Facebook's 5 million advertisers used two main desktop interfaces, Ads Manager and Power Editor, to spend over $100M/day. Power Editor was built for Facebook's largest advertisers and Ads Manager was for everyone - the smallest to the largest advertisers. Over time, the interfaces became more similar in what they were able to do, but different in how those tasks were performed.
“Convergence” is what the team called a two and a half year project to combine these interfaces. Converging Facebook's two largest ads surfaces gave Facebook's partners and advertisers a single interface, the new Ads Manager, which improved product performance, streamlined advertiser workflows, and helped teams to move faster. The new Ads Manager combines the best of the old Ads Manager and Power Editor for all advertisers.
My team used a stepwise approach in which one feature was combined, shipped, tested, and perfected at a time. I joined the project to work on converging publishing models about a year and a half into the larger project, and for the last six months of Convergence, became the lead designer responsible for delivering the final product.
I was the sole designer on a cross-functional team of between 4 and 8 engineers and a PM, collaborating with a content strategist and researcher. The entire redesigned publishing flow and the details of testing and launching the converged interface were my responsibility.
Converging publishing models was the biggest hurdle to overall convergence. The publishing model I shipped provided advertisers with the right checks and balances based on the weight of the edit they were making:
In the edit sheet where advertisers were bulk selecting and editing ads, edited items were automatically saved as drafts
The edited items could be published with no review, discarded, or bulk published
Edits and new items from an entire session could be reviewed and published in bulk using the button at the upper right
In the table, edits were live upon hitting ‘Enter’ on the keyboard or clicking ‘Publish’ in the cell
Advertisers found the new model efficient and helpful. In addition, this design increased publishes, edits, and revenue.
Publishing refers to how advertisers get an ad up and running to actually spend the money. The team had known from the beginning of Convergence that the biggest Ads Manager/Power Editor disparity was the publishing model:
Ads Manager: Edits and created items in Ads Manager either are “live” and delivering ads right away or are discarded
Live edits ensure efficiency and ability to handle single and small groups of items without extra clicks
Power Editor: Edits and created items in Power Editor are automatically saved to a draft state and must be reviewed before being “live” for delivery
Drafts provide security against spending unintended money and the ability to handle lots of items at once
The main goal of the project was to figure out how to keep the benefits of each model and minimize the learning curve. We needed a system that minimized mistakes and maximized efficiency.
This is Power Editor circa 2015. Power Editor gives advertisers the security of being able to double check their work, especially when they have a lot of items, before spending any money on it. Every change is saved as a draft that must be reviewed.
This is Ads Manager circa 2015. Ads Manager's publishing model is efficient in that it needs fewer clicks and gives advertisers instant feedback on what's running. When a change is made, it is immediately live, as shown below.
We came into this project with a couple of principles based on previous research. We learned when there were a lot of conditions for object state, people found the rules of what was published hard to understand because of lack of consistency. So our first principle was
Provide one consistent rule for what happens when an edit is made
The second principle I developed with the team came from wanting to minimize the learning curve for the largest advertisers — the power users with established workflows with drafts. The advertisers who didn't have drafts had often asked for them, so we decided on another principle:
Keep drafts: drafts provide protection from potentially costly mistakes
The draft model at the time only allowed advertisers to publish all changes at once. The one thing we were still worried about was the loss of efficiency for advertisers familiar with one-click publishing.
To help advertisers regain efficiency, I introduced subset publishing to give advertisers control over smaller groups of items. Advertisers would be able to publish a single item or a small group of items within the set of all changes at a time. Subset publishing would result in
Stability and performance advantages
Not having to Q/A items unrelated to the larger set
Easier onboarding for non-draft users
The subset (a single item, small group, or all changes) depended on which part of the interface they were publishing from. There were three ways to publish a change, each with a different level of precision
A button in the top right to review and publish all changes. This was a carry over from Power Editor, which already had this button
A button inline to show that an item has unpublished changes, and to publish the single item it describes
Buttons in an edit sheet to review and publish all items selected for editing. This was a modified carryover from Ads Manager, which allowed advertisers to publish or discard after editing but not save to draft
To give advertisers one rule to remember for how to publish from anywhere, we emphasized that for each change
The change is saved as a draft
It must be reviewed by the advertiser
The change can be published from the review step
As mentioned, this method relies on having everything go through a draft step. This was a change for the advertisers who hadn't had drafts in Ads Manager before.
