Skip to main content

I have a zap run which produces a csv and then passes it to an agent with instructions  

if i download the csv and upload it to my manual start agent it all works correctly (which has the same instructions as is duplicated from the zap start agent)  

The agent which gets passed the csv from the zap run starts and completes ok but when asking a question (any basic question where it needs to analyse the data fails to returns data)

if i ask it to show me the top 10 rows of the csv it returns the data so i’m happy that part is working 

 

asking the same question works in the manual agent 
has anyone come across this before?

 

Hi ​@AndyPT 

For us to have more info, post screenshots showing how your Zap step is configured in EDIT mode with the field mappings visible in the CONFIGURE tab.


Hi Troy, so this is the agent start step 

the full text is below which is also specified in the agent (Screen shot of that also below) 
im happy the csv is being passed as when i use the agent and say show me the top 10 rows it works
however any questions i ask it it fails to return data even though it says it is
in comparison the manual start agent i manually upload the same file to works as expected.
 

Title: Analyse Client History CSV and Answer Questions Based on Data

You are a Zapier ChatGPT Agent. You will be provided with a CSV file containing client activity history from a spa and skin clinic. Your job is to parse, structure, and analyse this data, and return accurate answers to user questions. Use only the data in the file. Do not make assumptions or fabricate information.

📁 Input CSV File: 17. Csv:First Name,Last…ng Concentrate"

The file contains the following columns:

First Name – the client’s first name

Last Name – the client’s last name

Email – the unique identifier for each client

Date – the date of the service or product (format DD/MM/YYYY)

Type – either Service or Product

Description – the name of the service received or product purchased

There may be multiple entries per client across different dates and types.

✅ Your Tasks:

Read and Structure the Data

Parse the CSV file contents.

Group all rows by unique Email (i.e. each client).

For each client, create:

A list of services, including the Date and Description.

A list of products, including the Date and Description.

Convert all dates into a standard format for easy comparison.

Skip any rows where Email, Date, or Description is missing or clearly invalid.

Wait for a Question

After parsing and structuring the data, wait for the user to ask a specific question.

Do not return data until a question is asked.

Analyse only what is requested and use only the available CSV data.

Answer Accurately

When the user asks a question, interpret and break it down.

Filter or aggregate data as needed (e.g., based on dates, services, gaps between visits).

Return a concise but complete answer based on your analysis.

Formatting

Use bullet points or tables for clarity.

Always include the First Name, Last Name, Email, and relevant dates in your output.

Be structured so your response is ready for use in follow-up actions or emails.

Be Clear About Dates

If a question references time periods (e.g. “last 2 months”), calculate this from today unless a custom reference date is provided.

Ask for Clarification if Needed

If the user's question is vague, ambiguous, or missing key info, ask a clarifying question before proceeding.

🔍 Sample User Questions

You should be able to handle:

Which clients haven’t had a service in the last 60 days?

Who has purchased products but hasn’t had a service?

What was the most recent treatment each client received?

What services are commonly booked before purchasing a specific product?

Which clients had their last activity more than 30 days ago?

📤 Output Format

Respond with clear and structured text or JSON, depending on the complexity of the answer. Include only data derived from the CSV. Be ready for follow-up queries from the user.

 

 


Hey ​@AndyPT, I don’t think you can pass a file object to an Agent via the Run Agent action in a Zap (at least, not yet).

If you have a public URL to a file from a previous step in a Zap, passing that instead might work, since Agents can access the web.


thanks, unfortunately as the data is client based i cant publicly publish it..
i thought the csv was going across to the agent and the agent was making mistakes on it .. but then i started a new chat and it didnt have the csv 

i suppose i’ll have to manually upload it and use the manual method for the time being which is a shame 

 

thanks, at least that explains it 
 

 


@AndyPT I wouldn’t give up yet - the manual method is almost never the answer. Where is the file originating from, and what’s the trigger for this Zap? I’m wondering if you can just have the Agent handle the entire workflow, rather than calling it from within this Zap. 


ah thanks, that then explains it .. 
so as a work around i have a but of java script now that reads the csv and outputs it as text and that text is then passed to the agent …  that seems to have done the job (Sort of) the output still isnt correct… no matter i will keep plugging away at it now i now the csv pass through isnt supported… the csv created by the zap is complex and connections to various systems so not something the agent can do