Tines Academy › Lesson
Get the data ready
Prepare the feedback for the dashboard, with clear rules for what to count, what to set aside, and how to handle repeated copies.
Objective: Define rules for preparing the data and check that every response is accounted for, using both totals and examples.
Know what’s in the data
Ticket-system data can have gaps or inconsistencies. A ticket might have no product area, more than one, or a label Maya’s team doesn’t use. The same response could also be loaded twice. Unless you decide how to handle these cases, the dashboard could count a response more than once or include it in the wrong group.
Customer comments need care too. Leaving out name and email fields doesn’t remove personal details someone typed into a comment. Keep that in mind when deciding who can open the dashboard in Lesson 3.2.
Start by defining what each field should contain. The table below shows the rules for Maya’s test data, which has one response per ticket. For your own data, use the values and counting rules applicable to your data.
Field
What it means
What to check
Ticket ID
The ticket the response belongs to
Count each ticket once. Flag records with the same ticket ID but different ratings for review.
Rated
When the customer submitted the rating
A date and time within the period loaded, including the current week so far.
Rating
How the customer rated the support they received
“Positive,” “neutral,” or “negative.”
Product area
The product area recorded on the ticket
Exactly one of “Billing,” “Account access,” “Shipping,” or “Mobile app.”
Channel
Where the support conversation took place
One of the channels Maya’s team uses.
Comment
What the customer wrote
Text or blank; comments may contain personal details.
Decide what happens to each problem
The diagram below shows how preparation accounts for every loaded response and which responses feed the dashboard’s charts. The loaded data stays unchanged.
Find the problems first
Choose the tab that matches your setup, then send the prompt in workflow chat. With a connector, fill in your team’s accepted values or provide them when Tines 3B asks.
Inspect the loaded feedback without changing it. Use the rule we agreed for identifying and counting each response.
Check for repeated copies or conflicting records, missing IDs, invalid rating dates or dates outside the loaded period, and values that don't match these rules:
- Rating: one of [your ratings]
- Product area: exactly one of [your product areas]
- Channel: one of [your channels]
- Comment: text or blank, if comments are included
Ask me for any accepted values I haven't provided. Don't assume that every value in the loaded data is correct.
For each problem, show how many loaded rows it affects and one original example. Use a real ticket or response link where available. Don't invent links.
If a row has more than one problem, make that clear so I don't mistake the problem counts for a total number of rows. Flag uncertain cases for me to confirm rather than guessing what they mean.
Inspect the loaded test feedback without changing it. Each fictional ticket should have one response.
Check for repeated copies or conflicting records, missing ticket IDs, invalid rating dates or dates outside the loaded period, and values that don't match these rules:
- Rating: "positive", "neutral", or "negative"
- Product area: exactly one of "Billing", "Account access", "Shipping", or "Mobile app"
- Channel: "email", "chat", or "phone"
- Comment: text or blank
For each problem, show how many loaded rows it affects and one original test response. Don't invent ticket links.
If a row has more than one problem, make that clear so I don't mistake the problem counts for a total number of rows. Flag uncertain cases for me to confirm rather than guessing what they mean.
Review what the check found
Read the inspection report in workflow chat.
Check a few examples
Open an example of each reported problem.
Compare it with the original ticket or response. For test data, use the saved original shown in the report.
Does the field have the problem described?
Check unfamiliar values
If you don’t recognize a value, ask whoever owns the data what it means. Don’t change the rules just to make a warning disappear.
If you can’t confirm it yet, leave it for review.
If something doesn’t match
Show Tines 3B the reported problem and the original example. Ask it to explain the difference, then check the updated report.
Prepare the data
Leaving out responses can change the chart’s numbers. The summary below the chart will show what was left out and why.
Send the prompt in workflow chat. Include any corrections you’ve confirmed. Leave anything you’re unsure about set aside for review.
Prepare the feedback using the field rules and the way we agreed to identify each response. Keep the loaded data unchanged and save the prepared version separately.
Give every loaded row exactly one outcome: counted, set aside with a reason, or removed as a repeated copy.
- Apply only corrections I've explicitly confirmed. For example, "Login" can become "Account access" only after confirmation.
- Keep one copy of each response. Set conflicting records aside for review rather than choosing a rating. In the one-response-per-ticket setup, this includes different ratings for the same ticket.
- Set aside unresolved problems from the inspection, including missing identifiers, invalid or out-of-range rating dates, invalid ratings or channels, and missing, multiple, or unknown product areas.
Use only counted responses in the charts. The percentage is negative responses divided by all counted responses with a valid rating, for the same period and filters.
Below the weekly chart, add a compact data-quality summary showing how many rows were counted, set aside, or removed as copies, and why. Keep it less prominent than the chart. These outcomes must account for every loaded row once, even if it has several problems.
Check that the loaded data stayed unchanged and every loaded row was counted, set aside, or removed as a repeated copy. Show any rows with no outcome or more than one.
Check that these totals add up to the number of loaded rows. For the last full week, show the negative count, total counted responses with a valid rating, and percentage.
Show a counted response, a set-aside response and its reason, and a correction or removed copy if there is one. Compare them with the originals and the agreed rules.
Report what you tested, the results, and anything you couldn't check. Keep the rest of the build unchanged.
Check that it adds up
Read the summary and test results in workflow chat.
Check what happened to each row
Find the check showing that the loaded data stayed unchanged.
Each loaded row should have been counted, set aside, or removed as a repeated copy. Ask Tines 3B about any row with no outcome or more than one.
Add up the totals
In Browser, find the data-quality summary below the chart.
The totals should all match.
Compare a few examples
Check a few examples to see whether the decisions were right.
Example
What to look for
A counted response
It meets the rules you agreed.
A set-aside response
The reason explains why it was left out.
A correction, if there is one
It’s a change you confirmed.
A removed copy, if there is one
It’s another copy of the same response, not a separate rating.
Check one week’s percentage
Use the last full week. Check the totals against the counts in the test report. It should include positive and neutral responses too. It shouldn’t include responses set aside or copies removed.
If there are no counted responses with a valid rating, the percentage should be unavailable, not 0%.
If something doesn’t match
Show Tines 3B the example or number that looks wrong and explain what you expected. Ask for a fix, then repeat the affected check.
Ask it to record unfinished checks in Readme.
What to do if something still feels fuzzy
You defined the preparation rules and asked Tines 3B to apply them. You checked that each response was counted, set aside with a reason, or removed as a copy. Next, you’ll use the prepared data to build the rest of Maya’s dashboard.
Build it safely: You checked a few prepared responses against the original data and the preparation rules, rather than trusting the totals alone.