Every change or set of changes required three clicks to publish.
The team's researcher conducted a week-long, 60-participant diary study on Convergence as a whole. We got a lot of responses like this:
"Editing budgets and grids added an additional step of having to approve the change... for some actions (i.e. budget changes) . I found myself easily forgetting."
- a FB advertiser
While we caught some feedback that the model had a learning curve, it wasn't enough to take a step back and reconsider. It wasn't until we released Convergence to 10% of all advertisers that we realized we'd need to do some rethinking.
Creation of new items dropped by 4%, and editing of existing items dropped 11% in test versus control. Data logging showed this wasn't because people lost interest, but because they were not making it through all the clicks necessary to publish.
They thought they were publishing, but in many cases were not.
If advertisers' mental models are so far away from what we're serving them, it is not a good system. We went back to the drawing board and held off launch for the holiday advertising season.
The researcher and I collaborated on some followup research to find out what problems people were having and which types of advertisers were having them. We found:
Old Ads Manager users who weren't familiar with drafts were confused by the extra steps to get changes running
There were many complaints of too many clicks to make lightweight changes
Looking at the problem from an other angle shifted our thinking. We needed to tune the model to fit people's workflows rather than pursue consistency to the detriment of workflow and understanding.
I reworked the pieces into a mixed model with the amount of review depending on the weight of the editing task. These are videos of a prototype I made for lab research:
Drafts are present in workflows where draft friction is necessary for Q/A, like with bulk edits in the edit sheet or when publishing all drafts through the upper right corner button
Lightweight or optimization changes made in the table don't have the extra steps of review, but publish right away
While it was pretty obvious to publish a change made in a cell in the table, the buttons and what they should do in the edit sheet were hotly debated across the team. The edit sheet is where people make all their heavy duty edits to image and text, schedule, audience, and others.
While we were originally thinking of this as an online shopping experience with a checkout, one-click purchase option, and a clear cart, we finally settled on the model similar to email. I worked closely with the content strategist to make sure this was communicated well. People needed:
A way to tell people that their edits in the edit sheet were automatically saved to draft
A way to let people discard a draft they were working on
An explicit way to close without publishing and leave as draft
A way to publish immediately without a review step, similar to a ‘Send’ button in email
Even though there was inconsistency in whether something was saved as draft or not depending on the location the edit was made, people found this to be more intuitive because the result was based on the weight of the edit. Tests in the lab showed it was a success, so we did a gradual roll out to our advertisers.
The final product brought edits and creations back to the level they had been before Convergence. In fact, edits even increased and there was a positive effect on what advertisers were spending.
Publishing increase by 0.7%
Editing increased by 5%
I'm proud of this project because it:
Maintains all advertisers' existing workflows
Reflects their idea of the right mix of security and efficiency
Balances visual cues for object state and publishing process that advertisers from both interfaces could find meaning in
Makes sure the visual treatment isn't too heavy-handed or too subtle
Longer-term impact means a clearer ads creation and management process, and the ability of teams at Facebook to not only maintain but innovate on the ads interfaces.
Through this project I internalized an important design concept: Let go of ideas that aren’t giving people what they need. Paying closer attention to research and data could help me make sure I am designing for people’s actual mental model.
If I were to do this again, I would definitely also come up with some contingency plans for each main hypothesis. We spent a lot of time coming up with a better solution after the first idea didn’t work. It’s impossible to always predict every case of what could happen and why, but I could have considered other options for high-risk changes before something went wrong.