Key Features of EDMCS 18.10

Usually, when October rolls around each year, there are three things I am very excited about:

  1. The start of the NHL season
  2. Watching one of my favorite scary movies (Halloween ’78)
  3. Eating leftover Halloween candy for the next six months because we bought 200 pounds of it at Costco and only had 5 trick-or-treaters show up at our door

But this year, add #4 to the list: The October 18.10 release of Enterprise Data Management Cloud Service (EDMCS)! Trust me, this is a monster of a release (pun intended). This is the biggest release since 18.07 and here are some of the highlights:

  • New packaged adapter for Oracle Financials Cloud GL
  • Property Inheritance
  • Add Related Nodes Across Viewpoints
  • Download Viewpoint in Request Context
  • Request Load File Summary Statistics
  • Level Property
  • Metadata Deletion
  • Object Details Popup

Here are more details and insights on a few of these features:

Oracle Financials Cloud GL Adapter

This is a major adapter in terms of functionality. And like the packaged adapters for Planning and Budgeting Cloud Service (PBCS) and Enterprise Planning and Budgeting Cloud Service (EPBCS), you can create an EDMCS application for Oracle Financials Cloud GL using a standard wizard interface, identify a connected application or use a file interface, and reap the benefits of several built-in validations.

The registration process includes steps to identify basic settings (e.g. Active Languages, Active Trees, Multiple Active Tree Versions, and Max Depth), register multiple Segments and Trees, and add/modify/remove Financial Categories. From reviewing the Oracle EDMCS Administrator Guide, I identified at least 16 built-in validations for EDMCS applications based on the Oracle Financials Cloud GL Adapter. This is certainly a packaged adapter with a lot of “meat on its bones.”

Property Inheritance

For those who love Oracle Data Relationship Management (DRM), you likely used the inheritance features of that product extensively – whether it was the global, inheriting property definition or using functions like ParentPropValue(). Well now you have property inheritance in EDMCS.

Property inheritance is set during the Application Registration process by modifying custom properties within your dimension node types. The choices are “None” or “Positional” which work as you would expect. Inherited property values can be overwritten in a request for exception cases. The other feature I like is that inheritance supports Shared members for relationship-level properties, allowing a shared member to have a different inherited property value than its primary member.

NOTE: property inheritance is only available for EDMCS Custom applications. It is not available for EDMCS applications based on the PBCS, EPBCS, or Financial Cloud GL adapter.

Request Load File Summary Statistics

This nice little feature gives you immediate statistics and feedback as you load a Request File into EDMCS. The number of rows processed, loaded, and skipped are identified, and you can open the request file attached to the in-flight request to see details on why rows were skipped. This all occurs before you submit the request and provides helpful, proactive input as to the changes that will be processed in your request.

Key Features of EDMCS Image 1

Metadata Deletion

With 18.10, you can now delete custom properties and custom applications. This makes me happy, as I’m not ashamed to admit that as I learned EDMCS, I made plenty of mistakes in creating applications or properties that I could not un-do. Oh the shame! Instead, I had to resort to archiving the application to partially hide my guilt. But now you can delete those unwanted custom properties and applications. Hopefully, future releases will allow intelligent deletion of other “mistakes” involving data chain objects like node sets, hierarchy sets, and node types that are archived and no longer used/needed.

Object Details Popup

Another minor, but helpful feature. Hover your mouse over a viewpoint name/label, and a popup window will display the application and dimension being used in the viewpoint. Helpful to keep your bearings when you start to use maintenance views that span multiple EDMCS applications and dimensions.

Key Features of EDMCS Image 2

Conclusion

As you can see, 18.10 is a significant release and arguably the single largest release in terms of new features since EDMCS was introduced in January 2018.

The biggest feature? Definitely the new adapter for Oracle Financials Cloud GL applications. I recommend reviewing the Oracle EDMCS Administration Guide to understand the full power that comes with this adapter. It is quite interesting.

The feature I will use right away? Property inheritance. With my current project involving multiple Custom EDMCS applications, property inheritance is a welcomed feature that will significantly ease the maintenance of key properties throughout my maintenance views.

I’d love to hear any insights and feedback from you as we continue this crazy journey called EDMCS. And stay tuned for future blogs discussing new EDMCS functionality and lessons learned from current projects!

