are simulated on top.

Pictured: real daily oil by producer; shaded is the test window.

04Results

Less oil lost, fewer wasted trips

On heavy-oil polymer floods with pumps that wear and fail, we simulated the whole loop: Sigma raises an alert, a crew goes out, the well comes back. Here is what it recovered, compared with finding failures at the next test or from raw pump-signal alarms, and estimate accuracy.

7.1% → 4.9%
of production lost to failing pumps, compared with waiting for the next test
76% fewer
wasted call-outs than raw pump-signal alarms, for the same oil recovered
−21%
error in each well's monthly oil, against today's allocation
−56%
error in each well's monthly water, the number a waterflood is steered by

Oil recovered, and the trips it took

Production lost0%4%8%False alarms / yr01020Tests onlyTests only: 7.1% of production lost7.1%Tests only: 0.1 false alarms per 20 wells per year0Pump-signal alarmPump-signal alarm: 4.9% of production lost4.9%Pump-signal alarm: 18.2 false alarms per 20 wells per year18SigmaSigma: 4.9% of production lost4.9%Sigma: 4.3 false alarms per 20 wells per year4Perfect knowledgePerfect knowledge: 3.7% of production lost3.7%Perfect knowledge: 0.2 false alarms per 20 wells per year05-well batteries, daily data, per 20 wells per year

Heavy-oil polymer floods, 5-well batteries. Same crew, same repair policy; only the alert behind it changes. Sigma recovers as much oil as a raw pump-signal alarm with 12 fewer call-outs a year per 20 wells, and 2.2 points more production than waiting for the next test.

Where the money comes from

Pumps that fail partway cost a heavy-oil field several percent of its production when they're only found at the next test. Sigma finds them in days instead of weeks, and gets most of that oil back.

False alarms are also costly. Raw pump signals are noisy and drift over time, leading to false alarms that send crews to service wells that are fine. Sigma’s intelligence processing can much more accurately discriminate a real change from signal noise, so that failures are caught with only a fraction of the wasted trips.

This is the size of gain operators see in the field when they move to well-by-well surveillance, with deferment falling from 4% to 1.8% in one published case (SPE-APOG 2021).

The estimates behind it

Scored on a year the model never saw, well by well and month by month, against the known volume.

Change in monthly per-well error, compared with today's proration

OilWater−60%−40%−20%0← lower error than today’s prorationSynthetic, 8 fieldsSynthetic, 8 fields: oil error -21% vs proration−21%Synthetic, 8 fields: water error -56% vs proration−56%Synthetic, noisy pump signalSynthetic, noisy pump signal: oil error -17% vs proration−17%Synthetic, noisy pump signal: water error -50% vs proration−50%Synthetic, quarterly testsSynthetic, quarterly tests: oil error -17% vs proration−17%Synthetic, quarterly tests: water error -53% vs proration−53%Heavy-oil polymer flood 1Heavy-oil polymer flood 1: oil error -14% vs proration−14%Heavy-oil polymer flood 1: water error -11% vs proration−11%Heavy-oil polymer flood 2Heavy-oil polymer flood 2: oil error -11% vs proration−11%Heavy-oil polymer flood 2: water error -5% vs proration−5%Norne field modelNorne field model: oil error -70% vs proration−70%Norne field model: water error -63% vs proration−63%Volve, real daily ratesVolve, real daily rates: oil error +1% vs proration+1%Volve, real daily rates: water error -12% vs proration−12%

Water improves on every field; oil on six of seven. The gain scales with the problem. On the smooth polymer floods, where proration is already within a few percent, Sigma still cuts the error by 11 to 14%. On pumped wells like Alberta's it cuts oil error by a fifth and water error by more than half.

