TLDR, the report is https://nemtracker.github.io/, please note, my experience with BigQuery and Google stack is rather limited, this is just my own perspective as a business user .
Edit : 20 Sept 2019, Data Studio use now BI engine by default for connecting to BigQuery, now the report contains the historical data too.
I built already a dashboard that track AEMO Data using PowerBI, and it is nearly perfect except , the maximum update per day is 8 time, which is quite ok ( direct Query is not an option as it is not supported when you publish to web, actually it is support but rather slow) , but for some reason, I thought how hard would it be to build a dashboard that show always the latest Data.
Edit : 23 Sept 2019, actually now, my go to solution for near real time reporting is Google Data Studio, once you get used to real time time, you can’t go back.
The requirements are
Very minimum cost, it is just a hobby
Near Real time (the data is published every 5
Export to csv
Free to share.
Ideally not too much technical, I don’t want
something to build from scratch.
I got some advices from a friend who works in this kind of scenario and it seems the best option is to build a web app with a database like Postgresql, with a front end in the likes of apache superset or Rstudio Shiny and host it in a cheap VM by digitalocean , which I may eventually do, but I thought let’s give BigQuery a try, the free tier is very generous, 1 TB of free Queries per month is more than enough, and Data Studio is totally free and by default use live connection.
Unlike PowerBI which is a whole self service BI solution in one package, Google offering is split to three separate streams, ETL, the data warehouse (Biguery) and the reporting tool (Data Studio), the pricing is pay per usage
For the ETL, Dataprep would be the natural choice for me,( the service is provided by Trifacta), but to my surprise, apparently you can’t import data from an URL, I think I was a bit unfair to Trifacta, the data has to be in google storage first, which is fine, but the lack of support for zip is hard to understand, at least in the type of business I work for, everyone is using zip
I tried to use Data fusion, but it involve spinning a new spark cluster !!!! , and their price is around 3000 $ per month !!!!!
I think I will stick with Python for the moment.
The first thing you do after creating a new project in BigQuery is to setup cost control.
The minimum I could get for BigQeury is 0.5 TB per day
The source files are located here, very simple csv file, compressed by zip, I care only about three fields
SETTLEMENT DATE : timestamp
DUID : Generator ID , ( power station, solar, wind farm etc)
SCADAVALUE : Electricity produced in Mw
Add a table with partition per day and clustered
by the field DUID
Write a python script that load data to Bigquery,you can have a look at the code used here, hopefully I will blog about it separately
Schedule the script to run every 5 minutes: I am huge fan of azure WebJob, to be honest I tried to use Google function but you can’t write anything in the local folder by default, it seems the container has to be stateless but I just find it easy when I can write temporary data in the local folder (I have a limited understanding of Google function, that was my first impression anyway) , now, I am using google functions and cloud Scheduler, Google functions provide a /tmp that you can write to it, it will use some memory resources.
I added a dimension table that show a full Description for the generator id, region etc, I have the coordinates too, but strangely, Data Studio map visual does not support tiles!!!
Create a view that join the two tables and
remove any duplicate, and filter out the rows where there is no production
(SCADAVALUE =0), if there is no full Description yet for the generator id, use
the id instead
Notice here, although it is a view, the filter per
partition still works, and there is a minimum of 10 MB per table regardless of
the memory scanned, for billing BigQuery used the uncompressed size !!
One very good thing though, the queries results are cached for 1 day, if you do the same query again, it is free!
Create the Data Studio report : I will create two connections :
live connection: pull only today data, every query cost 20 MB, as it is using only one date partition, (2 Tables), the speed is satisfactory, make sure to disactivate the cache
But to confuse everyone there two types of caches, see
the implication is sometimes you get different updated depending if your
selection hit the cache or not, as the editor of the report, it is not an issue,
I can manually click refresh, but for the viewer, to be honest, I am not even
sure how it works, sometimes, when I test it with incognito mode, I get the
latest data sometimes not.
Import connection : it is called extract, it load the data to Data Studio in-memory database (it uses BI engine created by one of the original authors of multidimensional) , just be careful as the maximum that can be imported is 100 MB (non compressed), which is rather very small (ok it is free so I can’t complain really), once I was very confused why the data did not match, it turn out Data Studio truncate the import without warning, anyway to optimise this 100 MB, I extract a summary of the data and removed the time dimension and filtered only to the last 14 days, and I schedule the extract to run every day at 12:30 AM, notice today data is not included.
Note : Because both datasets use the same data source,
cross filtering works by default, if using two different sources (let’s say,
csv and google search, you need some awkward workaround to make it works)
Voila the live report, 😊 a nice feature shown here (sorry for the gif quality) is the export to Sheet
Schedule email delivery
although the report
is very simple, I must admit, I find it very satisfying, there is some little pleasure
in watching real time data, some missing features, I would love to have
An option to disactivate all the caches or bring back the option to let the viewer manually refresh the report.