Missed our earlier blogs on EDMCS, Cloud Data Management, and REST API? Be sure to check them out:

Using Subscriptions with EDMCS

As an earlier blog mentioned, the 18.07 release of Enterprise Data Management Cloud Service (EDMCS) delivered one eagerly anticipated piece of functionality: Subscriptions! And do not fear – these subscriptions are useful and do not involve a 1-year subscription to the Fruit of the Month Club (not that there’s anything wrong with that).

This blog post dives deeper into this new functionality, describes how it works, and highlights some lessons learned from utilizing Subscriptions with a current project involving multiple EDMCS Custom applications supporting multiple Profitability and Cost Management Cloud Service (PCMCS) applications.

Why Are Subscriptions Important?

Subscriptions are a huge step towards true “mastering” of enterprise data assets within a single master data Cloud platform. With EDMCS, it is important to build deployment-specific applications configured to the dimensionality requirements of the target applications to most effectively use the packaged adapters, validations, and integration capabilities. But in many cases, you also need to share common hierarchies across applications and avoid duplicative (that’s my big word for today) maintenance. After all, why have a master data management tool if you still must perform maintenance in multiple places? That’s just silly.

The answer to this dilemma is Subscriptions. By implementing Subscriptions, requests submitted to a primary viewpoint will automatically generate parallel subscription requests to subscribing viewpoints to automatically synchronize your hierarchy changes across EDMCS applications.

Note

This comment is important: “automatically generate parallel subscription requests.” EDMCS will not update a target, or subscribing, viewpoint behind the scenes with no visibility or audit trail to what has occurred. A parallel Subscription request will be generated along with the Interactive request that will be visible in the Requests window, along with the full audit trail and details that you find in an Interactive request. Even better, the Subscription request will generate an email and attach a Request File of the changes.

Nerdy Details

Views and Viewpoints

The first thing to really think about is the View and Viewpoint design of your EDMCS applications. Subscriptions are defined at the Viewpoint level, so you need to identify the source and target viewpoint for your business situation. With my current project, I have multiple EDMCS applications supporting multiple PCMCS applications. While the dimensionality is similar across the applications, the hierarchies vary, especially with the alternate hierarchies. So, it has been important to isolate the “common” or shared structures that should be synchronized across applications into their own viewpoint so that a subscription mechanism can be created.

Node Type Converters

You will likely need to create a node type converter. If the source and target viewpoints do not share a common node type, you must create a node type converter for subscriptions to work. In my situation, I had already created node type converters since I wanted to compare common structures across EDMCS applications, so the foundation was there to readily implement subscriptions.

Permissions

To create a Subscription, the creator must have (at a minimum) View Browser permission to the source view, View Owner permission to the target view, and Data Manager permission to the target application.

The Subscription assignee (this is the user who will “submit” the subscription request) must have (at a minimum) View Browser permission to the source view and Data Manager permission to the target application.

Creating a Subscription

Once the foundation is in place in terms of viewpoints, node type converters, and permissions, the actual creation of a subscription is easy.

Inspect the target viewpoint (the viewpoint that is to receive the changes from a source viewpoint via subscription), navigate to the Subscriptions tab, and click Edit. From there you can select the source viewpoint, the request assignee, and enable Auto Submit if needed. Save the subscription and you are all set.

  • Currently, there is no capability to edit an existing subscription. You must delete and add a new subscription to effect a change.
  • Any validation errors for your subscription will appear on this dialog as well. These are documented nicely in the Oracle EDMCS administration guide.

Using Subscriptions with EDMCS Image 1

Auto-Submit and Email Notifications

Emails will be generated and sent to the Request Assignee, whether Auto-Submit is enabled or not. The email will include details such as the original request #, the subscription request #, and how many request items were processed or skipped.

Using Subscriptions with EDMCS Image 2

Note

  • Remember, the subscription request will have a Request File attached to it. View the request file attachment to see details on why specific request items were skipped.
  • The request file is not attached to the email itself, only to the request in EDMCS.

Lessons Learned

Like I mentioned earlier, the foundation is important to making subscriptions work. And it all boils down to design and ensuring the building blocks of that foundation are in place:

Design, Design, Design!

