Voice of Customer (VOC)

Voice of Customer is a platform that helps product teams gain insights from inbound customer data. It enables companies to capture, analyze and act on customer feedback quickly and easily.

Comment

Inspiration

Hello Judges! Welcome to our P.O.C of what we call the V.O.C but don't just take our P.O.V

"We have a community manager who is in-charge of consuming appstore reviews, support tickets send to product and reply to then. She then summarizes learnings each weeks and sends themes (good and bad) to product teams.” - Product Team at Yahoo Mail (listen to 2-min clip)

“Product Ops is literally going through and like reading through verbatims of like hundreds and hundreds of, of things. Like she can't go through literally everything. She checks all the inbound channels, outbound channels. For one very specific thing - driver feedback” - Product Team at Lyft (listen to 1-min clip)

“I don't think like 1 PM can do that unless they're like dedicated to this (analyzing inbound feedback) type of work, especially with the amount of customers that we have to listen to. So our Product Support Team triage all product support tix and really surface to me like the big things that I wanna put into my roadmap like 10% of our support tickets talk about how how difficult it is to create a budget” - Product Team at American Express (listen to 2-min clip)

“In my dream world, there would not need to be anything that's manual. I can just go to a dashboard and see are we trending up? Are we trending down? What were the themes (of support tickets) in the last week? Because there's some sort of intelligent tagging of all the different tickets so I can figure out, what sort of areas we're causing more volume or more complexity. And then they'd have some sort of way of flagging. Like here's a few cases you should go look at or here was something that surprised us. And they (Product Support) wouldn't even have to send me anything. I could just go and self-serve to get this data.” - Product Team at Netflix (listen to 2-min clip)

With ever-growing digital transformation, we see that product teams are on constant pressure to build for the needs of the customer. However, when you are serving millions of customers, it is challenging to listen to the Voice of the Customer. There is more feedback than ever; sales teams ccing you on emails where they lost deals, support ccing you on support tickets for complaints/bugs, success team sharing customer feature requests, your colleagues sharing ideas on Slack, execs asking for top-down requests, and of course the market itself pushing you in strategic directions. Well, where does all this feedback live? It lives in those respective team's tools: salesforce, gainsight, email, zendesk, slack, appstore, twitter, etc and is not easily available when product teams needs to make decisions.

What it does

Voice of Customer consolidates user feedback into a single repository. Doing so, product teams can make custom-centric decisions, not just follow their gut, the loudest, or the highest paid person in the room. VOC also allows product teams to avoid reading every feedback that is coming in; instead, we can provide a dashboard to always answer the question going into strategic meetings: "what are our top 10 user problems atm?"

How we built it

Prototyping

We decided early on we wanted to use docker/docker-compose to build locally and use Kubernetes/GCP for a cloud deployment because that is what the engineers were used to using.

We started with some prototyping in a jupyter notebook to engineer the prompts for GPT-3 and the web scraping code in a more casual environment.

V1

Once we had that relatively down, we moved it over to a simple flask app with celery on the worder side using redis as the message broker. Again this is what the BE engineer was used to so that was easy to spin up.

Once we had a working v1 locally for both BE and FE, we spent time deploying the first version to the cloud which took some time/multiple attempts.

V2 - Final Version

We played around with the data and chatted about what other graphs we wanted to implement and we even added a new GPT-3 prompt to do some filtering on noise in these crowded data sources. We then deployed V2 which has more graphs and more filtered data in insights

Challenges we ran into

Engineering perspective

Deploying to the cloud took a few attempts because none of us are seasoned Devops people so the BE engineer had to figure some stuff out but it was resolved relatively quickly. the Docker image for the inference/ML service started out at a whopping 7 gigabytes but we took time to get that down to < 3 gigs. Also we tried to do a cron job and it worked locally but we couldn't get it to work in the cloud so that was difficult. c'est la vie.

Product perspective

It was challenging to get the team to empathize with the problem, which is standard in real life. I, the PM, had the voice of the customer in the head and I can relay that to the team, but it is not as powerful as the team listening to the actual customers. What helped the team was listening to clips from product leaders at established companies and finding trends among these calls. From then on, we were all on the same page. (hint hint: this is too where Voice of Customer AI can help).

Accomplishments that we're proud of

Engineering perspective

Scraping data from three sources and automatically labeling them with feedback is usually something done over time with humans in the loop. All team members know this because we've been involved in this process at various levels. It's cool to see something like working so well in < 48 hours because it speaks to how much this industry will evolve in the AI revolution.

Product perspective

We established trust relatively quickly, which was key for a weekend hackathon. This helped us have hard conversions about critique and feedback to progress. One of the best things we did was draw hard lines when creating the teams on each person's function: Shiv as Product, Kevin as FE, and Sinan as BE. This helped in ownership responsibilities.

Overall, we added fun and play throughout our experience together with pauses to huddle on Slack, send memes, jokes, and even casually roast each other (mostly Sinan, hehe.)

What we learned

Engineering perspective

Docker images for ML services can get huge without proper foresight and care

Product perspective

Breakdown the problem into bite-sized chunks of customer value. We knew we couldn't not get proprietary data from product leads of their Zendesk, Salesforce, or Slack data, so instead we went with data that is open to public: Twitter, Appstore, Reddit, G2 Crowd, etc.

What's next for Voice of Customer

We will setup follow up calls with Yahoo-mail, Netflix-identity, Lyft-driver, American Express-budget, Amazon-pets, Paddle-payments, Angel List-syndicates to present our prototype based on their feedback and ask these teams if we partner together to do a 6-month pilot with them.

Built With

Share this project:

Updates