An option to trigger email delivery based on alert, (for example when a measure reaches a maximum value), or at least schedule email delivery multiple time per day.
Make data Studio web site mobile friendly, it is hard to select the report from the list of available reports.
Google Data Studio support for maps is nearly non existent, that’s a showstopper for a lot of business scenarios
Kelpler.gl is a Geospatial data analysis
by Uber !!!, in this short blog we show
how to build a map with a time attribute and then showing a time playback slider,
it is very and straightforward process, all you need is to format the data as a
This data was randomized, as an example I show the main step to build a solar farm, piling, tracker, module then Electrical works, but obviously you can try it with road , pipeline projects, any project really that have geographic data
Please note the data is rendered using your local computer, so it is totally safe
The map is here,
keep reading if you want to understand how it was built
I attached a sample dataset here;
the field are self-explanatory.
Just be careful with Excel for some reason, when you open
the csv file, it changes the date format, make sure the format is like this
And because in a medium sized solar farm, the total pile
will be around 50,000 Piles, if you want just to show 4 main stages, the total
number of rows will be 50,000 * 4 = 200,000 row, PowerBI will not support that
( Tableau shine in this scenario),
The documentation is very clear, please read it to experiment with your own data
If your dataset has this fields names
X_lat, X_lng, Kepler.gl will automatically recognised it as a point layer
2-Define the legend
We select colour based on the field status ( select the palette
3- Select time field
The filed should be recognized as time otherwise, it will
4- Add satellite as new map Style
For some reason, it is not there by default, but it is
trivial to add
You can play with the slider, either moving the end date
or a window, up to you
6-Now you have multiple options for sharing
You need to have a mapbox
account, it is virtual free to medium usage !!!
For this example, I will share it using Dropbox, but you can export to HTML and securely share it either per email or hosted in sharepoint
This is one of the best piece of software
that I had the privilege to use, nothing to install, free , simple and straight
to the point !!
I have been asked to produce a simple construction report, we need to show the last 4 weeks of actual progress data and 6 weeks of forecast and to make thing a little bit complex the average installation since the start of the project, nothing special three measures, average to date, install per week and forecast per week
Obviously, it is trivial to be done in Excel using named sets,
if you don’t know what’s named set and cube formula is, you are missing the
most powerful reporting paradigm in Excel, a good introduction is here,
and there are plenty of resources here.
Unfortunately named set is not supported yet in PowerBI, you
can vote here,
Just for demonstration purpose, if you try to add those three
measures to a matrix visual, PowerBI just repeat them for every time period, obviously
that’s not good at all, the actual
installation make sense only in the past and the forecast has to be in the future,
there is no option to hide a measure if there is no value in a column and even
if it was possible we need to show the average installation independently of
the time period, anyway this the report when you add the three measures
and because I already learned a new trick on how to dynamically add measures to a matrix visual in PowerBI, I was tempted to try and see if it works in this scenario.
So, let’s see how it can be done using the disconnected table
Create a disconnected table with two columns
Order and status
Add a calculated column,
As the cut-off date change at least three times a week, the week number change accordingly, we can’t simply hard code the dates, instead let’s add a new calculated column, which will just lookup the week date from a master calendar table based on the order, when the order is -4 it will return “average to date”, I added a dummy 0.5 order just to add an empty space between actual and forecast ( cosmetic is important)
Who doesn’t love the glossy Level 1 reports our project produce. But really, when you look into these beauties, really understand the difficulty that goes into them. What follows is first a description of what a typical Level 1 report is, and how we can structure our excel based data to be a bit smarter.
This is by no means a fully comprehensive guide on this subject. It is instead just a primer to get us thinking about how we feed data into our reports.
Who doesn’t love the glossy Level 1 reports our mega construction projects produce. But really, when you look into these beauties, do you really understand the difficulty that goes into them. What follows is first a description of what a typical Level 1 report is, and how we can structure our excel based data to be a bit smarter (which is the real message to this article).
Interspersed with hopefully be a few key strategy points which can guide your work.
I’ll then showcase how you can take what will now be structured data and upload into a powerBI visual (although the process to capture the data into any database and drive any visualization tool would be the same)
Strategy – Don’t be afraid to use excel (not everything needs to be automated)
Key Elements of a Level 1 Report
Cost and Progress
Here we are presented with:
Overall progress curve
Cost & Commitment curves
Some may argue what to lead with – for me its always %. No bigger value highlights where your are more than what % are we. Not displayed on the image above is a data series reflecting how many people are have and comparison against planned. People achieve progress. Its impossible to talk progress without talking how many people we have. The graphs provide enough enough context to allow for discussions about productivity without having to muddy the waters
The cost sections should include visibility into what our final forecast costs will be (and comparison against baseline). Underneath that key metric are a few sub items such as how much contingency we have, a few cost curves associated with spend profiles and commitment profiles.