  • The importance of dimension, view, and viewpoint design cannot be overstated. For each dimension, evaluate the primary and alternate hierarchy content and identify what will be shared across dimensions or applications and what will be unique to each dimension and application.
  • Based on that analysis, carefully design your viewpoints to enable subscriptions across EDMCS applications for hierarchies that truly need “mastering.”
  • As early as possible, identify the EDMCS user population along with permission levels for applications and views. This is important to identify the appropriate “Request Assignee” for your Subscriptions. I recommend creating a security matrix identifying each user and the permissions each will have.
  • Without a clear and well thought out design, you will find yourself constantly re-doing your views and viewpoints which, in turn, will cause constant rework of your subscriptions. The “measure twice, cut once” adage certainly applies here!
  • I am a big proponent of standard, consistent naming conventions to improve the usability and end user experience. The same holds true for Subscriptions. Consider using a standard naming convention for your viewpoints so it is clear which viewpoints have a subscription. It’s not obvious – unless you Inspect the viewpoint – that a subscription exists.
    • One approach I’ve been using is to name my source and target viewpoints identically with a special tag or symbol at the end of the target viewpoint name to indicate a subscription is present. I’m sure there are other and probably better ideas, but I find the visual cue to be helpful.
    • Perhaps in the future, Oracle will display subscription details when you hover over a viewpoint name (hint hint).

Node Type Converters

  • Ensure you have node converters in place
  • Make sure your node type converters are mapping all required properties.
    • I ran into an issue where updates to one property in my source viewpoint were not being applied to my target viewpoint via subscription requests, but all other property updates worked fine. The reason? I had recently modified my App Registration and added this property to a dimension’s node type. But my node type converter had already been created and wasn’t mapping or recognizing the new property. Once I updated my node type converter, the problem was solved.

Troubleshooting

  • The request files attached to subscription requests are a valuable troubleshooting tool. Status codes and error messages are included in these Excel files that are extremely helpful to determine why your request was not auto-submitted.
  • Inspect the Subscriptions on your viewpoints. Any validation issues will be displayed and are easily addressed. Typical Subscription validation errors include:
    • The request assignee no longer has the correct permission levels
    • The viewpoint no longer is active
    • A node type converter is missing

Conclusion

I have been looking forward to the subscription functionality in EDMCS and am pleased with it so far. Subscriptions are easy to configure, can be configured to auto-submit if desired, and generate emails to remind the requester a request has occurred and to act if the request was not submitted or request items were skipped. EDMCS Subscriptions are a big step forward to enabling true mastering of your enterprise data management assets!

Catching up with EDMCS

Last time, in the Wonderful World of Enterprise Data Management Cloud Service (EDMCS), we discussed initial impressions of this exciting new Oracle Cloud product and highlighted some early functionality enhancements.

But do you realize how much functionality has been added to EDMCS since its initial release in January 2018? The short list is impressive:

  • Enhanced node alignment/location in side-by-side viewpoint compares
  • Exposed REST API operations including dimension imports/exports and request creation/submission
  • Enhanced searching across members (name and descriptions) and data objects
  • Lifecycle management of data objects
  • Incremental imports
  • Viewpoint download from selected node

Furthermore, in the areas of REST API and metadata integrations, Tony Scalese, Vice President at Edgewater Ranzal and Oracle ACE, has written several blogs. These posts were written from the perspective of hands-on, real-world experience by working with one of three customers accepted into the Oracle EDMCS Early Adopter program. The blog posts include:

In this post, I’d like to highlight another feature that was recently added to EDMCS: enhanced request load files.

Enhanced Request Load Files

In the initial release, EDMCS provides a mechanism to perform bulk updates to EDMCS hierarchies – the Excel request load file. While the feature immediately had some advantages over its distant cousin in Data Relationship Manager (DRM) (action scripts), there were limitations. Primarily, EDMCS would only recognize the first tab or worksheet in an Excel file.

Well that has been fixed! Request load files can now contain multiple worksheets, and EDMCS will recognize all of them (provided the worksheet names match your viewpoint names of course). Additionally, EDMCS will automatically select all valid worksheets to load into EDMCS when loading a request file. This makes it very easy to download viewpoints to Excel and build a request file containing updates for multiple viewpoints to bulk upload at one time.

