Wednesday, 12 August 2015

Adblocking could be the saviour of high quality journalism

Reading the latest news, adblocking is heralded as the harbinger of the online apocalypse.

Many internet users have decided that they don't like to be tracked and profiled.

They don't like to be irritatingly diverted from the article that they're trying to read.

They don't like auto-play videos.

They don't like ads for things they've already bought, following them around the internet.

And they especially don't like their mobile data allowance being chewed up by advertising.

All of this means that the user base of adblocking software is on an upward trajectory and publishers are starting to worry.

According to PageFair, 21% of UK internet users use an adblocker. Amongst younger, tech-savvy audiences, the rate of blocking is much higher.

For publishers, this is a serious problem, because they get paid per thousand eyeballs viewing ads on their webpages.

For example, if I visit The Guardian, I see this

That banner at the top is generating a small amount of money per visitor for The Guardian.

With a blocker? Poof! It's gone. The page resizes into the gap and you'd never even know it had been there. The Guardian gets no money at all for my visit.

As an aside, Ghostery gives me a list of every tracker that's logging my visit to The Guardian and many of these will subsequently try to follow me around the web and see where else I go.

It's quite a big list...

That's FIFTY EIGHT separate trackers, just from a single visit to The Guardian homepage. No wonder some users want to opt out.

So adblocking continues on its upward trajectory, gains increasing penetration into mobile devices, and in the near future hits and then passes 50% of all users. All of our national newspapers go bust. It's inevitable, right?

Well no, actually, I don't think so.

I think adblocking could be very good for high quality journalism, but there'll be some short-term pain before we get there.

First a little bit of technical info...

Adverts are fairly easy for users to block, due to the way that they're bought and sold and embedded onto web pages. When you visit a newspaper's website, it loads the article you want to read from its own server and makes a request to some other server(s) for adverts. If you keep a big list of the servers that are used to deliver advertising, you can block them in your browser. The article loads and the adverts don't. Simple.

Now imagine the adblocking rate is 100% of internet users. What happens?

Any user, visiting any website, sees no third party adverts at all and publishers get no money.

But advertisers still want to talk to the The Guardian's readers. Of course they do, The Guardian's readers spend a long time staring at those pages and they're an educated bunch, with disposable income. Advertisers like people with disposable incomes and will work quite hard to get their brands in front of them.

Advertisers also no longer have an option to place their ads programmatically - splattering them across the web on loads of different sites and trying to follow individual users around - because as soon as they try to do that, the adblockers kick in and their messages disappear.

Suddenly, there are big online advertising budgets available, but no way to spend them programatically across lots of smaller websites.

So newspapers start selling advertising space the old fashioned way. They agree to place adverts on their site - delivered from their own servers, woven into articles and much tougher to block - and they agree a price to do it.

They might even negotiate that deal on the telephone. How quaint.

It's a model that looks a lot more like the traditional way of selling advertising space. Individually struck, higher value deals with large companies, rather than automated small-ads scattered across the web.

Why doesn't that already happen? Well sometimes it does, but currently, large proportions of online advertising budgets are used to buy programmatically. If you've got a choice between negotiations with individual newspapers, or bunging some cash at an ad exchange on the promise that the exchange will find your exact target audience wherever they are on the web and show them your ad, then of course you'll pick the automated route.

Adblocking takes away the automated route. Users don't like the tracking it entails and in increasing numbers, they're opting themselves out.

If you can't track and follow an individual user onto sites where its cheap to advertise, then you're going to have to buy your audiences in bulk, in the places where they congregate.

Adblocking doesn't leave an internet barren of advertising. It leaves large websites, with high quality audiences, in an incredibly strong position to negotiate their own deals.

For newspapers, it removes a huge number smaller websites from their competitive set and puts them back in the much stronger position that they used to enjoy.

Adblocking at scale would redirect online budgets towards those sites that can guarantee large, desirable audiences for advertisers. It could be the saviour of high quality journalism.

Thursday, 6 August 2015

Is Microsoft's Power BI a Tableau killer?

Have you heard of Betteridge's law of headlines? Then you already know the answer to this post's headline. No it's not.

Why not? Good question. And there is one feature where Microsoft's Power BI  manages to lay a glove on Tableau, but we'll save that for later. First, why are we here?

My last post on Wallpapering Fog was part reminiscence, part rant and part lament about where it's all going wrong with Microsoft Excel. You see, I used to like Excel a lot and mostly, I really don't any more. It's still very useful, but in terms of new features, the bad stuff is starting to outweigh the good.

I had a comment on that post from the design leader on Power BI - which is very flattering - asking if I'd tried the general availability version of their new software and I hadn't, yet. I have now. My IT department is also pushing Power BI so I really needed to take a proper look.

