Applying Generative AI with Swantide
Designing an AI-supported middleware CRM tool to attract larger RevOps customers.
Company: Swantide
Role: Staff Designer
Dates: November 2022 - November 2023
Technology: Converted from Material UIv3 to Material UIv5 with backend in Typescript, Python, and Go
Location: United States (remote)
Overview
Swantide is a middleware software application that automates the setup, configuration, and management of RevOps customers' Salesforce instances. The application enables go-to-market (GTM) teams and startup founders to do this quickly and at a fraction of the price. For every $1 spent on the setup of Salesforce, it is estimated that companies spend $8 on the maintenance of that setup. Swantide drastically reduces this cost.
As their first full-time design hire, I supported several business goals and a few product team goals. The business goals were as follows:
- Support AI features projected by the leadership team and the board
- Successfully sell to larger GTM teams, leading to larger sales prices per client
- Find ways to increase usage from an annual setup to monthly or daily usage
Results
- My work helped Swantide secure a partnership with Salesforce.
- Sales team sold to larger companies and secured their first five-figure deals in 2023 Q3.
- Sales team had the largest of new customer sign-ups for 2023 Q3 due to the release of Metadata Catalog.
- Regular updates and improved user experience secured the company's first three-year locked contract in 2023 Q3.
- Company won Best App for Everyday Use at Salesforce Ben AI Demo Jam 2024.
Process
Scaling Swantide Meant Supporting More Sophisticated Customer Needs
When I started at Swantide, I worked with the CEO, the Head of Customer Success, and the Head of Engineering on key business outcomes that were required for the company. I used this framework to help organize the product and business goals of the company and regularly updated the product through regular customer and internal company feedback:
When using this framework, the 'how' was not as clearly identified, but it did facilitate discussions with the leadership team to lock down clear outcomes they wanted for the near future and further out into the future. I revisited these outcomes regularly as I worked with the CEO and engineering team and as we gained a better understanding of our target customers.
Introducing Agile Practices to Bridge Alignment at Release
In addition to the business goals, there were internal team goals that the CEO and Head of Engineering also wanted to address:
- Provide designs that could help build alignment and ensure solutions are applying standard UI/UX best practices.
This was to help solve the problem that the CEO did not have visibility and predictability into what experiences were being released. The engineering team had mostly newly experienced engineers who were designing and building unvalidated solutions that often did not apply standard UI/UX patterns (design debt) and it was causing issues with scaling Swantide. - Be an advocate and user of the CEO's new Linear backlog.
The engineering team had not been using a backlog, but the introduction of my role posed a real need to provide a means to track work. I needed it to help with understanding the company priorities in real time and manage parralell tracks of work. - Establish regular feedback channels with key stakeholders and customers.
My designs incorporated the regular feedback from the CEO, sales team, customer success team, and customers while also addressing the overall product goals. It was important to balance all of these inputs as I determined a design system that would scale to the appeal and needs of larger customers.
Focusing on More Sophisticated, Technical Customers
When I started at Swantide, many of its customers fell into the maturity 1 and maturity 2 customer profiles (see maturity diagram). Their engagement with the Swanttide application was very high (daily) at the very start of their Salesforce setup journey, but dwindled after completing their Salesforce setup (either once or twice annually). Although we wanted them to return to the application more frequently, the reality was that if Swantide did a good job in the setup, then customers had very little reason to return unless they wanted to make changes to their Salesforce instance. They also typically had other matters within their business that needed their focus and could not prioritize enhancing the capabilities of their setup any further than they already had.
The board wanted our CEO to show that Swantide could attract customers that could pay higher annual subscriptions and to show that Swantide had value past the setup of Salesforce. In order to do this, the team felt they would need to focus more on maturity 3 and more specifically maturity 4 customers. These customers had larger teams, more specialized needs, and larger spending budgets. With more specialize roles and new capabilities to support them, the team felt we could increase MAU and potentailly DAU metrics.
