What Was the Manual Process Before Automation?
The client, a growing fitness company, had a two-step manual workflow for generating client meal plans. A staff member would visit one website to retrieve macro data, then manually enter that data into a second website to generate a full meal plan complete with meals, directions, ingredients, and a grocery guide. That output then had to be imported into Adobe and formatted into a branded, customer-facing document. End to end, this process took up to an hour and a half per client.
With exponential growth over the prior year, the company's biggest constraint was manpower. Every new client added another 90-minute block of manual work, which made scaling extremely difficult.
What Was the Original Automation Plan, and Why Did It Change?
The initial approach was to connect both websites directly via API so the systems could communicate without any manual steps. The first website had full API access, but the second website's API was severely limited. It only allowed creating or deleting a company record, which was not sufficient to automate meal plan generation.
Rather than abandoning the project, the team pivoted to what they call a PDF builder solution. This approach still automates roughly 80% of the process, with one small manual step remaining on the client's end.
How Does the PDF Builder Solution Work?
The team built a custom landing page where the client can enter client details such as name, age, and macros and click to generate a meal plan. For the fully automated version, the system pulls macro data from the first website and feeds it forward automatically.
For the PDF builder flow, the client logs into the second website, generates the week-by-week meal plan PDFs (up to 13 weeks), and uploads those PDFs to the custom landing page. From there, the process is fully automated. The uploaded PDFs are sent to the Google Gemini API, which analyzes each document, extracts the structured meal data (including items that span across pages), and formats everything needed to generate the final branded output.
Once the client clicks generate, they can step away entirely. The finished document downloads automatically in their browser when it is ready.
How Well Does AI Handle Unstructured PDF Data?
One of the more impressive aspects of this implementation was how well Gemini handled messy, unstructured content. Meal data that bled from one page onto the next was correctly identified as belonging to the previous entry and was included in the right place. This kind of contextual reading of unstructured documents is what made the PDF builder approach viable as a substitute for a direct API call.
What Time Savings Did the Automation Deliver?
The previous manual process took up to 90 minutes per client. With the new system, a full 13-week meal plan generates in approximately 10 to 13 minutes, and that window includes the time the client spends opening and reviewing the finished PDF. A shorter plan covering just 4 weeks runs in roughly 4 to 5 minutes.
That represents an 8x to 9x reduction in time per client, and the staff member is not actively involved during that window. They click generate and move on to other work.
What Data Privacy Guardrails Were Put in Place?
Because Gemini is a public model accessed via an API key, the team was careful about what data gets passed to it. Only meal content such as ingredients and preparation steps was included in the PDF uploads. No customer-facing or personally identifiable information was fed into the model. This is a critical point: confidential or restricted data should never be submitted to a public AI model, even when access is controlled through an API key.
How Can This Automation Be Extended?
The generated meal plan PDF could be automatically stored in the company's platform or CRM so staff never need to manage downloads manually. The completion of a meal plan generation could also trigger downstream automations such as CRM updates, client notifications, or billing workflows. Each automation built this way becomes a reusable component that can be connected to future workflows without starting from scratch.