Transforming Wellness with AI
Wellhub
Innovative companies have embraced technology to redefine the meaning and the medium of wellness. Wellhub, a global corporate wellness platform, stands at the forefront of this transformation. Founded to revolutionize workplace health and fitness, the company connects over 18,000 businesses with top-tier wellness resources across 11 countries.
Trainiac, a key division of Wellhub, specializes in creating personalized fitness experiences by matching companies and individuals with expert fitness professionals. The Trainiac platform goes beyond traditional fitness solutions, leveraging technology to create meaningful connections between trainers and clients. By focusing on personalization and data-driven wellness, Trainiac has carved out a unique niche in the corporate wellness market, addressing the growing demand for tailored, accessible fitness solutions in workplace environments.
While Wellhub and Trainiac innovate in the wellness services they offer their clients, they also innovate in how they build and operate those services. With AI agent automation, they are reimagining how AI can transform the complex operational backend workflows that are integral to their business.
In the span of just 6 months and zero software development, Trainiac automated two critical human-centric workflows using AI agents implemented on the Thunk.AI platform. |
Trainiac runs a two-sided digital marketplace service, connecting talented fitness professionals with consumers who seek fitness instruction. The service relies on two critical workflows that lie at the intersection of human judgment-intensive work and repetitive, manual work:
Certifying and onboarding fitness professionals
Create compelling multilingual content about fitness for consumers of the service
Both workflows have so far been managed and executed by trained employees, making them expensive to maintain and scale. Partnering with Thunk.AI, Wellhub has developed AI agent applications, known as ‘thunks,’ to automate these workflows. Now these fully deployed automated workflows significantly improve response times, lower costs, and increase the quality of the onboarding process.
80% decrease in content generation and translation expenses
15%-20% reduction in processing time for new trainer onboarding
Greater scale and higher quality with constant team size
Workflow Process #1: Trainer Onboarding
Trainiac’s process for bringing new trainers onto the platform was labor-intensive and time-consuming.
Personal trainers submitted detailed application forms with their credentials, professional experience, gallery photos, sample training plans and exercise demonstration materials for assessment.
The partnership team then manually reviewed every application:
Verifying certifications and qualifications
Checking their understanding of the Trainiac platform by assessing their training plans
Assessing trainer expertise
Verifying the quality of their submitted photos to ensure they met quality, safety, and focus standards
Finally, crafting detailed acceptance, follow-up or rejection emails to the trainers.
In some cases, application re-submissions from trainers had to be re-evaluated
The team processed approximately 10 applications daily, with each review demanding significant time and attention to detail.
The Wellhub team used the Thunk.AI platform to automate the process.
They created a thunk to handle trainer onboarding submissions.
They used English to define a set of submission fields and the logic to check those fields.
They described an aggregated pass/fail check across all the AI-computed results and how to construct and communicate the result back to the trainer
They connected existing submission forms from multiple global markets to the thunk via a simple webhook interface
All of this was done rapidly, taking advantage of the self-service capabilities and no-code design of the Thunk.AI platform. Every form submission from a trainer is now run through this AI agent automated process, while the Trainiac team simply checks and approves before the final communication back to the trainer.
When a trainer submits an application via the form, it is sent to the thunk where it is immediately processed by the AI agent. The AI agent handles a diverse range of inputs, processing, transforming and understanding the content. This is one of the very useful capabilities of the platform. It can handle a wide range of content formats: text-based certifications, document URLs, image screenshots or audio/video files. It can also handle a variety of different input languages. In all cases, the AI agent transforms the input into a structured, consistent representation.