Early Customer Interviews, a Card Sort Study, Heap Analytics and Feedback from Leadership Identified Several Key Areas Where We Could Focus Our Product's Direction
I conducted several studies (interviews, card sorting, and usability testing) to better understand how current RevOps customers (maturity 1 and 2) and potential RevOps customers (maturity 3 and 4) thought about managing their entire revenue cycle (from lead generation to customer retention) and how it would impact the Swantide's product direction:
- Maturity 3 and 4 customers wanted to self manage their Salesforce instances, using Swantide.
These customers typically had larger teams with more specialized roles. As a result, they had the time to learn and configure Swantide, tailoring it for their needs. This resulted with less dependence on our Customer Success team. - RevOps professionals valued observability and monitoring.
Because these teams were larger, there was a need to report data up to the VPs in their company (usually sales and marketing). Having a tool that could assist in this communication requiement would be highly valuable. Ideally, it reduces the time they spend in collecting and displaying the information. This meant design decisions that supported scale and more information on the screen for the technical functionality they expected from the product. - Updates to their CRM’s instance posed the greatest risks.
Updates and changes to their Salesforce instances often caused bugs that could result in a costly (time and money) overhaul and review of the system. Mitigating this risk was highly valuable. - Understanding their metadata was key to updated configuration and maintenance of their CRMs.
Some companies were spending close to $30,000 per contract to have consultants organize, define, report, and clean their metadata. - Metadata was a new term to RevOps professionals that started their careers on the sales side.
This term only aligned to more technical users (maturity 4 customers) who were familiar with configuring their CRMs. We learned there was a large population RevOps professionals who were learning these more technical terms while managing their teams. - Basic usability of the product needed to be addressed.
Analytics from Heap indicated that Swantide had legacy UX patterns that were not standard (design debt), causing problems for our customers and putting more work on our Customer Success team. I needed to manage incorporating updates to the design system to solve for this design debt problem while simultaneously releasing new features.
The Experience Configuring and Managing Metadata was Our Path Forward
Ultimately, we determined as a company that metadata visibility and users' ability to reliably make changes to the metadata was going to be key to attracting maturity 3 and 4 customers. In order to support this strategy, the product could not heavily rely on our small Customer Success team to solve all of our users' problems. We needed a product that could scale with demand of our target customerbase. That meant that while we built these new metadata and generative AI capabilities into the Swantide application, we also needed to improve the usability of the application, driving more customer self reliance.
6 Full Product Solutions Launched to Support Business and Customer Needs
While at Swantide, I was able to contribute to six product solutions that incorporated these requirements:
- Create solutions that support Swantide's generative AI goals.
- Fix legacy UX concerns and make the product easier to use.
- Reduce the time Customer Success spent with solving customer challenges and enable customers to self manage.
Data Uploader
MVP Launch: March 2023
Problem Summary: Customers often struggled to clean their Salesforce data. As a result, they often have duplicates or metadata for particular objects that are missing data.
Solution Summary: Data Uploader was a free tool offered to Swantide customers to allow them to upload, update and upsert data related to Accounts, Leads, Opportunities, and Contacts objects within their Salesforce instance. The team already had an existing version of Data Uploader, my job was to improve the experience and allow customers to Upsert and Update data. Originally a nine step modal series, I combined several steps into a single page and left two more modals to clean the data and confirm the changes transmission to Salesforce.
Outcome Summary: Feedback on the tool was that "it was much easier to use" and that they loved that they could now update and make changes to their existing data sets. All Swantide customers were using it to update their data sets and they appreciated that it came with the Swantide product.
RevOps Search Experience
MVP Launch: June 2023
Problem Summary: Swantide customers struggled to locate workflows and edit them, taking up Customer Success team's valuable time (averaging around 30-45 minute calls).
Solution Summary: We launched an AI-powered search that helps Swantide customers find and edit their Swantide Workflows (templates that automate the setup and configuration of customers' Salesforce instances).There were two versions of of RevOps Search Experience that needed to be supported: (1) an AI search experience prior the incorporation of AI chat feature and (2) an AI search experience incorporated with RevOps AI Assistant later in the year.
Outcome Summary: The search feature reduced workflow search questions to nearly zero and reduced call times related to the topic to less than a minute, but it highlighted a new problem - customers started asking for help on debugging implementation mistakes. Our customers were becoming more self-reliant but it created newer problems for us to solve.