Sigma beats other methods

Oil-40%0%50%100%D017 baselineceilingWater-40%0%50%100%D017 baselineceilingSigmaSigma: oil error 15.0%, 92% of the possible improvement over D01792%Sigma: water error 4.1%, 98% of the possible improvement over D01798%SOTA KFSOTA KF: oil error 15.2%, 86% of the possible improvement over D01786%SOTA KF: water error 4.6%, 87% of the possible improvement over D01787%Std KFStd KF: oil error 16.7%, 51% of the possible improvement over D01751%Std KF: water error 6.0%, 62% of the possible improvement over D01762%ML VFMML VFM: oil error 33.3%, -334% of the possible improvement over D017-334% ›ML VFM: water error 10.2%, -17% of the possible improvement over D017-17%D0170%0%

Share of the possible improvement over D017, synthetic fields. Zero is today's proration and 100% is a filter given the true noise settings. Sigma captures 92% on oil and 98% on water, ahead of a state-of-the-art Kalman filter tuned to each well, and pulls further ahead on water, on the heavy-oil floods, on daily estimates and on tracking pump drift. The ML meter falls below proration for lack of training data. SOTA KF, state-of-the-art Kalman filter tuned to each well. Std KF, standard Kalman filter. ML VFM, machine-learning virtual flow meter. D017, today's proration.

One well, month by month

050100150200250123456789101112Month of the scored year (oil, m³ per month)ProrationSigmaActualMonth 1: actual 71, Sigma 76 (95% band 58–94), proration 93 m³Month 2: actual 130, Sigma 102 (95% band 79–133), proration 95 m³Month 3: actual 158, Sigma 166 (95% band 135–207), proration 197 m³Month 4: actual 108, Sigma 118 (95% band 97–137), proration 158 m³Month 5: actual 80, Sigma 82 (95% band 70–98), proration 55 m³Month 6: actual 66, Sigma 73 (95% band 62–86), proration 94 m³Month 7: actual 96, Sigma 91 (95% band 65–126), proration 79 m³Month 8: actual 104, Sigma 79 (95% band 57–107), proration 73 m³Month 9: actual 95, Sigma 88 (95% band 67–113), proration 109 m³Month 10: actual 99, Sigma 80 (95% band 64–101), proration 92 m³Month 11: actual 96, Sigma 96 (95% band 78–117), proration 95 m³Month 12: actual 60, Sigma 72 (95% band 58–88), proration 82 m³

One synthetic well (well 28), scored year. Average monthly error: proration 26%, Sigma 11%. The shaded 95% band held the actual volume in 12 of 12 months.

Between tests, the error grows more slowly

0%10%20%30%test day1–34–78–1415–2122–3030+Days since the well was last testedProrationSigmatest day days since test: proration 9.1%, Sigma 8.3% daily oil error1–3 days since test: proration 12.3%, Sigma 10.6% daily oil error4–7 days since test: proration 16.7%, Sigma 12.9% daily oil error8–14 days since test: proration 21.0%, Sigma 15.1% daily oil error15–21 days since test: proration 25.4%, Sigma 17.3% daily oil error22–30 days since test: proration 27.9%, Sigma 19.0% daily oil error30+ days since test: proration 30.2%, Sigma 19.8% daily oil error

Daily oil error by days since each well's last test, synthetic fields. After a month without a test, proration is about 30% off and Sigma about 20%.

When the pump signal is noisier than its spec

0%10%20%30%40%50%2× spec4× spec6× specReal noise in the daily pump signal, vs its stated specD017Std KFSOTA KFSigmaPump signal 2× spec: D017 18.9%, Std KF 16.7%, SOTA KF 15.2%, Sigma 15.0%Pump signal 4× spec: D017 19.1%, Std KF 28.6%, SOTA KF 16.1%, Sigma 15.9%Pump signal 6× spec: D017 19.1%, Std KF 44.6%, SOTA KF 16.5%, Sigma 16.1%

Monthly oil error, synthetic fields. Conventional reconciliation trusts the stated spec and gets worse than doing nothing. Sigma reads the noise off the data itself and holds steady.