Schedule and Narrative
The schedule aspects of a Level 1 report are always tricky. Do we need to only display the final project milestone? For me, on major projects no single DATE has any meaning. Thus even on a Level 1, I still prefer to include 10-15 dates that represent some key aspect of the project. All dates should be compared against what we said last month to highlight current month variances, and dates should be compared against our project baseline (or whatever current approved version thereof).
The narrative section of a Level 1 can nearly always be updated by simply reading the progress, cost and schedule tables. Just put words to the graphs. Key adders here are insights into RISKs. What may come in the future that will alter what we are saying today!
As always, safety metrics are also usually front and center. For me, this has always been a difficult aspect of our jobs. A political correctness that is forced into our reporting. Don’t get me wrong, safety is the most important aspect of a project. So, including a safety table somewhere on the Level 1 is always done. For this article, I want to instead focus on the key project control elements and data integration.
Level 1 Data Structures
So, we all know what a Level 1 report looks like, and I would fathom we can all mostly agree these are the elements included and can be rolled out as a standard for any major construction contractor. Most of our reports likely already report this information in some manner or another. The entire point of this article is that we should really focus on entering the data in a smart data centric way so that if you want to automate anything down the line, you have the foundations to do so.
At this stage, I don’t want to talk about the source data used to generate your summaries. We can leave that for a later discussion.
Key Data Domains
We are aiming towards consistency here and want to actually represent all the data required for our key Level 1 chart to be housed in a database. Therefore we need to have structure.
Strategy – Do not focus on systems, focus on DATA
A critical strategic element in my approach is that I do not care what systems you use. Our reporting is not a function of our systems (at least in this step 1 phase). We instead need a structure from which we can extract data and as easily as possible, move that data into a template or format in which we can drive our level 1.
If you go down the path to seamlessly integrate source systems with a Level 1, you unwittingly constrain yourself.
Typically our (time phased) progress data will be sourced from Primavera. There are other systems where the progress data may live, but again, that isn’t the focus of this article – I don’t care where it lives and neither will any seasoned project controls manager. We just need to know it exists and has a common structure
Here, a few key notes, use a consistent data format. The above structure is how all your progress data should be housed, not just Level 1. All time phased data, all the way down to Level 5 detail items should be managed in a data structure, not a fancy formatted excel file. Trust me, updating a table such as the above will serve you in the long run. Even if your data is fully managed inside a system such as P6 or PRISM or ECOSYS or COBRA, you should be able to at least extract Level 1 into the format defined above.
You guessed, we can capture our Level 1 cost data in exactly the same format
In the graphs we are building, there are only 11 specific datasets. Only 4 of these require update on a period basis. So again, we really boil this down to something simple.
Strategy – Do not over complicate anything in your Level 1 layer
The implementation of the specific data model I have outlined above fits the strategic approach to keep your level 1 simple. Any project can implement this data model for Level 1 with without any integration into source systems. Level 1 can be updated by the project controls team doing a few copy-pastes into excel to capture project wide data. Again, I would assume your teams already do this, but perhaps end up copying this data into various other corporate systems as well.
Again, we are keeping a simple approach and only capture the required information.
Here, we are forced into a different structure. So whereas the cost and progress data can fit the same data model (as seen above), we will need a different template for schedule dates. We will typically be using Primavera, as such this model fits P6, but the idea is universal.
I do not believe this information can ever be fully automated from our scheduling systems. These paths will continually be adjusted. The planning lead will always refine what activities are being tracked to be displayed on the Level 1. Behind the scenes, there are tricks upon tricks to pull the dates, however, again, we are talking about the data layer here, not necessarily HOW you get the data into this format.
It is entirely possible to have the assignments encoded into P6 activity codes. Therefore, it would be possible to integrate your Level 1 data directly into either the source P6 database, or an XER export. In my experience, any automation that is attempted in this arena (for Level 1 data), is futile. We are only talking 10-15 key activities. Let you lead planner sort out how they get the data into this format. Again, our strategy is to not over complicate this. If the data is provided to a digital team in the format about, you are for all intents done.
The model above only captures the finish dates. If added visuals with simplified GANTT charts are needed in your Level 1 (and will be discussed in my next Level 2 article), you would have to edit the above.
The nice value of the above structure is that we have effectively created an interface, an integration layer, between what will be P6 data and our dashboard. The list of what activities can easily be edited by way of a sharepoint list. Then, in your data model, you can link on scheduleID to pull the relevent date data (I suspect many do this).
