Customer Feedback Loop: Google Forms to Sheets to AI Analysis to Automated Actions

Google Forms & Sheets

ChatGPT API

Make.com

Gmail / WhatsApp

Why most trades businesses are flying blind on feedback

Most engineers leave the job assuming it went well. The truth often sits in the customer's head and never comes out

Most trades businesses know whether a job went well by feel. The engineer comes back, says it was fine, and the office moves on to the next one. If the customer was unhappy, you usually find out one of three ways. They ring up and complain. They leave a bad Google review. Or they just never call again.

All three of those happen too late. By the time a complaint phone call lands, the customer has already told their partner, their neighbour, and probably a WhatsApp group. By the time a one-star review goes up, it is on the internet for the next decade. By the time they have gone quiet, you have lost a repeat customer worth thousands over the years and you do not even know it.

The fix is not complicated. You ask. You ask quickly, you ask in a way that takes them 30 seconds, and you do something useful with the answer. Done properly, this becomes a feedback loop: the customer tells you something, the system reads it, and either the owner gets pinged about a problem or the customer gets sent a Google review link. Either way, you act on it within the hour.

The good news is the tools to build this loop are already in your hands. Google Forms is free. Google Sheets is free. The Make.com free tier covers 1,000 operations a month, which is enough for a small business doing 200 jobs. The only paid bit is the OpenAI API call, and even that comes in at pence per response. The whole stack runs for under £25 a month for most outfits.

What you are actually building

The Make.com scenario builder shows the whole flow visually. Drag, drop, connect, done

The architecture has four layers. Layer one is the collection point: a Google Form that takes the customer 30 seconds to fill in. Layer two is the storage: a Google Sheet linked to the form that captures every response in a row. Layer three is the analysis: a Make.com scenario that picks up new rows, sends each response through the ChatGPT API, and writes back a sentiment score plus a category tag. Layer four is the action: the same scenario reads the sentiment and decides what to do, whether that is alerting you, sending a review link, or just logging it.

The reason this works for trades, where most software is overkill, is that you are not buying a survey platform. You are gluing free tools together with a small amount of automation. Once it is set up, it runs itself. The customer fills in the form, the row appears, the AI reads it, the action fires. You do nothing unless something needs your attention.

I built the first version of this at Help me Fix to pick up on patterns we were missing in tenant repair feedback. The technical bit took an afternoon. The business value showed up in the first week when we caught a complaint about a contractor that would have otherwise become an angry email to the housing association three days later. That is the whole point. Speed.

This pairs neatly with anything else you already have running. If you have read the heat pump monitoring integration guide or the AI lead response setup, you already have most of the pieces. The OpenAI account, the Make.com account, the Google Workspace login. You are just adding a new scenario.

What you need before you start

Before you build anything, sort out the accounts. None of this works if you skip the credentials step and try to wing it later. Here is the checklist.

You need a Google account, ideally a business Google Workspace one but a free Gmail account works fine for the form and sheet. You need a Make.com account, free tier is enough to start. You need an OpenAI API account with a small amount of prepaid billing on it, £5 is plenty for the first month. You need somewhere to send alerts, which is either Gmail, WhatsApp via the Make.com WhatsApp module, or Slack. And you need a way to trigger the survey, either a Zapier or Make trigger from your job management software or a manual SMS link.

The OpenAI prepaid billing trips people up. You sign up at platform.openai.com, add a card, top up by £5 or £10, and that gets you API access. This is separate from a ChatGPT Plus subscription. You do not need ChatGPT Plus to use the API. The API charges per token, and at 2026 pricing GPT-5.4 sits at $2.50 per million input tokens and $15 per million output tokens. A typical sentiment analysis prompt is under 500 tokens. You will spend pennies, not pounds.

Step 1: Build the Google Form

Keep the form to four questions or fewer. Anything longer and response rates collapse

Go to forms.google.com and create a new blank form. Name it something the customer will recognise, like "How was your visit from us today?" Keep it short. Four questions is the upper limit, three is better.

Question one is the NPS question: "How likely are you to recommend us to a friend or colleague?" Use a linear scale from 0 to 10. This is the only quantitative question you need.

Question two is the open feedback: "What is the one thing we could have done better?" Long text answer. This is the gold. The AI will analyse this.

Question three is optional but useful: "Anything you want us to know?" Long text again. Some customers will leave question two blank but write a paragraph here.

Question four, if you want it, is the job reference number, prefilled from a URL parameter if you can manage it. This lets you tie responses back to specific jobs, engineers, and dates.

Settings matter. Turn off "Collect email addresses" unless you actually need it. Turn off "Limit to 1 response", because some customers might want to come back later. Turn off the progress bar. Set the confirmation message to a plain thank-you that does not ask for anything else. The whole experience needs to feel like 30 seconds, not five minutes.

Step 2: Connect the Sheet

In the Google Form, click the Responses tab, then click the green Sheets icon. This creates a linked Google Sheet that auto-populates every time someone submits the form. Name the sheet something obvious like "Customer feedback responses".

Every form submission becomes a new row in the sheet automatically. No connector to build, it just works

The default sheet has one column per form question, plus a timestamp column. Add three more columns to the right of the existing data. Call them "Sentiment", "Category" and "Action taken". These will be populated by the Make.com scenario later. Leave them empty for now.

One thing to lock down at this stage is the column order. Once Make.com is wired into specific columns, do not reshuffle them or your scenario will start writing data to the wrong place. If you need to add new questions to the form later, add them to the end of the question list, not the middle.