This also means you need to be careful! Since EDMCS auto-selects any matching worksheet name, if you were not paying attention, you could accidentally load outdated requests from a worksheet. But you can still delete any unwanted request items prior to submitting the request, if you catch them first.

Catching Up on EDMCS 1

While you could always load multiple request files in a single request since the initial release of EDMCS, this feature is a nice usability and productivity enhancement. It works great for situations such as adding a node to a primary hierarchy/viewpoint and inserting it into an alternate hierarchy/viewpoint, all from the same request.

Conclusion (and a teaser)

While EDMCS is the new kid on the block in the Oracle EPM cloud space, it’s exciting to see how it’s quickly closing the gap with new functionality being added regularly! REST API operations, enhanced request files, and the other enhancements mentioned above show how far EDMCS has come in just 6 months.

But wait, there’s more!

The 18.07 release of EDMCS looks to be a HUGE release chock full of new features, including one I am especially excited for: subscriptions!

Look for more blog posts coming soon to discuss the subscription functionality and utilizing EDMCS for a Profitability and Cost Management Cloud Service (PCMCS) implementation.

An Exploration of the EDMCS REST API

Recently my team and I had the opportunity to implement Oracle’s newest offering – Enterprise Data Management Cloud Service (EDMCS). EDMCS for those of you who are not familiar provides a cloud-based solution for managing master data (also referred to as metadata) across the organization.  Some like to refer to EDMCS as Data Relationship Manager (DRM) in the Cloud, but the truth is, EDMCS is not DRM in the Cloud.

EDMCS is a completely new vision of what master data management can and should be. The architect of this new cloud offering is the same person who founded Razza Solutions which was the company that developed the product now known as DRM.  That is important to know because it ensures that the best of what DRM has to offer is brought forward.  But, more importantly, it ensures that the learnings and wish list of capabilities that DRM should have are in the forefront of the developers’ minds.

Ok, now let’s get back to fun stuff. In the 18.05 patch for EDMCS, the REST API (v1) was exposed for public usage.  The documentation for the REST API can be found here:

https://docs.oracle.com/en/cloud/saas/enterprise-data-management-cloud/edmra/rest-endpoints.html

As I highlighted in the previous post Troubleshooting Cloud Data Management Metadata Load Errors, I had developed an automation routine to upload EDMCS extracts to both PBCS and FCCS using FDMEE and Cloud Data Management.  We had been eagerly awaiting the REST API for EDMCS to finalize this automation routine and provide a true end-to-end process that can be scheduled or initialized via a single action.

Let’s take a quick look back at the automation routine developed for this customer. After the metadata has been exported to a flat file from EDMCS, the automation would upload a copy to the PBCS and FCCS pods, launch Cloud Data Management data load rules which would process the EDMCS metadata extracts, run a restructure of the database after all dimensions had been loaded, and then send a status email alerting the administrator of the result.  While elegant, I considered this to be incomplete.

Automation, in my view, is a process that can be executed without user interaction. While an automation routine certainly has parameters that must be generally maintained, once those parameters are set/updated, the automation cycle should not be dependent on user input or action.  In the aforementioned solution, we were beholden to the fact that EDMCS exports had to be run interactively; however, with the introduction of the publicly exposed REST API in the 18.05 EDMCS patch, we are now able to automate the extract of metadata from EDMCS.  That means we can finally complete our fully automated, end-to-end solution for loading metadata.  Let’s review the EDMCS REST API and how we did it.

The REST API for EDMCS is structured similar to other Oracle EPM REST APIs. By this, I mean that multiple REST commands may need to be executed to achieve a functional result.  For example, when executing a Cloud Data Management data load rule via the Data Management REST API, the actual execution of the data load rule is handled by a POST call to the jobs function with the required payload (e.g. DLR name, start period, etc.).  This call is just one portion of a functional requirement.  To achieve an actual data load, a file may need to be uploaded to the cloud, the data load rule initialized, and then the status of the data load rule be retrieved.  To achieve this functional result, three unique REST API executions would need to occur.