Too often, narrative comments are shuffled between parties via email, entered into several documents, edited, customized, etc. The project controls team is always struggling sourcing commentary from various sources, and in my experience, we end up entering in something ourselves.
Level 1 data structures have to fit into these complications. In this realm, sharepoint offers a canned solution by way of sharepoint lists.
Strategy – If Technology already exists, use it
Strategy – Technology can be used in innovative ways – use a mashup mindset to use existing technology in a new way
I find that sharepoint lists offer unparalleled capabilities for commentary. However, for lists to be really functional, they need to be embedded into FORMS or some routines that provide export functionality
In this example, I have mocked up a simple INFOPATH form that could represent our sharepoint fields. The sky is the limit when it comes to existing technology that can automate the capture of this type of commentary.
The value adder here is that instead of allowing unstructured comments (via email or manually marking up a word , excel or power point file), we have structured comments that are housed in a database and that database can be updated in a distributed manner using WEB based technologies.
The above would be a web based form which will be updated by the associated responsible parties. However, we can’t quite import a form into our data model. When the above form is filled out, the data will be stored in a data model (which we will have to design first before we can even build the form above). Thus, what we are looking for is something akin to the below
The above is just a table in an excel file, but again, when we house data in this format, it can naturally flow into a database. That is what we need to focus on. Even in our excel reporting world, if you can capture commentary in this tabular data centric way, you can still link to it from your main dashboard tabs to be “smarter” in how information is managed.
Strategy – Focus on the DATA! (I can’t say this enough)
Everything we do can be captured in a data model. Every report we design should be able to pull direct data out of a data structure. Thus, before we add anything to reports, first consider the entire flow of data required.
Putting it all together
At no point in time in the above have I had to rely on a source system. However, I have been able to take a typical Level 1 report and extract everything from it. I have taken this data and outlined a data model (in simple form) that can drive not just 1 project, but an entire corporate endeavor in this space.
As with everything, nothing novel here. Many companies already have systems that capture some of this information. This is more just a thought experiment for those that perhaps do not have a clear data model that supports level 1 reporting. It also highlights the discussion topics of “what are the manual steps” – because there will be manual steps in getting the data into the right format.
For me, everything above has to be manual at some point up or down the food chain. Your projects and portfolios need to have the discussions about where this type of Level 1 data is housed. If all projects already have this data in consistent databases, all you need to do is query that source. Everything discussed here is system independent. You can easily generate these data tables by way of query a source system directly (if you can), but I have not limited or require that approach
Strategy – Whatever you do, allow for flexibility
Even though my data model is entirely excel based, the data structure is very powerful. I can, in quite automated steps, import and convert these datesets into a more database model and thus gain value from dashboards that wouldn’t be custom for your project, but could drive an entire portfolio (and when you see how this scales to Level 2 data and beyond, the worlds your oyster).
If you actually want to proceed with a dashboard, and if you have your data as outlined above, here is what you can do with it. In fact, I would recommend that your source tab in excel that is driving your dashboard looks like the below.
The above data isn’t “immediately” friendly for digital reporting. A few transformations are required. The key steps involved are (the below was done as just an example using PowerQuery)
Unpivot the Timephase date columns
Pivot the the “SeriesName” column to create a unique “Column” for each dataset (this is need to create unique lines on our dashboard graphs)
At this stage, we have a nicely formatted table and we can now import into PowerBI. The intent here is not to showcase a beautiful Level 1 dashboard in PowerBI. My intent is more to showcase the data structures need to drive a dashboard. With the above data, we get pull each data series into graphs, tables, cards, KPI metrics, etc.
Our model has tagged each record with a “As-Of” date. Thus you can utilize this structure to have your dashboard display ALL prior months by way of a slider or select. Given more advanced skills, you can also pull out metrics about current incremental values vs what we said last month. Although, I feel those metrics are best served in Level 2 report where more detail is available.
Apologies for the look and feel below, I just pulled in the data to showcase that indeed you can drive a dashboard with what is effectively just a few lines of data that every project already has. We can bring together cost, schedule, progress, and commentary quite easily and in a very data friendly way.
For me, there is no substitute for an excel based dashboard. The value in this for me is ensuring that when I produce a Level 1 Dashboard (in Excel), I should give consideration to ensuring my data is structured appropriately. This gives us a fighting change to perhaps go down the path of creating a more digital world. It also allows for perhaps more flexibility in dealing with Level 2 data to maybe have some real automation of rolling up of data.
Level 2 obviously. I hope to showcase how the same ideas and concepts here can also help you structure your raw excel based Level 2 data to perhaps be better utilized in a more digital world