Building a conversational app for enterprise users

Lessons learned from launching an enterprise-grade Slack app to a global team with 2M users

According to The Help Desk Institute, 58% of corporate employees say they “hate calling the help desk” and 74% say they’d prefer a chat-based alternative if it worked. On average, it costs large enterprises $25 per help desk call… but only $1.50 for a call to a self-service channel via automation.

To help, we’re proud to be launching our app for Slack today. Thanks to the Slack platform, we developed it in about a month and have been testing and refining it with support from millions of employees at companies like McDonald’s and adidas over the past eleven. It’s finally ready.

Astound customers have large volumes of support tickets and are metrics-driven. They measure success based on how many calls are deflected “left” from experts to self-service channels. They have complex infrastructure, rely on networks of vendors and outsourcers, and have service level agreements (SLAs) with end-users that involve penalties for downtime.

Here are the top five lessons we learned launching our enterprise-grade Slack app…

Lesson one: “Better” competes with “familiar” in large enterprises

Employees in large enterprises are accustomed to specific tools and processes. New ones must not just be better. They need to be so much better that the high switching cost associated with internal training, updating policy docs, and replacing old systems is justified. Re-training millions of employees isn’t trivial.

We rely on onboarding capabilities in Slack to push tutorials to users and sense when they’re frustrated. We also focus ruthlessly on the user experience — we analyze every click, question, and reply to make sure the experience using Astound is always better than calling the help desk.

Lesson two: Enterprise employees are global

They don’t speak one language or have a common culture. They live and work in different time zones. We needed to adapt the virtual agent to accommodate all of the ways employees expect support. We needed to be better than the best live agent every time.

To achieve that, we developed a platform that lets Astound admins configure dialogs and monitor traffic to understand usage patterns. Using AI and natural language processing, we automate the process of improving dialogs so every interaction is natural, efficient, and leads to the best answer.

Lesson three: Work happens across workspaces

Large enterprises use Slack differently from small teams. Related conversations happen across workspaces. Shared services like IT and HR require dedicated channels accessible from any workspace. We rely on Slack Enterprise Grid capabilities like enterprise shared channels to ensure answers to common questions are accessible globally.

Lesson four: Cultivate power users

We conducted focus groups consisting of five to seven “power users” weekly throughout development and user acceptance testing (UAT) phases. This is the single biggest reason our launch was successful.

Without asking users, we never would have appreciated the priority of custom slash commands, compressing dashboard widgets before sharing as attachments, or unfurling custom URLs. Create artificial opportunities to engage and empower users. They’ll provide feedback before launch and evangelize your app to everyone else when it’s live.

Lesson five: Engage InfoSec early

Data security is our single most important product feature. Large enterprises employ whole teams to ensure access rights are defined and enforced by every app. For example, every trouble ticket or HR case may contain personally identifiable information (PII). Visibility of PII must be restricted to its owner and the fulfiller of the request.

To ensure our app was CISO-ready, we proactively solicited security questionnaires from each of our design partners. We had our app pen-tested and we remediated vulnerabilities. We also implemented security procedures to restrict access to our GitHub repos, harden our multi-tenant infrastructure in AWS, and log security incidents. We waited to launch until we felt confident we would pass any reasonable security audit.

To ensure no PII is shared in public Slack channels, we rely on Slack integrations with vendors like Okta and AirWatch to manage authentication tokens and comply with policies. Astound AI machine learning models exclude PII and are trained on everything else so the accuracy of answers improves over time.

The Slack platform made it easy for us to develop an enterprise-grade app

Developing on the Slack platform accelerated our time to market by months. We have ambitious plans to extend the Astound virtual agent to other service domains beyond IT and HR. To get there fast, we’ll continue to invest in Slack knowing the foundational components we need — cross-workspace sharing, integrations, role-based access, and secure storage — are available now and were developed based on the same enterprise principles we follow at Astound.


Dan Turchin is the Chief Product Officer and co-founder of Astound. He’s passionate about building great teams that build great products that solve hard problems that change lives. And he can recite Jabberwocky. Follow Dan on Twitter @dturchin.