Sandbox Capabilities
MVP Launch: September 2023
Problem Summary: Debugging changes made to customers' Salesforce instances is a time intensive process. These changes sometimes resulted in a Salesforce instance completely crashing, leaving GTM teams running blind with their data. Swantide's Customer Success team spent hours to days helping our customers with these issues.
Solution Summary: Swantide's Sandbox solution allowed its customers to develop, to test, and to train without compromising the data and applications of their production org, preventing the need for our customer success team to undo the problem. Swantide customers could switch between their production and sandbox environments within the app, giving them the ability to test their Salesforce changes before pushing to production and limiting unwanted breaks in their systems.
The team released two Sandbox solutions: 1) a stop-gap solution that allowed customers to switch between a production environment and a sandbox environment and 2) a solution in the new Account page to serve as a location for customers to manage their account settings.
The stop gap solution was a light weight solution that allowed our team to safely test our sandbox product. During this time, the frontend code was going through an overhaul and we needed a lightweight way to provide product value before incorporating AI related features.
Outcome Summary: Upon release of the MVP stop gap solution, our Customer Success team had a 92% reduction of requests to fix full crashes in customers' Swantide instances. This was considered a huge success. Because of the value, we validated that having the ability to self manage their instance was useful and an Account page was necessary. Thus, I started to incorporate an Account Details page into the new AI designs that were to be built later in the year.
My Catalog (Metadata Catalog)
Initial Design Finished: March 2023
MVP Launch: August 2023
Problem Summary: Revenue Operations teams hire consultants to document the metadata of their Salesforce instances. These hired staff can cost up to $30,0000. Documentation of metadata helps RevOps teams reconfigure and debug issues related to the changes made to their Salesforce. Solving this problem would attract larger customers for Swantide.
Solution Summary: My Catalog (Metadata Catalog) is solution to a data dictionary for Salesforce customers. The catalog gives Swantide users a way to organize and visualize every component of metadata in their Salesforce system without the need of these hired consultants.
I designed a way to organize the metadata and to drill down to the details of each metadata component. Eventually, this capability would allow Swantide admin users to see how particular components were impacted from changes to the Salesforce configuration. If admin users can see it, then it is believed that our AI agent could help serve as a copilot to better understand each metadata's details.
Auto-Documentation Powered by Generative AI
Initial Designs Finished: April 2023
MVP Launch: August 2023
Problem Summary: Customers could not easily see how their Salesforce instances were impacted when changes were made. It was even more imperative when they made changes that caused their Salesforce instances to crash. These unwanted changes to their Salesforce instances could cost thousands in dollars and weeks to solve, because it was difficult to see how the metadata was impacted.
Solution Summary: Auto-documentation applied to My Catalog automatically records how and where a component is used throughout their Salesforce instance, as well as, see what the field, object, flow, validation rule, or apex class does in simple business terms without the need of a hired consultant.
Swantide tracks and manages the metadata automatically and visually displays this information in the My Catalog - a data dictionary that allows Swantide users to see the details of their metadata and changes that happen to them after the configuration of their Salesforce instance is updated.
Outcome Summary: Metadata Catalog and Auto-Documentation saved RevOps teams a significant amount of time (nearly 3 months of cataloging work for roughly 1000-5000 components per company for the initial launch) and money (around $30,000 per consult). It resulted in major sales for the company (first five-figure deals with the largest number of new customer sign-ups for a quarter, and the first three-year locked contract in 2023 Q3) and secured a partnership with Salesforce. These two capabilities along with Swantide's Sandbox capability became a strong suite of features that the sales team pitched to customers.
RevOps AI Assistant
Estimated MVP Launch: intended January 2024
Problem Summary: RevOps AI Assistant is in-app chat that provided automated, intelligent support for customer's simple questions about Salesforce best practices to complex debugging scenarios. The RevOps AI Assistant was a tool intended to be a troubleshooting aid that allowed customers to easily seek help for simple questions like asking what a particular flow does to better understanding why a particular workflow stopped working.
Swantide's Customer Success team would spend hours to days trying to help customers better understand the root reasons for their bugs. The hope was that the AI Assistant would reduce this time significantly. At the start we were hoping for 15-20% drop in debugging specific Customer Success calls.