A bit of background if you're unfamiliar with Wallpapering Fog - I've been using Tableau (full-fat and Public) for four years and love it, but wouldn't say I'm wedded to it. I'll be very comfortable jumping ship if something better comes along, but for now Tableau is the best BI software on the market by quite a distance. I think using the best and looking for better is a healthy attitude to take.

Before Tableau, I used Excel Services. It was rubbish. Hopefully Power BI is better.

Onto the test. Is Power BI any good?

I went to and signed up.

I got a button that said "Get Started" and clicked it.

The website paused and said "working on it".

Hmmm. I haven't asked you to do anything yet. What could you possibly be working on? As a (reluctant) SharePoint user, I know that "working on it" message all too well. This is not the most promising start.

Still, it's gone in a couple of seconds and I've promised to be as objective as possible with this review. We're in, let's move on.

I'm deliberately going to pile into Power BI without reading any instructions at all, because that's exactly the approach I took with two of its competitors - Qlik View and Tableau - and with both, I was able to make good things happen pretty quickly. I only needed help later, as more advanced features came into play. That's the benchmark.

This will also be an early review and I haven't tested Power BI extensively. I may have missed things, but seeing what you can achieve with a piece of software in 24 hours is a useful exercise. In my experience, great software - like Photoshop or Tableau - will blow you away immediately and then keep on giving as you discover more depth. Before I'd invested any time in Tableau and long before I'd got involved with its community, you can read that it did that.

What to do first? I need to connect to some data so we can put Power BI through its paces. Let's have a look at the options.

I can load a csv file or spreadsheet, but decided to have a crack at an API connector and was presented with a slightly strange array of options.

Acumatica (who?) and Circuit ID (also who?) but not the usual array of social, stock price and economic data connectors. Odd.

Anyway, Google Analytics is there. Everything connects to Google Analytics. Let's try that.

Oops. Five minutes of Power BI thinking about things and then I gave up and refreshed my browser.

Maybe I'm being too ambitious (I'm not, I managed to do this in the competition's software). I loaded the example retail dataset.

A dashboard appeared!

If I'm being picky, some of the formatting is a bit scrappy - especially considering it's the only bundled example - but it's a dashboard. Let's not be too picky yet.

I can click around, highlight and filter and go into edit mode, but after a couple of minutes, I found a button for "Power BI for desktop" and downloaded it. I had hopes of more connectors for local databases and did find those. After I eventually installed it.

Yeah, so like I said. Eventually installed. Including rebooting a work PC that doesn't hurry itself to reboot, that little lot took the best part of twenty minutes.

Why don't I already have the latest version of IE? Because it's rubbish. I use Chrome and Firefox.

On the plus side, from the desktop, the Google Analytics connector works.

From this point on, the niggles stopped and Power BI was... Fine. Not good, not great. Fine.

I loaded some traffic data from Google Analytics and drew a couple of charts. I noticed that dates don't automatically drill from years, through months, to days like they do in Tableau and so manually created a month field.

Power BI's got a weird distinction between measures and columns, so creating that date column involved a brief false start. Apparently dates can only be a column, not a measure. What's the difference? Tableau switches effortlessly between dimensions and metrics, which is a very useful feature. You can split data by a numeric field (using it as a dimension), or you can add that numeric field up. At the point you create the field, the distinction doesn't matter - it's just a column of data and you choose what to do with it afterwards.

Creating charts was straightforward enough, but quite limited. Microsoft have ditched the rows and columns interface from Pivot Tables (that Tableau adopted and supercharged) and gone with a vertical interface. It works, but it's nowhere near as intuitive.

That list of options for summing, averaging and counting is again... fine. Just the basics. No frills.

Enough Google Analytics. I restarted Power BI and connected to one of our advertising datasets on SQL Server. 350k rows that describe what different companies have been spending on advertising over the past few years, to see how Power BI would handle a little bit more data.

It coped fine. Connecting was slightly counterintuitive because you have to tick a little box next to the table name, that's not obviously a tick box. Just clicking on the table name brings up a preview, but leaves the "Load" button greyed out and leaves you scratching your head for a bit.

Data loaded, I quickly put together this summary view of UK advertising spend.

You can filter across the whole page, or just on one chart. If you click on something - a date, or a category - then every element on the dashboard immediately filters to what you've selected.

I like that sort of universal filtering behaviour, when I can control it, but it doesn't look like you can here. Everything filters when you click something, whether you like it or not. That has the potential to confuse the hell out of non-technical end users of your dashboard as they idly highlight a date and most of the data on the view suddenly vanishes.

