If you’re awake in 2023, then you feel the buzz around AI and all the ways it will affect our lives in the coming decade. ServiceNow, in particular, is placing an emphasis on how AI will transform how we work and develop applications on the ServiceNow platform. Yansa Labs has always been on the bleeding edge of ServiceNow development, and we’ve been early to experiment in applying AI to our own internal processes before rolling those practices out to our clients.
At the 2023 Knowledge conference, a team of software engineers from Yansa Labs took home the CreatorCon Hackathon top prize for Generative AI, building a clever tool for summarizing SDLC content such as release notes. This exercise impressed on us how AI might automate the daily tasks here at Yansa that are repetitive, arduous, or easily overlooked.
One area where we could use help is marketing – social media, in particular. We’re humble developers, and touting our wares on social media does not come natural, but it is necessary. How could we apply AI to help us here?
Enter Tracy Social AI!
As a proof of concept, we set out to build an app on the ServiceNow platform to grow our online presence by automating the generation and publication of social media posts. We named a few requirements:
- Make it extensible – With the ever-evolving AI landscape, we don’t want to tie ourselves to one AI provider – rather, we want to enable ourselves to plug into the latest AI technologies, whatever they may be. Similarly, we want to be sure we can use our solution across any now-or-future social media platform.
- Make it flexible – At Yansa Labs, building flexible tools that work under any circumstance is the name of our game. If we’re going to build something, we’re not cutting corners – we’ll build it configurable and flexible to work for any customer in any industry.
- Make it easy – Demystifying complex processes is an important part of the software we build. This solution needs to be set-it-and-forget-it, and it needs to ‘just work’.
What to call it was easy. We already have a name for the virtual assistant that performs our simple automations at Yansa Labs – we call them Tracy. This would be Tracy for Social Media using AI = Tracy Social AI!
From there, we started building. A few days later, out popped our first prototype, and it worked pretty well! Integrating with OpenAI for AI functions and Ayrshare as the shim between ServiceNow and social media platforms, we were able to provide simple prompts and generate simple posts for Twitter.
However, we quickly ran into limitations:
- OpenAI doesn’t know everything! Particularly when it comes to a small enterprise like Yansa Labs, OpenAI doesn’t know our history, our culture, our reputation, or our plans. Also, much of what we need to post about involves new products or unreleased services. We needed to be able to educate OpenAI about these things.
- The devil is in the details. Tracy would frequently create ‘good enough’ content, but we need it to be ‘great’. We needed the ability to be very detailed about personas, the context in which the content is created, instructions for a specific post and post timing. Even after the content is generated, we needed the ability to perform automated transformations to do things like add @mentions, insert helpful links, and remove any questionable content.
- It still takes too much time! Even with our app on ServiceNow, somebody still needed to create each prompt, click the buttons to generate the content, review it, and click the buttons to post. More automation would be necessary.
Over the course of several weeks, we enhanced Tracy Social with an array of reusable components to aid in our quest for the best AI-generated content and automations:
- Context Statements, Instructions, Prompts, and Personas that allow us to create standardized pieces of instruction one time and then reuse them across multiple posts, helping to streamline the process and maintain a consistent voice.
- Content Transformers to intercept, edit, and augment AI-generated content with specific details for our use-cases.
- A Scheduler to automate periodic generation and posting of content at a cadence that works for us – lights out, light speed!
- An Approval and review process to allow us to, if desired, review and approve the final draft before it gets out the door.
Finally, because we’re Yansa Labs and this is ServiceNow, we went the extra mile to make it both easy and as powerful as possible. A beginner can get started with just a few clicks, but experts can code, referencing their unique data within ServiceNow to drive AI-generated content that is unique to their organization. It looked great!
But, would it work in real life, and what would people think?
We started letting Tracy out into the wild. If you follow us on LinkedIn, you may have noticed an up-tick in our chattiness over the past couple of months. You’ve been a part of our live testing phase! Since August 1, 100% of our content on LinkedIn has been generated via Tracy Social AI. In that time, we’ve posted every 3-7 days.
Tracy is reactive to the work we’re doing. While some content is posted periodically to keep our followers engaged, other content is generated based upon changes in our ServiceNow environment. For example, posting about our Recertification Services when the latest ServiceNow release hast just reached General Availability, or posting about recent product updates based upon our product release schedule.
So, how is it going? As developers & engineers, we sometimes cringe at the salesy social media content that is generated. However, we have to push ourselves out of our comfort zone to be successful in that space. In general, Tracy has done a great job of keeping us relevant and engaged with our followers on LinkedIn – far better than we’ve ever done ourselves. The numbers don’t lie! Since we began, we’ve seen these changes to key metrics:
- 778% increase in followers
- 1000% increase in reactions
- 375% increase in reposts
Our number of Impressions & Clicks increased from a flat-line & immediately picked up speed – it has stayed high as we’ve maintained a consistent posting cadence. All of this in only 2 months time!
What started out as a proof-of-concept for Yansa Labs has now become an important part of our marketing strategy. We’ll continue to leverage Tracy for our social media presence. We’re also looking at ways to let Tracy handle internal business communications around release schedules, periodic task assignments, and more via our internal Slack channels. I can assure you that this blog post was completely human-generated. However, because of the flexibility we built into Tracy Social AI, it could be used to automate blog content and publication. So, maybe in the future?
Of course, once we realized how much Tracy was able to help us, we wanted to offer these capabilities to our own customers. Today, you can find Tracy Social AI as a certified application on the ServiceNow Store. We’d love to have you try it for free. It’s great fun to setup and test. Who knows, maybe you’ll find that Tracy will become an important part of your marketing team?