Set up a second tab in the sheet called "Dashboard". This is where you will build a simple summary view, with rolling NPS, response volume, and a count of negative responses by category. You can build this with basic Sheets formulas, no scripts needed. AVERAGE for NPS, COUNTIF for the rest.

Step 3: Add the AI analysis layer

This is the bit that makes the loop intelligent. Without AI, you have a glorified survey. With AI, you have an analyst reading every response within seconds and tagging it for you.

The Make.com scenario reads new sheet rows, sends each one to OpenAI, and writes the result back automatically

Log into Make.com and create a new scenario. Add a Google Sheets module as the trigger, set to "Watch new rows" on the Form Responses sheet. Set it to run every 15 minutes. You could go more frequent but 15 minutes is fine for feedback and keeps your operations count down.

Add an OpenAI module next. Use the "Create a Completion" action, pointing at the GPT-5.4 or GPT-5.1-mini model depending on budget. The prompt is the important bit. Here is the structure I use.

System prompt: "You are a customer feedback analyst for a UK trades business. Read the feedback below. Return a JSON object with three fields: sentiment (one of Positive, Neutral, Negative), category (one of Quality, Punctuality, Communication, Pricing, Cleanliness, Attitude, Other), and summary (one short sentence summarising the feedback). Output JSON only, no other text."

User prompt: Pass the NPS score, the "one thing we could have done better" answer, and the "anything else" answer, concatenated with line breaks. Tell the model the NPS scale runs 0 to 10.

After the OpenAI module, add a JSON parser to break the response into its fields. Then add a Google Sheets "Update row" module that writes the sentiment, category, and summary back to the same row the trigger picked up. Now every new feedback response gets analysed and tagged automatically within 15 minutes of being submitted.

Step 4: Wire up the automated actions

This is where the loop closes. The AI has classified the response. Now the scenario needs to decide what to do based on the classification.

Add a Router module after the JSON parser. A router lets you split the workflow down different paths based on conditions. Set up three paths.

Path one is the negative path. Filter for sentiment equals Negative OR NPS score less than 7. When this triggers, send a WhatsApp message or email to the owner or operations manager. The message should include the customer's name, the job reference, the category, and the AI summary. Make it actionable: "Negative feedback from [customer] on job [reference]. Category: [category]. Summary: [summary]. Call them within the hour."

The negative feedback alert arrives on your phone within 15 minutes of the customer pressing submit

Path two is the positive path. Filter for sentiment equals Positive AND NPS score 9 or 10. When this triggers, send the customer an email with a direct link to leave a Google review. This is the moment they are most likely to follow through. The follow-up is short: thank them, link to the review page, and stop. Do not ask them to do five other things.

Path three is the neutral path. Everything else. No alert, no action, just log the row. These are the responses that do not need a human in the loop. They are useful in aggregate, not individually.

For the trigger that sends the survey out in the first place, you have two options. If you use a job management platform with webhooks, like ServiceM8, BigChange, Commusoft or Tradify, you can fire a Make.com scenario when a job is marked complete that sends an SMS via Twilio or WhatsApp with the survey link. The same scenario can prefill the job reference field. The BigChange and Make.com integration guide walks through the webhook setup for that platform specifically. If you have not got that setup yet, a manual trigger works fine: an admin clicks a button in Sheets, the scenario sends the survey. Less automated but a starting point.

What this costs per month

Pricing transparency, since most integration guides hand-wave at this. Here are the real numbers for a small fleet doing 200 jobs a month, with a 25 percent survey response rate, giving you 50 feedback responses to process.

If you scale up to a larger fleet running 1,000 jobs a month, you might need the Make.com Pro plan at £18.82 a month to handle the volume, and the OpenAI bill goes up to around £8 or £10. The SMS bill scales with you, sitting around £40 to £50 a month at that level. The total still comes in well under £100, which is a fraction of what a feedback platform like Trustpilot Business charges you to do less.

Testing and rollout

Run the scenario manually first with a handful of test responses before letting it loose on real customers

Do not roll this out to all customers on day one. Run it manually with five or ten test responses first, ideally using real feedback you have collected before. Submit the form yourself with both positive and negative example responses, watch the scenario run, and confirm the AI is classifying correctly.

Pay particular attention to edge cases. What does the AI do when the feedback is sarcastic ("yeah, brilliant, only three hours late")? What about when the comment is ambiguous ("it was alright I suppose")? Tweak the system prompt until the AI's classifications match what you would call them yourself. This takes 30 minutes of iteration and pays back for years.

Once you trust the classifications, enable the actions path by path. Start with just the negative alert. Let it run for two weeks. Then enable the positive review request. Let it run for another two weeks. By the end of the month, you have a battle-tested loop that handles itself.

Once the system is running, build the habit of looking at the dashboard tab in the sheet every Monday morning. Five minutes a week is enough. You are looking for trends, not individual responses. A category that suddenly spikes, an engineer whose jobs are tracking lower than the team average, a pattern in the times of day where complaints cluster. The AI does the heavy lifting on individual responses. Your job is to spot the patterns above them.

What tradespeople are saying

Real discussion from the community and from Make.com and Zapier user forums where people are building exactly this kind of automation.

Recommended videos

The Ultimate Guide to Sentiment Analysis using the Open AI API

Analytics with Adam

How to use GPT4 with Google Forms

Prompt Circle AI

ChatGPT Sentiment Analysis in Google Sheets

Sheets Ninja

Using ChatGPT for classification and sentiment analysis in Google Sheets

walzr

How to use ChatGPT for Sentiment Analysis (Google Sheets)

AlmaBetter

ChatGPT Inside Google Sheets: Support Replies Based on Sentiment

Coefficient

Frequently asked questions

My verdict