In keeping with the imposed filtering, you can drop a filter element onto the view (top right on my little dashboard) to make it more obvious what's going on, but you can't attach that filter to specific dashboard elements. You filter everything, or nothing.

If you look closely at my dashboard, you'll see that the dates (which are slanted, yuk) are in the wrong order. That's because in our SQL Server, they're stored as strings.

I tried to make a new column that would be a correctly formatted date and this happened.

What's wrong with that? I don't know, it looks fine to me. Apparently Power BI doesn't know either, because the error message is blank.

I tried calling it Formatted_Date, removing the space. Nope.

I tried a few other ideas. Nope.

Maybe DATEVALUE doesn't work like in Excel. That would be daft.

I gave up.

Overall, Power BI feels like it's doing the bare minimum. You can drop charts, text and maps onto the screen. You can sum and average data. You can format axes and change colours. That's about it.

There are features to create variables (when they work) and to connect data sources together, but BI software will stand or fall on the front end and how you are able to present data back to a user.

Drawing on a couple of my own recent projects, this is Tableau. So's this. Both built in the free, Public version. As far as I can see, you've got no chance of producing visualisations like these in Power BI. It's not outright bad, it works, it's just miles behind the competition.

Power BI hasn't come out swinging. It's a cagey, cautious entry into data visualisation that seems competent, but nothing more than that.

Before we go on, I should mention price, because it's important. Full fat Tableau Desktop is over £1000 a copy. It bloody well ought to be good.

Full fat Power BI is $9.99 a month. These two pieces of software aren't targeting the same market.

It's fairer to compare Power BI with Tableau's free offering - Tableau Public.

Tableau Public is full-fat Tableau Desktop, with database connections stripped out and you can only save your visualisations to Tableau's cloud. We've established that in terms of the sophistication of visualisations you can build, this means Tableau will blow Power BI out of the water. What about when you publish your dashboard to the cloud, for others to see?

Data access and refresh is where Power BI wins over Tableau. It can connect to database sources and APIs and auto-refresh from those sources so that your online dashboard stays updated without you needing to do anything. This functionality is free and for $9.99 a month you can do those refreshes at high frequency, with a bigger data storage allowance, though one that only takes it up to the allowance that Tableau offers for free.

OK Microsoft, you just got my attention. It might be just a line chart and a map, but an auto-refreshing line chart and a map is interesting.

What's the catch?

Well if I had an auto-refreshing dashboard, I'd want to embed it into this blog, or into, but you can't do that without paying $5 a month for SharePoint Online. I use SharePoint at work and it might just be the worst designed piece of software I've ever come across. I'm not even sure it was designed. No, I'm not paying my own money for it.

For now, if you want to build a sophisticated, good looking visualisation, you need Tableau.

If you want to build a more basic visualisation that refreshes itself, then Power BI is an option, but you'll only be able to share it on, not embed it.

There are hopes within the Tableau community that Public will get Tableau's new web data connector, which is coming in the next mini-release. If that happens, then Power BI is dead in the water, because its auto-refreshing (at a low price) USP will be gone.

In a business context, Power BI is in a potentially strong position. If your company has already bought into Microsoft's Office 365 and SharePoint stack - as mine has - then Power BI will integrate with that fairly cheaply and allow users to publish visualisations to each other. I wouldn't be at all surprised to see it gain a fair amount of traction.

Unfortunately, as analysts, that will mean investing our time making the argument that Power BI's competitors, while significantly more expensive, are significantly better. Tableau and Qlik View are changes to how analysts work and massive boosts to their capability and efficiency. Power BI is not that. If you just want to automatically publish a table and chart of financial results to the exec team though, it will certainly be useful.

Overall, Power BI is a 6/10 product and in places it doesn't feel finished. It's useable, but limited. Give it a try, but don't expect too much.


I haven't mentioned Power BI's other innovation of natural language search. The idea is that users can type in "sales in France" (or similar) and the dashboard will show them that.

I haven't mentioned it because it's a gimmick. Cortana (I assume it's got Cortana's engine) isn't the Starship Enterprise computer and it's not going to be intelligent enough to be useful. You'll see this message a lot.

Unfortunately, I already have experience of this feature playing really well in a controlled demo, where the salesman knows it will work. It's going to make it harder to explain to an excited senior exec that Power BI isn't really very good, when "you can talk to it and it just understands!"

Footnote 2:

In this review, I may have said Power BI can't do something, which it actually can. If that functionality comes from bolting on Power-something-else, then I'm not interested. Power BI, PowerView, PowerPivot... the Microsoft data ecosystem is a bit of a mess at the moment and it needs a more coherent offering. Power BI should work out of the box as a single download, as its competitors do and that is how I've tested it.