One engine, tuned per field. Sigma computes its setting from each field's own first year of data, so the results above use the setting that suited each field. Sigma is built for pumped onshore batteries like Alberta's, where wells are tested every few weeks and allocated by proration, and it carries over to offshore fields run on chokes, as it did on Volve's real rates, where it improved water accuracy. Sigma's benefit on individual batteries can be forecast from historical data from that production site.
05Value

What it's worth to your company

Estimate how Sigma can improve your own operation by dragging the sliders below. The estimates are based on Sigma's results on heavy-oil fields calibrated to Alberta production data. Contact us to get a more accurate estimate based on your data.

How you monitor wells today
Typical battery size
Losses prevented

Failures caught early, fewer wasted trips

Oil recovered each year0
Call-outs per year, change0
Net value per year0

Production gained

Online optimization of injection and lift

Extra oil each year0
Its value0

Losses prevented uses the results above against the monitoring you chose, for your battery size. It counts the change in production lost, in call-outs, and in C$300 confirmation tests per well. Against pump-signal alerts most of the value is trips avoided, and against tests it is oil recovered. Production gained applies your chosen uplift to your production. Operators have published 1 to 5% from pump set-point optimization and about 2% more oil with 22% less injected water from pattern-level injection reallocation (SPE-187468). Cleaner allocation for royalty and partner accounting comes on top and isn't counted here.

Both figures assume full-scale implementation of both phases, the daily well estimates and the optimization workflow built on them.

06Running the waterflood

From knowing every well to running the flood on it

Most of a waterflood's decisions run on numbers that are weeks old. Which patterns get the water, how hard each pump runs, which well gets tested next. With a daily rate and confidence band on every well, Sigma turns those into daily recommendations, so the water goes where it most efficiently moves oil, the pumps sit at their best drawdown, and the crew's day starts with a ranked list.

SHALE CAP ROCK HEAVY OIL SAND water / polymer high-permeability streak: injected water bypasses the oil and reaches P2 Injector P1 P2water cut rising P3pump wearing Battery meter + test separator Optimizer next phase Sigma estimate daily rate ± band, every well rates + uncertainty daily data recommendations 1 reallocate injection 2 tune speed 3 flag & rank 4 test P2 next
One loop, four levers. Daily data from the battery feeds the estimate. Its per-well rates and confidence feed the optimizer, which sends recommendations back to the injectors, the pumps and the test schedule.
1

Injection optimization

Move water or polymer toward the patterns that respond and away from the ones that only cycle water. Operators report about 2% more oil with 22% less water injected from this alone.

2

Pump set-points

Speed and stroke, well by well, to hold each well at its best drawdown without pumping off or over-lifting water.

3

The morning list

Which wells to adjust remotely, which to visit, and which are worth a rig, ranked by the oil at stake.

4

Tests where they count

Test the wells Sigma is least sure about first, instead of on a fixed rotation, so every test separator hour buys information.

07Getting started

One battery, twelve weeks, a number you can check

Nothing to install. We work from a copy of the data you already keep, and you see what Sigma would have caught on your own wells before you commit to anything.

Weeks 1–2

Data review

What's recorded, and how often.

Weeks 3–6

Calibrate

Fit on early history only. Later months stay sealed.

Weeks 7–10

Parallel run

Daily estimates beside your process, checked against new tests.

Weeks 11–12

Report

Accuracy, alerts and oil saved, and a go / no-go.

You send
Well testsDaily battery volumesOn-hours & downtimePump speed / fluid levelInjection ratesTwo years of history
You get
  • Daily oil and water for every well
  • Monthly allocation that closes to your meter
  • Drift flags on pumps whose signal has gone stale
  • A value ledger with every alert, what was done, and what was at stake
You decide
No measurable value, nothing further owed.

And we tell you why it didn't work on your battery.

Pump speed and fluid level are often logged weekly rather than daily. Weekly still works; we measured both cases and will tell you in advance what to expect from yours.

Next step

Send one battery. We'll tell you if it's a good candidate.

Two years of well tests, battery volumes, operating hours and whatever pump history you have. We first check whether those signals carry enough information to make a pilot worthwhile, then replay them to show which changing wells Sigma would have caught, how many days sooner, and what was at stake. That first look costs you nothing but the export.

Check one battery See the results again