Next, the AI agent reviews the submitted trainer form against the pre-defined acceptable criteria. The AI agent doesn't just check boxes — it intelligently evaluates each submission component, much like a human reviewer would, and provides its assessment along with its rationale. For instance, the trainers have to submit a small gallery of photos and these n$eed reviews. The thunk designer specifies the guideline for review as follows:
Trainer gallery photos must have: appropriate clothing, minimized background clutter, no photos with children or minors and a preference for action or exercise-related images |
That’s it! Notice how the guidelines are exactly how they would be specified to a human employee who would be expected to exercise judgment and subjectivity in making decisions. Yet, that is sufficient for the AI agent to process the photos submitted and decide if the guidelines are being met.
After all of the form submission properties have been individually checked, the AI agent also generates a final comprehensive assessment for every trainer. Now, a team member reviews the AI’s work, makes necessary adjustments, and finalizes the assessment.
The final workflow stage involves intelligent communication management. The AI agent, again supported by human-provided communication templates, drafts personalized communications in the same language as trainer submissions. It shares the assessment along with precise guidelines for next steps, providing an excellent draft for the human reviewer to assess and email.
This automated AI thunk dramatically reduces manual processing time, while it also ensures consistent, high-quality review of all trainers worldwide. Human error is minimized. The team no longer has to manually download, read, and process inputs of various formats and content. This allows team members to focus on strategic initiatives in engaging and growing their trainer base.
As for the trainers, they now enjoy the benefits of being onboarded rapidly and generating revenue faster.
Workflow Process #2: Content creation
Equally important to Trainiac is the process of crafting content about various exercises to share with end-users of the service. It is traditionally labor-intensive, time-consuming, and expensive—requiring significant human effort. Because the service is offered in many countries, the content also needs to be translated into 6 different languages.
Crafting each exercise description used to take approximately 20 minutes from a fitness expert, and required clear explanations of proper exercise form, key training tips, and the associated benefits. Beyond creation, translations into multiple languages added further time and incurred external service costs, with multilingual support demanding significant resources to ensure accuracy and brand consistency across diverse audiences.
The Wellhub team used the Thunk.AI platform to automate the process.
They created a thunk to create content for exercises.
They used English to define a simple workflow (describe the exercise, describe three tips to explain how to do the exercise safely, and three benefits to your body, and translate into different languages), along with a few sentences of descriptive logic to guide each of the steps.
Once again, using the self-service capabilities and no-code design of the Thunk.AI platform, the thunk was built rapidly and modified easily.
One critical aspect of the workflow definition is to describe the technical accuracy required for fitness instruction, optimize the content for specific platform constraints such as length of content, and reflect the experience level of fitness instructors on the platform.
The Thunk.AI platform provides features to define constraints, policies or requirements that must be met as the work is done. For example, the designer of a thunk can describe constraints on the output of a workflow step, such as “generated description must be under 2 minutes long when recorded”. This guidance is used by the AI agent while both doing the work and while checking the work. Once the work is complete, there are also affordances in the user interface to make it easy for human reviewers to examine the quality of work.
After the initial content generation, the thunk generated precise translations of the content across six key languages: English, Spanish (two different dialects), Portuguese, German, and Italian. This isn't just word-for-word translation, but a sophisticated rewriting that considers cultural nuances, local fitness terminology, and platform-specific communication styles to preserve Wellhub's unique brand voice.
Now that the thunk is live in production, the Trainiac team simply submits new exercise names regularly and lets the thunk automate the work. By reimagining content creation using AI agents, Wellhub has transformed a potential operational bottleneck into a productivity gain and a strategic advantage for its Trainiac product team.
Outcomes and Impact of AI Agent Automation
The implementation of these two workflows in production have delivered significant operational improvements for Wellhub.
With the trainer onboarding workflow, AI automation has provided tangible and measurable benefits:
|
Akshay Ahooja, Head of Business Operations and Partnerships at Wellhub, captured the additional intangible benefits perfectly:
"The quality of the Trainiac service rests largely in the quality of the direct training that coaches deliver. These are professionals in the wellness field with varying degrees of experience and expertise. The ability to scale these partners significantly, with the same team in place, while not just maintaining but improving quality is a big challenge. Thunk.AI met that moment of need by helping us automate the onboarding workflow so the team can focus on more strategic areas."