To export metadata from EDMCS to a flat file using the REST API, the following needs to be executed:

  1. Get the dimension information for the EDMCS application from which metadata will be exported
  2. Execute an export of the dimension(s)
  3. Determine the status of export
  4. Download the export to a flat file

Let’s explore each of these in a little more detail. First, we need to get the dimension IDs for the application from which we will be downloading metadata.  This is accessed from the applications function.

https://docs.oracle.com/en/cloud/saas/enterprise-data-management-cloud/edmra/op-v1-applications-get.html

When executing this function, the JSON object return includes all applications that exist in EDMCS (including those archived). So the JSON needs to be iterated to find the record that relates to the application from which metadata needs to be exported.  In this case, the name of the application is unique and can be used to locate the appropriate record.  Next, we need to query the JSON object to get the actual dimension id (circled in red).  The dimension ID is used in subsequent calls to actually export the dimension.

Great, now we have the dimension ID. Next, we need to execute the REST API call to export the dimension.

Automated Metadata 1.docx

https://docs.oracle.com/en/cloud/saas/enterprise-data-management-cloud/edmra/op-v1-dimensions-dimensionid-export-download-post.html

You will notice that when you access this POST method, the dimension ID from the previous step is required:

/epm/rest/v1/dimensions/{dimensionId}/export/download

The JSON object returned from this execution contains minimal information. It simply provides the URL to the next required REST API execution which will provide the status of the execution.

Automated Metadata 2.docx

With this information, we can check the status of the export using the jobRuns function

https://docs.oracle.com/en/cloud/saas/enterprise-data-management-cloud/edmra/op-v1-jobruns-jobrunid-get.html

The JSON object returned here provides us the status of the export invoked in the prior step (in yellow) as well as a URL to the actual file to download which is our last step in the process.

Automated Metadata 3.docx

Once the export job is complete, the files can be streamed using the URL provided by the REST execution in the prior step.

https://docs.oracle.com/en/cloud/saas/enterprise-data-management-cloud/edmra/op-v1-files-temp-fileid-get.html

And there you have it, a fully automated solution to download metadata to flat files from EDMCS. Those files are then provided to the existing automation routine and our end-to-end process is truly complete.

And for my next trick…let’s explore some of the different REST API tools that are available to help you in your journey with the EPM REST APIs.

 

Troubleshooting Cloud Data Management Metadata Load Errors

In my last post, I highlighted a solution that was recently deployed for a customer that leveraged Enterprise Data Management Cloud Service (EDMCS), Financial Data Quality Management Enterprise Edition (FDMEE), and Cloud Data Management (CDM) to create an automated metadata integration process for both Planning and Budgeting Cloud Service (PBCS) and Financial Close and Consolidation Cloud Service (FCCS). In this post, I want to take a bit of a deeper dive into the technical build and share some important learnings.

Cloud Data Management introduced the ability to load metadata from a flat file to the Oracle EPM Cloud Services in the 17.11 patch. This functionality provides customers the ability to leverage a common platform for loading both data and metadata within the Cloud.  Equally important, CDM allows metadata to be transformed using its familiar mapping functionality.

As noted, this customer deployed both PBCS and FCCS. Within the PBCS application, four plan types are active while FCCS has the default two plan types.  A design decision was made for EDMCS to create a single custom application type that would store the metadata for both cloud applications.  This decision was not reached without significant thought as well as counsel with Oracle development.  While the pros and cons of the decision are outside the scope of this post, the choice to use a custom application registration in EDMCS ensured that metadata was input a single time but still fed to both cloud applications.  As a result of the EDMCS design decision, a single metadata file (per dimension) was supplied with properties necessary to support each plan type.

CDM leverages its 23 “dimensions” to store metadata information for processing. Exactly like data, metadata is imported using an import format into the CDM relational repository.  Each field from a metadata file is aligned to a CDM dimension field.  The CDM Account dimension always represents the target application member name and the CDM Entity dimension represents the parent of the member.  All other fields can be aligned to any of the remaining 21 dimensions.  CDM Attribute dimensions can be utilized in the import and mapping process but are not available for exporting to the cloud application.  This becomes an important constraint especially in a multi-plan type deployment.  These 21 fields can be used to store any of the properties required to successfully load metadata to the target plan type.

