InstaUP uses AI technology to boost engagement; it automatically selects the most engaging photo, edits and crops the image, suggests hashtags, and automatically posts at the most impactful time.
Lauren Kainski, Tanay Chaturvedi, Seth Bergeson (MBA Candidates, Foster School For Business)
Carlo Del Mundo (Xnor.Ai), Kiron Lebeck (Phd Candidate Allen School), Rosa Thomas (AWS), Max Smiley (Microsoft) Arthur Liang, Ben Ihrig
Class Project In Entrepreneurship Course On Starting A Business And Pitching To Venture Capitalists
InVision, Adobe After Effects, Adobe Premiere, Sketch, Illustrator, Keynote, Google Slides, Instapage
A Slide Deck Pitch, Executive Summary And Financial Outline
This project highlights my experience designing for an A.I. algorithm and working with engineers and business students.
With recent changes to Instagram’s algorithm, which prioritizes posts based on past engagement, organic reach has decreased significantly. Only an estimated 10% of audiences see content. In this new environment, both influencers and marketers need more tools to increase their engagement. How might we create better tools for influencers and aspiring influencers to optimize engagement?
Who is an Influencer? An influencer is usually defined as an Instagram user with over 2K to 10K followers and earns on average $180 per post.
Finding Influencers & Signal
Targeted micro-influencers on 3 verticals
We created an instagram crawler that generated 1,105 contacts in the Beauty Influencer category. We reached out to these contacts. We also employed Instapage with targeted Facebook ads to measured clicks through scraping and targeting instagram influencers on Facebook. Our primary audience was Female and Males 13-19.
*Drove 60 Clicks to the InstaUp website for an average Cost per Click of $ 0.34*
To understand the marketplace we looked at a broad range of products.Instagram has 800M users and ad spend will increase by 88% this year to $6.9B. Seventy percent of businesses (15M) use Instagram in their marketing efforts. Ad spend on influencers will increase by 50% this year to $2.4B by 2019.
We reviewed 74 products. Competitors provide similar tools as InstaUP but none of them analyze the 80M photos posted per day on Instagram nor do they integrate the four functions InstaUP provides. Competitors fall into three categories: social media management, analytics tools, and targeted solutions. Many freemium photo editing apps exist, and influencers and marketers often use several of these apps in conjunction with scheduling or analytics apps. InstaUP integrates these core functionalities into one intuitive app while also leveraging the billions of photos on Instagram to refine its algorithm.
85% influencers measure success from likes/follows and 46% from comments
50% + of influencers spend 30 mins to 3 hours per post
54% of Instagram users want to be full-time influencers, but aren’t yet
Conducted 21 Interviews
We spoke to Instagram users who had between 2,000 to 208,000 followers. Conducted expert interviews with the the former Marketing Manager of OfferUp and Co-founder of TBH (which was sold to Facebook in 2017).
01. The product must make people feel like it is offering new knowledge and insights.
"You're selling the idea that machine learning can provide more value, augmenting their photo selection process. But actually, it's the _belief_ of the outcome that attracts people, users don't care how it's done under the hood."
CoFounder of TBH
02. Algorithm changes at Instagram cause a lot of unrest and confusion. Influencers are not clear as to why some posts do better than others.
Sometimes a picture will go viral and you would never have guessed it and other times a picture will only reach a fraction of your following.
@your_daly_dose_, 60.8k followers
I struggle to understand why some of my pictures do better than others. This product would help me be more proactive with my content.
@Shwetarolaniya, 17.7k followers
03. Influencers want to create unique, authentic content that doesn't feel repetitive.
“The hardest part is definitely generating new and exciting content. There's SO much out there now and a lot of it is repetitive. I can only look at so many photos of Glossier You before I just get sick of it and don't even want to think about the scent. I know my page is probably more saturated than your average beauty enthusiast who follow beauty accounts, their friends, their favorite restaurant, etc. But it gets hard, and it makes it hard to set yourself apart."
04. Content creation was time intensive and many influencers wanted to spend more time engaging directly with followers or managing administrative tasks, especially with cross platform demands.
"Aside from testing, writing the blogpost it takes around three hours, on top of that comes about one hour of picture taking and editing and another hour for scheduling promotional tweets/Instagram/Google+ etc., In total between 4-5 hours. Videos take around six hours, accompanying blog post included."
InstaUP’s algorithm uses convolutional neural networks (CNNs), a popular computer vision technique, for image selection. To evaluate the technical feasibility of CNNs on the influencer space, InstaUp's first CNN model focuses on self-portraits ("selfies") since well-curated datasets for selfies already exist. InstaUP's CNN is trained on 46,836 selfies. An image is scored from 0-100% with scores closer to 100% suggesting higher engagement.
CNNs provide a data-driven foundation for inferring "how" engaging an image will be, but CNNs do not adequately describe "why" an image will be engaging. To close this loop, InstaUP uses Google's Cloud Vision API to further analyze image content based on smile, eye contact, clarity, position, empathy, posture, and movement.
Training the Model
I worked with the engineers to look over the photos selected by the model. I noticed the algorithm seemed to prefer smiling photos which seemed antithetical to the goals of many instagram followers. We retrained the model so it was not as biased towards smiles. Ultimately the plan was for the product to scrape actual Instagram images so the photos would be more in line with the users goals.
Creating a Product
The group was committed to using machine learning, but had no other clear ideas about the design of the product. We considered a very lean product with a quick to market strategy as well as a more robust design with a number of features. We decided to add a few existing core features to the machine learning to automate the whole Instagram posting process. Auto-scheduling and hashtag optimization are existing features in a few available products and the engineers felt they would be able to recreate these features with a little bit of time.
To foster better communication to my team I created a wireframe.
Selling the Metaphor
From user interviews I hypothesized that people wanted a multi-purpose app that could offer a variety of professional quality experiences like a photo studio app on their phone to streamline posting to Instagram. I used this story to create the visual design. The colors in the app were based on walls in a photo studio and the navigation language followed the metaphor of a studio.
the most impactful photograph with custom filters and auto-filters based on the machine learning
offers a list of popular hashtags currently trending on Instagram
recommends the perfect time to initiate your post
We were able to follow up with two influencers for feedback One user was very excited about the features, but struggled to maneuver the app, the other was suspicious. I conducted an additional usability test with two users and made several core changes:
01. Created seamless integration where the best photo is pre-selected in your camera role so you don't need to upload anything
02. Added filters based on user feedback
03. Replaced words like "analysis" with photography terms and icons
04. The preview feature was deemed irrelevant because such a feature exists in Instagram
Working with a combination of engineers and business students, on a short time frame, meant we were all rushed. There was a tendency to skip the design process. It would have been great to have more time with our users to understand their needs. Often my colleagues were satisfied with customer validation and did not necessarily want to examine the user's responses. The engineers were consumed by testing the validity of machine learning system, but did not have a sense of how the product would work. At the end of the quarter an engineer came to me and mentioned that it would have been better to wait on the engineering until the product was roughly designed. I took this a big win for design! We all learned a lot about the importance of each other's field and the thinking of venture capitalists. (Not to mention hearing Andy Jassy speak was a treat). It really enjoyed building up some business acumen. I hope to work with cross-disciplinary teams in the future and believe this experience makes me better equipped to suss out a viable product.