With the content generation workflow, AI automation has provided significant and measurable productivity gains:
|
While there are different ways to implement automation with AI agents, the choice of platform makes a difference. The Thunk.AI advantage derives from two fundamental platform capabilities – self-service design and the “CHARM” framework for enterprise scenarios.
The value of the self-service no-code design of the platform is obvious. All logic in a thunk is expressed in natural language. This enables a high-fidelity and cost-efficient translation from the requirements of the workflow owner in an organization to the AI agent executing the workflow in a thunk.
The “CHARM” framework describes five key dimensions in which an AI agent automation platform has to deliver capabilities in order for AI automation to go beyond a cool prototype and actually be deployed at an enterprise like Wellhub. In all five of these dimensions, the Thunk.AI platform provides the necessary capabilities for enterprise adoption.
Compliance:
Thunk.AI has simple and effective natural language mechanisms to define policies that govern the content and style of all generated outputs. Likewise, there are mechanisms to define policies to check if submitted content is “acceptable”. The runtime AI agents verify and enforce compliance with these policies.
Human-in-the-loop collaboration:
The platform is fundamentally designed for human and AI collaboration as they drive together towards desired outcomes. This makes it easy for workflow owners and team members to maintain critical human oversight of any work done in a thunk, by an AI or by a team member. The platform records all activities by AI agents or human users, and thus provides a clear audit trail and a transparent decision-making process. In addition, human reviewers are able to provide corrective feedback of the AI agent work, creating learning examples to guide future processing.
Automation:
The platform is designed for seamless integration of AI and human expertise. Automation is not an all-or-nothing choice. Workflow owners can choose how much AI agents should automate or should they wait for human consent to move work forward. For example, in Wellhub’s trainer submission thunk, a human provides instructions to review trainers and then the AI agent is in the driver’s seat when reviewing elements of an application. Once this is done, the human takes over to review the AI’s assessment and make a final judgement on the trainer’s acceptance to the platform. Then, they switch seats again, with the AI drafting appropriate communications before the human reviews and sends the final emails to the trainer.
Reliability:
Reliable and consistent behavior is essential for an automated system in production. Thunk.AI achieves this through a combination of mechanisms. At the lowest level, all responses from the AI models are constrained to follow structured expected response formats, and this is checked and validated. All partial progress is recorded into structured data state, which is again verified for validity. Finally, there are multiple mechanisms for the workflow owner to define constraints that specify expected behavior and outcomes. This combination of clear instructions, consistent execution of the instructions, and rigorous verification of outputs distinguishes the Thunk.AI platform and allows Wellhub’s team to trust the automated work done by the AI agents.
Modularity:
As with software programs, AI agent automation “programs” also suffer from the downsides of large monolithic definitions — lack of reuse, difficulty in comprehension, inadequate quality control. On the other hand, when logic is composed from smaller reusable modules that can be built, tested, shared, composed, and reused, the overall solutions become more reliable and less brittle. Unlike most other AI automation platforms, Thunk.AI encourages design of modular AI components that can be composed to form larger end-to-end automations. In the Wellhub thunk that evaluates trainer samples, there are many smaller modules of logic that define how each trainer submission field should be processed, and then there are other logic modules that define how they should be composed. This step-by-step modularity of workflow logic plays an important role in the overall stability of the automation in production.
By reimagining its workflow processes with AI, Wellhub has demonstrated how AI agents can transform operational challenges into competitive advantages. As Wellhub continues to integrate AI agents into its other operations, the future of wellness platforms looks increasingly intelligent and efficient.
Beyond the wellness industry, this case study reveals how AI agent automation is starting to drive productivity and transform a broad range of operational processes.
WellHub
To learn more about the various wellness services offered by Wellhub::
info@wellhub.com
Thunk.AI
To learn more about the Thunk.AI platform:
info@thunkai.com