🤖 Meet ‘Goldie’

🧩 The problem

Customers were going straight to our support team for help with simple issues, driving up costs and overwhelming the team. We needed a way to intercept common questions and empower users to find the answers themselves before they submitted an email to our support team team.

💡The plan

We wanted to add some friction to how people could contact support, without making it a frustrating experience. The idea was to serve information directly to customers so they could quickly get back to working on their project. After looking into industry trends and our own customer behaviors, we landed on a chatbot as the right solution.

🧠 My role

As the content designer, I led the conversation design, mapping out bot IA + flows, and governance planning for the bot.

🔍 Finding the right problems to solve

To make the bot genuinely useful (and not just a gatekeeper), I partnered with Research and Customer Operations to review hundreds of support cases. We grouped them into themes and pressure-tested each one against the ✨Magic Formula✨:

✅ Top driver of support case volume.
✅ Solvable within 1–2 emails with support.
✅ Something the customer could solve on their own, without needing to contact support.

If the issue met all 3 of these criteria then we knew it was a good candidate for a bot answer. This framework helped us prioritize what the bot should handle and ensured we were solving real pain points in a user-friendly way.

🛑 Challenges

Chatbot fatigue

If customers are trying to contact support, it’s likely they’re already feeling frustrated. We wanted to avoid making it worse by having a bot that felt like it was blocking people from getting help.

To help with this, I worked with the support team to draft questions we could train the bot on. This way we were using real customer language, using words and phrases they would use when writing into support about the top topics. This resulted in answers being more findable which led to getting help faster.

Account specific support + security

Another challenge is that certain levels of support are only available on certain plan types. To keep information inside accounts secure, the support team can’t disclose too much information about accounts to anyone other than the owner of the account.

I created 2 different sets of flows, one for logged in users and one for logged out. This way we could cut back on noise for users by tailoring the categories and answers to their specific needs.

For example, logged out users would see the flow about what to do if they couldn’t access their account which logged in users wouldn’t need to see.

✍️ Bringing Goldie to life

From there, I wrote 100+ conversation flows focused on:

• Making the bot easy to navigate with a clear IA and distinct categories.

• Keeping responses short, clear, and friendly.

• Using language that customers use to describe their issues when writing into support.

I also created a style guide to align tone and structure across all flows and documented a governance plan so content designers and bot managers could work together on updates post-launch. When we eventually launched to more SurveyMonkey products, these resources made everything go a lot smoother and ensured consistency between the bots.

💥 The Impact

The bot successfully deflected 28% of cases from making it to support—saving the company over $700K/year in support costs. We rolled out 3 more bots for other SurveyMonkey products and support teams, continuing to lower costs across departments.

💰Bonus: the automated refund project

After the initial success, we looked into ways Goldie could help with the #1 case driver: refund requests. Typically, this is a manual process where support needs to review each case, check eligibility against certain criteria, then process the refund and respond to the customer.

I partnered with product, engineering, and customer operations to design a refund bot that could automate that process.

Unique challenges

• Customers who want a refund are already frustrated, so we knew this experience had to be as seamless as possible. Especially when a bot was involved.

• This is the first time Goldie would be looking into customer activity + taking an action on their account. We had to align very clearly with engineering what copy would trigger which actions.

The result was over $100k savings in customer support costs in the first 3 months!