Disagreements on AI Assistant Strategy
Out of all the AI solutions we were building, the RevOps Assistant was the most complex. The CEO wanted to go in the direction of an in-app chat service that provided automated, intelligent support for customers’ simple questions about their Salesforce instance to support for best practices when debugging issues in their instance. In its most ideal state, it would be a debugging collaborator for Swantide customers, specifically when bugs arose from changes to their Salesforce.
I originally believed a chat experience that sat as a layer on-top of the main Swantide app was the least risky approach that still provided the same business and customer value. If the team built the chat experience over the application, we could potentially use it in the Native Salesforce application or any other partnering company as an added layer of help for our customers. This would make the chat more versatile and flexible for our users and potentially meet them where they needed the help most - in the app that he/she was using.
The CEO; however, insisted the chat should not feel like an addition on-top of the Swantide application, but rather an integral and primary part of the experience. She felt the team could later create a separate chat experience that could sit on-top of Salesforce later in the roadmap. Although I felt this required more iterations of work, I also understood that she had conversations with the board.
I created a right-side activity panel that enabled RevOps users to ask their queries while still interacting with the relevant content in the application in the left main-content responsive panel. This activity panel could also handle AI driven information about pages, the Swantide App's AI workflow search experience for those customers who did not purchase the AI chat, and notifications. I encouraged the CEO that we take my designs and validate them with customers if this type of embedded chat would be valuable prior to launch. If customers would find the chat more useful when in Salesforce, then it would be a lot of work for the engineering team to also build the solution embedded into Swantide.
Winning Best App for Everyday Use at Salesforce Ben AI Demo Jam 2024
Unfortunately, my time at Swantide came to a close before learning what direction the team took. They did; however, showcase the embedded chat at the Salesforce Ben AI Demo Jam 2024 where they won by majority vote Best App for Everyday Use.
Learnings when Developing with AI
Need Time to Validate the Reliability of AI Specific Products
Swantide’s AI specific products, RevOps Search Experience and RevOps AI Assistant, required additional work around validating the reliability of our generative AI solutions. We ran frequent tests with our Customer Success team and selected friendly customers to ensure the accuracy of our AI solutions. We were aiming to release the product when it was close to 80% accurate but the reality was that it was closer to 65% accurate. Training the models and identifying edge cases (measure and learn) became part of the job. The hope was that if we released the product even at 65% reliability, it would continue to learn and improve.
Participants tested our large language models, feeding each model specific questions or search queries. We then requested that they evaluate the answers the models provided, giving each response a rating and a reason for that rating. The quality and reliability of responses were evaluated against the types of inputs the participants provided. Once launched, our Customer Success team kept a log of incorrect answers from Search and the plan was to do the same for AI chat. The product and engineering team would then review these incorrect instances and improve them.
Balancing AI Innovation and Product Value Execution
At Swantide, we had a clear company directive to pursue AI—specifically the development of an AI Assistant. While this was an exciting direction, I often questioned whether it was the most effective way to serve our customers in the near term.
Auto-documentation proved to be incredibly helpful, while debugging introduced significant time and financial costs. Accuracy was critical—not just for usability but also for maintaining customer trust and preserving our brand. As a small startup, we were constantly balancing the desire to launch AI-supported features with the need to prioritize features we already knew how to build well.
One example was Sandbox, a highly requested feature from our customers. In fact, it was arguably more valuable to them at that time than the AI Assistant. However, the majority of our development resources were directed toward shipping a rough version of the AI Assistant to signal progress, rather than fully investing in Sandbox. The result was that both solutions—AI Assistant and Sandbox—had delayed launches.
Customer expectations are rising—not only do they want AI, but they want it to be accurate, reliable, and fully integrated. In my future AI roles, I plan to thoughtfully evaluate when and how AI capabilities are introduced. The goal should be to balance innovation with execution, and to invest in reducing the time-to-accuracy. Ultimately, shipping is no longer the hard part. Testing, verifying, and refining for accuracy when it matters is where most of the effort lies today—and that’s where successful product teams will differentiate themselves.