Work management systems suck

Why assignment of work is important:

In the previous workflow article, we talked about work identification. Here we will talk about linking work identification to work assignment.

A system to assign work reduces time & resources searching for what to do, enables understanding of the total work everyone does on a daily basis and creates pathway to decision intelligence (what should be done to resolve the issue)

Without a system to assign work, an operation will be heavily reliant on human detection, which creates a bottleneck in the system around time and capabilities.

MAIN POINTS SUMMARIZED

1. Current work management software is manual based, creating human bottlenecks in the operation

2. Automating the assignment of work HAS to be built on an accurate and intelligent system. Creating a “logic/exception” based system fails due to amount of false positives and false negatives assigned to the user, thus creating frustration and lack of trust.

3. Linking time series data to work management & work documentation enables future decision intelligence.

The 3 different levels of work assignment

1. Worst: Scada alarms & manual scrolling through wells

• Traditionally all work is dictated by field operator (which isn't bad in its own right). But this is usually because

(A) There isn't a great way to combine all the requests/issues/alarms/optimization in one spot and done intelligently.

(B) Usually the issue is not specifically known and therefore problem identification starts at the site because the system is not smart enough to figure it out...but what if a system did!

(C) If there is no assignment system then how does the operation have any oversight on what work should have taken place vs what work was done

2. Not great: Pump by Exception / Control rooms / Work management software

• "Pump by exception" is currently set up to fail mainly because the system is not “intelligent”. Creating a bunch of “rules” creates a lot of wells assigned to operators that are really false positives (see below pictures). This creates a lot of frustration at the field level and actually decreases field operations performance (trust us…we can talk days and days about this one!)

• The first step of pump by exception is not visiting every well every day, which is the obvious first step. If that's the case, the next question is, what wells should be visited?

• What is an exception? SCADA alarms, short-term production decreases? I have seen operations that have 300 alarms in one area in one day. How is that helpful & who is keeping track of what action was taken on what alarm? In our analysis across multiple operators, we have found scada alarms to have a 75% false-positive rate! That means that alarms are more often unactionable than actionable. This results to everyone being less efficient than more efficient, as they are being assigned a well that does not have an issue!

• There is no feedback mechanism from users to recorrect the "system" and for the system to learn

• Control rooms are a better way to align what work needs to happen and assign the work to the right resource. However, this is still very human directed and creates a bottleneck in the system with their time & capabilities

• Work Management Software is a better system to track work that should be done and work that gets done. However, this still funnels through the human bottleneck of people assigning work vs a system. As wells increase, G&A still increases with this model.

The normal work management philosophy measures what is complete, at Tasq we measure what is effective. These are two totally different things.

3.Best in class: Machine Learning based (This is where Tasq thrives)

• Model based: Logic (if/then) rules can only take you so far as data is inherently messy. Creating models to understand normal conditions and make recommendations reduce the noise and enhance accuracy of the system. There are a lot of false positives created from poor production targets & scada alarms that lead to unnecessary site visits. Tasq increases the accuracy of work distribution by understanding the specific data signatures, then assigning the specific action.

• User feedback: THE key component of utilizing models to enhance work assignment is giving users (that receive the job) the "ability to re-train the model" so assignments have an automated continuous cycle improvement. Without correction there will be a lot of frustration! Below is an example of a well target model getting corrected by the user after the well was optimized to create a new baseline.

• Connected data sources: Utilize multiple data sources in a single model to discover what the issue is, not just a non-descriptive alert. Is this issue a worn plunger or a bumper spring issue (each have different action paths). Maybe a well doesn't need to be visited and a setpoint can be changed remotely, maybe a well should directly be scheduled to workover). This enables work to be automatically directed to the right person to action the job vs always going through the field operator for a site visit, thus reducing work out of the system.

Use cases of model driven approach at scale:

A. Understanding if a well is producing what it should be producing

B. Understanding abnormal operating conditions

C. Understanding what other work needs to be done (PM, scheduling, workovers…..)

D. Understanding if a well is optimized or not

(this will be a whole other article coming soon!)

‍

Lasting thought

How many systems does it take up to do one workflow....Scada alarm → spotfire → excel → Field Data Capture → operations report. The problem is everyone is using +5 different source systems and still being uneffective at projecting what action should take place to get the well to an optimal state. Excel to document/communicate what jobs should be done today, Scada to review manually review wells. No way to label data. No way to document work and tie it back to time series data. Not a great way to manage FDC inputs. Not a great way to utilize the expertise of your rock solid field technician. No way to scale the learnings. In effect, no way to learn at scale.

NEXT UP

How does would a work management system have a continuous improvement loop? How can an organization learn from data & effectiveness to drive decision intelligence?

‍

‍