The problem with personalization

The challenge of personalizing the news is the heterogeneity of our personal interests and the weakness of the signal we expose to the recommendation engine.

As an example, below are four stories from around the country I was interested in today, for reasons ranging from ‘possibly obvious’ to ‘unknowable’ by an algorithm. And by ‘unknowable’ I mean how likely would a machine be to affirmatively pick any of these four at a very low signal to noise ratio.

The dot-Boston domain is now open
Why am I interested? I was working at the Globe when we originally bid on and won the rights to sell this TLD. (I proposed renaming Boston.com to com.Boston. Just because.)
Could a machine have affirmatively predicted my interest? No.

Micro-apartments proposed for former mill building on Saco Island
Why am I interested? We have relatives that live in the complex and I spent some time working in the area. Could a machine have affirmatively predicted my interest? Not likely.

NBC moves 130 Premier League games to streaming service
Why am I interested? I am a big fan of the Premier League and watch games on NBC – but found this by random chance at our paper in Sacramento. Could a machine have affirmatively predicted my interest? Yes, I probably leave a wide paper trail on this topic.

Heavy traffic, cellphone service disruption expected in Charleston for total solar eclipse
Why am I interested? We will be in town for the eclipse. Could a machine have affirmatively predicted my interest? If the algorithm knew my calendar, correlated the travel with the geographically-specific eclipse event and put the two together – yes. In the current reality, no.

So what’s the ‘problem’ here? Personalization depends mostly on observed web behaviors. Much of our interest in the news is based on real life experiences and events. So to provide me a list of recommended stories you need to know not what I clicked on last week, but where I lived in 1998 and my level of interest in urban planning issues.

So even though Facebook knows ostensibly everything about me, and Twitter is packed with people I know/trust and rely on for news – neither of those platforms or any other app I am aware of is going to identify those four stories on a given day and surface them in a unified newsfeed.

In fact I will will be suitably impressed if a machine will ever be able to perform at that level of serendipity.  But, if you invent it, I would pay for that convenience as a service.

Product Management and proverbs

I can’t remember the first (or last time) I heard this proverb but I am convinced it is an allegory for product management and technical debt.

For want of a nail, the shoe was lost;
For want of the shoe, the horse was lost;
For want of the horse, the rider was lost;
For want of the rider, the battle was lost;
For want of the battle, the kingdom was lost;
And all for the want of a horseshoe nail

Newspapers are still steam powered

Yesterday, the BBC podcast “50 Things That Made The Modern Economy” covered the invention of the electric dynamo. The episode was partly based on a 1990 academic paper by economist Paul A. David who studies “scientific progress and technical change.

His thesis is that the initial lack of productivity gains related to the introduction of the electricity in factories was the common practice of simply replacing the one centralized steam engine with an electric motor. That approach maintained the disadvantages of steam while bringing none of the value of electricity.

The paper is here but read the excerpt below and mentally find and replace steam/electric with print/digital.

The proximate source of delay in the exploitation of the productivity improvement potential incipient in the dynamo revolution was, in large part, the slow pace of factory electrification.

The latter, in turn, was attributable to the unprofitability of replacing still serviceable manufacturing plants embodying production technologies adapted to the old regime of mechanical power derived from water and steam.

Thus it was the American industries that were enjoying the most rapid expansion in the early twentieth century (tobacco, fabricated metals, transportation equipment and electrical machinery itself) that afforded greatest immediate scope for the construction of new, electrified plants…

 

 

Shining a light in Plato’s Product Office

One of my favorite things about ONA is the session pitch. It is a great chance to wonder, “do I actually have anything to say?” but more importantly “what do I want to learn more about?”

The suggestion box opened recently and I am still thinking about this post from October on the challenges of building great products at legacy news organizations.

Here’s the pitch:

Shining a light in Plato’s Product Office

We can all agree – creating great products requires an understanding of and empathy for our users. That is a core tenet of Design Thinking and all Human Centered Design approaches to product development.

But empathy for users (readers, viewers, customers?) is not enough. The products we develop are an emergent property of the people, processes and cultures of our teams. And creating great products in any organization, especially one with a hundred-plus years of history, requires understanding ourselves as much as it does users.

We will talk about Plato’s Cave, empathy, fundamental attribution error, and the unique challenges of shepherding ideas through the complex web of intra-organizational boundaries, stakeholders, silos and workflows that are typical in modern media companies.

