Energy and Predictive Analytics


SafeRock uses predictive analytics and machine learning algorithms to forecast oil and gas production for public and private operators. Our system contains custom decline curves for over 15,000 wells in Eagle Ford and Permian. We dynamically generate forecast new well output, as well as EUR and future production for any collection of wells.
E&P: we help management identify growth opportunities, accurately measure EUR values and production performance, and improve internal budget and resource allocation to maximize ROI.
Service Providers: we help providers optimize resource allocation for their teams and projects, and track and compare performance of past wells.
Investors: We help investors detect anomalies, investigate production problems and identify key performance issues that should be discussed with management. Our deep well-by-well analysis improves investment decisions.
For brands and retail, our predictive algorithms accurately forecast total item and category sales. 

SafeRock spots Carrizo’s (NASDAQ: CRZO) stock surprise 2 months in advance

News Update
News Update: 
February 28, 2018 
On February 26, 2018, Carrizo’s (NASDAQ: CRZO) stock tumbled -23% and the company reduced the value of its Eagle Ford EUR assets by -14%. SafeRock had identified Carrizo’s production anomalies in December, two months earlier. 


For investors, our system detects production anomalies, generates dynamic type curves, and identifies bad wells. They can investigate production problems and be ready to question management on their earnings calls. They can use this deep well-by-well understanding to improve investment decisions. 
E&P management can identify the greatest growth opportunities and do a better job of budgeting resources where they are needed.
In the Oil and Gas sector our system:
  • Calculates EUR for oil and gas wells
  • Automatically spots anomalies, outliers, and re-completions
  • Forecasts expected production for the next 1, 2, and 6 months
  • Uses machine learning to calculate an optimal ARPS decline curve for each individual well
  • Project: System beats market consensus to the punch and forecasts gas and oil production

    – A top-rated US energy analyst firm wanted to improve production forecast accuracy for public E&P operators in Lower 48.

    -Using purely public data, we built a unique process using proprietary algorithms that takes well-level monthly oil and gas production data from E&P operators and builds dynamic models for each and every well.

    -The new system improved accuracy of forecasts by 2/3.
    -It dramatically shortened calculation time for estimating EUR and production forecasts for oil and gas.
    -It provides KPIs, well production profile, and production forecasts for oil and gas.
    -The process proved to be a true data-driven process that completed forecasts by 1,000% faster than before.

  • Project: System beats Experts in forecasting accuracy

    -The challenge was to improve the accuracy of EUR estimates so that our system could be more accurate than EUR calculated by engineering experts.

    -We agreed to build an automated system for calculating well-specific EUR. The ARPS model was used as a base.
    -We had engineering EUR estimates of several hundred wells, and the goal was to estimate this more accurately than an expert using the same data.

    -The new system was more accurate than expert estimates over 2/3 of the time in three different sets of well comparison.

  • Project: Custom Type Curves created for any set of wells

    -Type curves are used almost universally in the energy industry to estimate area production, new well output, represent production profiles, in corporate reports. However ,their limitation is that it cannot distinguish on a finer basis.
    -Our goal was to generate dynamic type curves for any given set of wells.

    -We built an automated system to calculate a custom Type Curve for any given set of wells.
    -ARPS model was used as a basis for calculation.
    -The significant advance was to taken any set of wells _ from 2 to 2,000 – and automatically generate representative type curves.

    -We succeeded in building dynamic type curves that were more accurate than existing static curves in predicting output.

  • Project: Fast Big-data System rapidly models one new Well per minute

    -The challenge was to turn around well estimates faster than traditional methods while maintaining accuracy.

    -System took standard public data, made APRS based models, and ensured best fit to historical production.

    -Generated dynamic well models at the rate of one per minute.
    -This process is systematic and automated, and produces unbiased results.


For Brands and Retail companies our system:
  • Integrates marketing and merchandise planning with online traffic to reduces out-of-stock
  • Accurately forecasts retail sales using predictive algorithms and machine learning
  • Improves both marketing ROI and return on ad spend