What publishing platforms should share

The first question I got asked after moving to The New York Times was whether running a single large site was easier than running dozens of brand sites. A year and a half into the job, my answer was that it was not easier. It was different work, with different rewards and different kinds of pain. The shape of the complexity had changed; there was no less of it.

I had spent years at Condé Nast responsible for the technology behind GQ, Vogue, The New Yorker, Wired, Epicurious, Glamour, Architectural Digest and many other brand sites. The count drifted a few sites either side of sixty, depending on regional editions and microsites. I moved to the Times in May 2011. Comparing the two jobs has forced me to rethink where complexity lives.

Suppose a photo editor has chosen a picture and cropped it to draw attention to the person whose story it tells. The picture is ready. Then someone adds another place to display it, with a different shape, and the editor has to start again. Some of that work may be necessary. Some may be work the publishing system should have remembered.

Publishing platforms should share work that remains useful across products while leaving room for decisions that belong to a particular audience. That requires more judgment than putting several websites on one codebase. A shared codebase can still ask an editor to make the same decision repeatedly.

Common capabilities, different products

Each publication in a portfolio has a voice, an editorial leadership, an audience and a commercial team. The work I remember most vividly was deciding where the shared platform ended and the brand-specific layer began. A publishing system could be common. A taxonomy could be mostly common with brand-specific extensions. A video player could be shared with skinning. A fashion commerce integration and a recipe tool had different jobs, as did their homepage tools. Technology should let each brand keep what makes it distinct and share what can be shared underneath.

Every shared system became a negotiation. When one brand wanted a feature another did not, someone had to decide whether it belonged in the shared layer with a toggle, in an extension through an agreed interface, in a brand-specific fork, or not at all. Those choices accumulate over years. A setting can accommodate a useful difference; a growing collection of special cases can make the common system harder to understand and change. Get the decisions right and the portfolio scales. Get them wrong and you end up with sixty different snowflakes wearing the same branding on the login page. Make the choices while the people who understand the products can still explain why the differences exist.

A large publication faces its own problem of coherence. Readers expect navigation, accounts, subscriptions, search and section fronts to belong to one recognizable product. Behind that experience, Opinion wants things on an article page that Sports does not. The Magazine needs layouts the daily report cannot share. Recipes have requirements a general news CMS may not cover, and interactive journalism produces work a standard template cannot render. A live report brings another set of editing and presentation needs. The work includes deciding which concepts mean the same thing across those experiences and which differences the platform must support.

Shared services also concentrate responsibility. A breaking story can put heavy load on services used across a publication. A change to a widely used article page can alter millions of readers' experience at once. If a shared paywall rule is wrong for thirty minutes, the mistake can reach the readers served by that rule across many sections. The reach of a decision depends on what shares it, not simply on the number of domains.

I think of the portfolio problem as integration and the large-publication problem as coherence. Across brands, the questions include common taxonomy versus brand-specific tags, shared versus separate identity, and cross-brand versus per-brand analytics. Inside one product, the questions include article-page variants, paywall rules serving different kinds of work, and APIs that must serve the web, iPhone, iPad, Android and third-party developers.

Both require negotiation between shared capabilities and specific needs. Counting websites tells me little about the difficulty of that work. One domain can contain several applications; many websites can depend on the same service. Neither model removes complexity. They relocate it into different dependencies, configurations and agreements.

Carry an editor's choices forward

Luke Vnenchak's June 17 account of Scoop, the New York Times content management system, offers a concrete example. Scoop manages content and publishes data for other applications to display. For photographs, editors draw a master crop and a square thumbnail crop. The system uses those choices to suggest the other crops, which an editor can refine.

The possibility of refinement matters. A crop expresses a judgment about what belongs in the picture. Suppose a wide photograph shows a speaker responding to someone in a crowd. A narrower version might retain the speaker's face and lose the person being addressed. The picture would still fit its allotted space, but something worth seeing would have disappeared. Reusing an earlier decision should give the editor a useful starting point while leaving that judgment open.

I would want the tool to retain the distinction between an adjustment for a particular use and a correction that should affect every use. If an editor changes the crop for a portrait-oriented page, that should not silently replace the crop chosen for a wider page. Conversely, correcting a mistake should not require hunting through every destination separately. These are different intentions, even when the editor's hand makes the same movement on the screen. A system needs some way to tell them apart.

Vnenchak's account also describes unfinished work on section fronts, the pages that collect a section's stories. At the time of his post, editorial ranking was tied to the desktop layout and then translated for mobile. The team was reconsidering how to represent importance so it could work more directly across layouts.

I find that a useful limit on the idea of reuse. Giving a story prominence involves more than assigning it a rectangle. If the same editorial priorities apply to a new layout, preserving those priorities could spare an editor from rebuilding the arrangement. If the new product serves a different audience, its priorities may need to change. A local reader's leading story need not be the national reader's leading story. Making both products obey one ranking would erase a legitimate difference.

Organize for the decisions people share

In his 1968 paper How Do Committees Invent?, Melvin Conway argued that an organization's communication structure constrains the systems it can design. For a shared publishing platform, I would start with the decisions that require people to talk to each other. Photo editors and the teams building new displays need an agreed meaning for a crop. Editors and application teams need an agreed meaning for prominence. Putting those groups in separate reporting lines does not remove their need to make these decisions together.

That relationship shows up in the work of organizing teams. In a portfolio, a central platform group and brand-aligned teams spend considerable effort deciding who owns what. In a large shared product, teams aligned to sections or capabilities need enough platform coordination to keep their work coherent. These are tendencies I have seen in organizing the work.

One consequence I did not appreciate before moving was that recruiting patterns differ. For a portfolio, I recruited engineers comfortable with diversity and context switching, and platform engineers who could say no politely and often. For a large shared product, I look for engineers who can hold a substantial shared architecture in their heads, care about coherence and ship into heavy traffic without flinching.

If you like integration problems, negotiated interfaces and a wide range of reader needs, a portfolio is rewarding. If you like coherence, scale and the craft of a product millions of people use, a large publication is rewarding. Both teach things the other cannot. The work I do on a given Tuesday at the Times is no less complex than the work I did on a given Tuesday at Condé Nast. The complexity lives in different rooms of the house.

Try a new destination

Before adding a shared publishing capability, I would ask an editor to prepare an already edited story for a proposed new destination. Watch what has to be done again. For each repeated task, ask whether the information needed to do it already exists in the system. If it does, find out why the new destination cannot use it. If it does not, ask whether something was lost in the earlier workflow or whether this destination requires a new choice.

The answers should change the design. A missing connection between applications calls for engineering work. A decision that was never recorded may call for a change to the editing tool. A different editorial purpose calls for room to make a different decision. Counting repeated clicks alone would miss the distinction: the same click can be needless repetition in one case and necessary editorial control in another.

This test also puts a useful burden on a proposal for a new shared system. Show what happens to an editor's work when another website or app joins it. There may be considerable engineering involved in supporting that destination. The editor should not inherit avoidable repetition because the software discarded a choice it could have carried forward. I'd rather have the editor looking at the next photograph.