Let’s consider this case study for a moment. The PBCS application has four plan types.  If a process were to be built to load all plan types from a single CDM data load rule, then we would not be able to have more than five plan type specific attributes or properties because we would not have enough CDM fields/dimensions to store the relevant information.  This leads to an important design approach.  Instead of a single CDM data load rule to load all plan types, a data load rule was created for each plan type.  This greatly increased the number of metadata properties and attributes that could be loaded by CDM and ensured that future growth could be accommodated without a redesign of the integration process.

It is important to understand that CDM utilizes the Planning Outline Load Utility (OLU) to actually perform the metadata load to the cloud application. The OLU loads metadata using merge (yes Planning experts, I realize that I am not discovering fire) which is important to understand especially when processing multiple metadata loads for a single application.  When loading metadata, there are certain properties that are Application level.  I like to think of these as being global.  Additionally, there are plan type specific attributes that can align (or not align) to the application level value/setting.  I like to think of these as local.

Why is this important? Well when loading a metadata file, if certain global properties are excluded from the metadata load file, the local properties (if specified) are utilized to default the global properties. Since metadata is loading using merge, this only becomes problematic when a new member is being added to the outline.

In this particular example, an alternate hierarchy with shared members was specified in one of the plan types. The storage property of the alternates was obviously set as Shared; however, when attempting the metadata load, the following error was encountered:

A Base Member cannot be changed to a Shared Member.

After much investigation (details to follow), I discovered that the global property should also be included in the metadata load.

While CDM utilizes the OLU to load metadata, it does not provide as much verbosity in the error information as the PBCS web interface (which also uses OLU) when loading metadata. Below is an example of the error in the CDM process log.  As a tangent, I’d love to check the logs without needing to open a Service Request.  Maybe Oracle will build an enhancement that allows that in the future (hint, hint, wink, wink to my friends at Oracle).

Baha Mar - Error Handling 1

So where do I go from here? Well, what do we know about CDM loading metadata to the cloud application?  We know that CDM uses the OLU to load a flat file generated by CDM.  Bingo!  The metadata file output by CDM is a good starting point.  That file is in the Outbox of the CDM application and can be downloaded in several different ways – CDM Import process (get creative folks), CDM process details, or EPM Automate.  Now we have the metadata file and can test to determine if the error is caused by CDM or the metadata itself.  It’s all about ruling out variables.  So, we take the metadata file and import it manually within the PBCS web interface and are able to replicate the error.  But now we have an important new data point – the line number from the metadata file that is causing the error.

Baha Mar - Error Handling 2

Now that we have actionable information, we can review each property and start isolating and eliminating different variables. We determined that this error was only occurring for new alternate hierarchy parents being added to the outline.  As a test, we added the global storage property and voila, the metadata load completed successfully.  Face palm!

Maybe this would have been obvious to folks with a lot of Planning experience, but there are plenty of folks learning the intricacies of Planning and Essbase (including our friends converting from HFM to FCCS), so I wanted to share a lesson learned in my journey of metadata integration using CDM.

CDM functionality for metadata represents two of the three primary operations of ETL. In my next post, we’ll dive deeper into how the extract component of ETL was accomplished to provide a seamless end- to-end ETL solution for metadata.

Cloud Data Management (CDM) and Financial Data Quality Management Enterprise Edition (FDMEE): A Case Study in Working Together

Why buy Financial Data Quality Management Enterprise Edition (FDMEE) when Cloud Data Management (CDM) is free?  As outlined in my recent white paper – FDMEE vs. Cloud Data Management – there are myriad factors that can drive the decision.  This blog post highlights how one customer gained a highly flexible and automated solution for data and master data management with an on-premise deployment of FDMEE in conjunction with Cloud Data Management.

This customer adopted a pure Cloud strategy as it relates to Enterprise Performance Management (EPM) procuring subscriptions to Planning and Budgeting Cloud Service (PBCS), Financial Close & Consolidation Cloud Service (FCCS), and Account Reconciliation Cloud Service (ARCS).  A diverse business, the customer has many unique operational systems with varying formats and charts of accounts.  So far, no reason why Cloud Data Management (CDM) can’t handle this requirement, right?  This is what CDM does – uses import formats and maps to consume and transform data – right?  Sure, but with caveats.  Notice that I used the word consume and not extract.  CDM does not provide the ability to link with on-premise systems to extract data.  Additionally, flat file data extracts that lack a consistent structure often cannot be natively consumed by CDM.