The Product Solution Stack Test

(tldr; A product strategy is only as strong as its weakest link.)

The Twitter dust-up this week ostensibly about digital having been a mistake for newspapers felt a bit like Groundhog Day (the movie.) New day, same argument, same result.

Steve Buttry and Mathew Ingram both posted eloquent rebuttals:

Buttry:  The newspaper industry’s colossal mistake was a defensive digital strategy

The colossal mistake that the newspaper industry made was responding to digital challenges and opportunities with defensive measures intended to protect newspapers, and timid experiments with posting print-first content online, rather than truly exploring and pursuing digital possibilities.

Ingram: Sticking With Print Would Not Have Helped Newspapers Avoid Death

Shafer’s piece assumes there was some magical Eden in which newspapers were unassailable, and that it would have been possible to remain there if not for the sin of pushing too hard into digital. But that simply isn’t the case.

Newspapers were never likely to dominate digital in the way they had print. Many of the causes are secular – the Internet is a supremely disruptive force that inherently flattens structures, eliminates middlemen and creates value at scale.

Newspapers on the other hand were middlemen built to extract local value from news and advertising. The worth of local journalism aside, the future of the print delivery business was always going to be problematic.

But blaming our travails entirely on those outside forces risks ignoring the mistakes we do own. And at the product and development level, we failed at some basic digital blocking and tackling. Those failures can be attributed to a lack of systems thinking in our strategies, and a lack of respect for just how difficult the ‘system’ of a legacy media company can be to navigate.

Developers have a concept of the ‘tech stack’ to describe the collection of software and subsystems required for a project. A typical example is ‘LAMP’ representing Linux, Apache, MySQL and PHP in a common combination of operating system, web server, database and application/template software for a web site. Each component is dependent on the other and works together as a single system.

In product development, we also attempt to design coherent systems. But unlike a standard software stack, product solutions are emergent properties that reflect the desires, needs and constraints of a wide range of constituencies including humans, machines, cost and time.

  1. Reader Need (desirability)
  2. Business Need (viability)
  3. Visual Design (on brand and articulate)
  4. User Experience (function, usability)
  5. Development Effort (feasibility)
  6. Systems Integration (between disparate services and tools)
  7. Workflow (process for creating & publishing)
  8. Culture (shared assumptions that govern behavior)

Note that some of those layers (desirability, feasibility and viability) are typical human-centered design concepts. The others represent the people and processes necessary to actually bring a product to life including designers, developers and journalists.  To succeed, any product must ‘pass the test’ of at least minimally satisfying the requirements for each of the eight layers in that stack. And, because the stack crosses multiple organizational boundaries, creative and strategic conflicts will typically abound.

And more to the point, each layer in the stack is entirely capable of vetoing a product decision considered ‘solved’ in any or all of the layers above it. So, solving for seven is still a failing grade.

For instance, readers would undoubtedly prefer (level 1. Reader Need/desirability) a news site with no ads. But that would violate the need for revenues (level 2. Business Need/viability.)

And the further apart the levels the more complex the negotiations and solution will be.

Imagine a preferred design that dictated the use of triangular photos on the homepage. We could undoubtedly build the template and create a script to automatically crop images and avoid manual effort. But triangular photos are such a radical departure from accepted journalism norms the newsroom would likely reject the idea. (Possibly rightfully so.) So, all of the work to develop a product solution that worked from levels 3 to 7 would have been wasted, because we failed to solve for culture (level 8.)

So why have newspapers struggled with digital?

  1. Our guide star is often an internal business need, not an external customer need.
  2. We have failed to modernize many of our backend systems (starting with the CMS) causing many product strategies to fail at that level.
  3. In the name of expedience we launch features and products that do not meet the minimum requirements of all eight levels of the product stack.
  4. Our organizations are often siloed in a way that can return ‘false positives’ in the product stack test due to the inaccessibility of data.
  5. We have not explicitly enough discussed the complexity of the full product solution stack, so we have under-estimated the difficulty involved.
  6. We have failed to invest in the product and development resources that could help identify and resolve these issues.

When we solve those issues and are building products that ‘pass the test’ then we can go back to arguing about the ‘original sin‘ of digital journalism. Until then, we are still learning how to do digital media and are not ready to write the post mortem. Time is short, but the solutions are at hand if we take advantage.