In this case, data needs to be loaded each day from numerous sources to support daily operational reporting.  The systems are a blend of on-premise, hosted, and Cloud applications.  The customer requirement dictated that any on-premise system should be connected directly to eliminate the need for a flat file extract to be generated daily.  Additionally, the hosted and Cloud applications are very industry specific and, in some cases, provided by very niche vendors.  The ability to modify extract formats was cost prohibitive or simply not supported.  As a result, several of these data feeds were not consumable by CDM without preprocessing/modification.

In light of the above requirements, the customer procured and deployed FDMEE on-premise.  The power of FDMEE allows a solution to be deployed that provides a direct connection to multiple on-premise systems as well as consume the flat file extracts from hosted and Cloud applications including Excel files (not in the required FDMEE/CDM format) and XML.  Because FDMEE on-premise supports scripting, we were able to greatly enrich the data integration cycle with full end-to-end automation including FTP downloading of hosted data, enhancement of the data integration cycle to detect data mapped to members not yet in PBCS or FCCS, dynamically setting substitution variables based on the processing day, running calculations in PBCS, and sending email status alerts to outline the success or failure of a data load cycle.

Although I am a huge FDMEE advocate, I recognize the value of Cloud Data Management and the benefits it provides in a case like this one.  This customer was one of just three participants in the Oracle Enterprise Data Management Cloud Service (EDMCS) program.  This means that they were able to use the software before it was publicly available – otherwise known as GA.  To participate in this program, one must recognize the absence of certain features and functions with the software.  The program allows the customer (and partner) to offer Oracle development and product management valuable input about the software and in some ways drive what features are prioritized within the product roadmap.

EDMCS currently lacks native connections to FCCS, but this will change over time.  So how does CDM help with loading metadata to FCCS?  In a recent update to CDM, Oracle included the ability to import a flat file into CDM and load metadata to a registered target application such as PBCS or FCCS.  John Goodwin gives a detailed overview of the technical setup.

FDMEE and CDM have come together in this case to provide a fully automated data integration process and an automated master data integration process.  Within EDMCS, a Custom application type was created.  The required properties for FCCS were built and attached to the multiple dimensions being mastered, and flat file exports were generated for FCCS.  We knew we were going to use CDM to manage the master data load process, but we had a decision to make – do we leverage EPM Automate or FDMEE as our automation hub?

We chose FDMEE.  Why?  Simply because a lot of automation assets had already been developed in FDMEE that could readily be reused for this process including execution of EPM Automate commands, a framework for leveraging the REST API (for PBCS and FCCS), and email alerting.  Additionally, we found the capabilities of EPM Automate to be somewhat limited.

For example, when you execute a CDM data load rule from EPM Automate, the process ID associated with the execution is not returned.  Why is that important?  Because in the event of a failure, I’d want to download the process log and attach it to the email so the user has information to address the issue.  Could I use the ListFiles command of EPM Automate to get the process log? Possibly, but it doesn’t account for potential concurrency, and I am not doing my job as a consultant if I build a process that can’t handle concurrent operations.  For reasons such as these, we leveraged EPM Automate when possible and the REST API as needed, and we wrapped it all together with an FDMEE process that could be executed on a scheduled basis or on demand simply by using the Script Execution functionality.

Let’s review the end-to-end solution.  In EDMCS, metadata is maintained for PBCS and FCCS.  The metadata is extracted to a flat file (.csv) after maintenance is completed and saved to a network folder.  From FDMEE, the master data integration process is initiated to upload the metadata files to FCCS and PBCS.  Cloud Data Management data load rules are initialized to process the metadata extracts.  In the event of an error, the CDM process log is downloaded.  Finally, an email is generated to alert the administrator of the data integration process status.

There you have it – EDMCS, FDMEE, and CDM working in concert to provide a seamless and elegant solution to data and master data integration for a customer that adopted a Cloud EPM strategy.  If you want to learn how you can enhance your Oracle EPM integration processes, contact us and we’ll be happy to discuss your options.