InputBuffer collects feedback from your users and sorts it for you into buffers. A buffer is a named topic group, such as Bug Reports or Billing. Each buffer has routing instructions, plain-English rules that say what belongs in it, and an AI reads those rules to decide where each new piece of feedback goes.

This page has two parts. The getting started guide has three steps: create two buffers, test how the AI routes a piece of feedback between them, and write the routing instructions that settle it. The second part, Working with buffers, covers how the AI decides, how to review its choices, and how to change a buffer.

Getting started guide

Before you start

You need...

• An InputBuffer account and organization. An organization is your workspace in InputBuffer. It holds your feedback and your buffers. If you don't have one yet, the getting started guide walks through signing up for a free account and creating one.

Test buffers

1. 1

   Create two buffers

   Open your organization and click Buffers in the left sidebar.

   The first buffer we'll create is Bug Reports. If you created it in the getting started guide, skip ahead to the second buffer below. Otherwise click Add Buffer in the top right, and fill in the two fields:

   • Name: Bug Reports
   • Description: Things that are broken or not working as expected.

   Click Create Buffer. Here's the Buffers page with Bug Reports on it, before any feedback has been routed into it.

   Buffers

   Bug Reports

   Things that are broken or not working as expected.

   0 feedback items

   Now add the second buffer the same way:

   • Name: Billing
   • Description: Charges, invoices, refunds, and plans.

   Click Create Buffer. The Buffers page now lists both buffers.

   The description you typed does two jobs. It tells your team what the buffer is for, and when the buffer is created we copy it into the buffer's routing instructions. So each buffer already has one rule, and step 3 sharpens it.

2. 2

   Test the routing

   Say a Yoyodyne customer writes "I click the download button on my invoice and nothing happens." The feedback mentions an invoice, and it describes something broken, so it could fit either buffer, or both. A piece of feedback can be in more than one buffer at once.

   You can test that feedback without adding it to your organization. On the Buffers page, click Test Routing. A form titled Test Your Routing opens. Paste the customer's sentence into Description, then click Test Routing at the bottom of the form.

   The result appears under the form. Here's one result you might get:

   → Routed to: Bug Reports, Billing  65%
   We'd assign this and mark it Needs review.
   The feedback describes a broken download button on an invoice, which fits both buffers.

   The result has three lines. First, Routed to names every buffer the feedback would go into, with the AI's confidence in that list as a percentage. Next comes what we'd do with the feedback. In this result we'd put it in both buffers and mark it Needs review, which means the AI was only partly sure. A later section, How the AI handles each piece of feedback, covers the other outcomes. Finally, the last line is the AI's reasoning.

   So far each buffer's only rule is its description, and neither description says what to do with a broken billing page, so the AI may not be very accurate. Step 3 gives it a rule.

3. 3

   Write routing instructions

   Suppose you want a broken page to go to Bug Reports and stay out of Billing, even when the page is a billing page. You tell the AI that by rewriting each buffer's routing instructions, so that each one says what belongs in the buffer and what doesn't.

   On the Buffers page, click the pencil icon on the Bug Reports row. The Routing Instructions box holds the description from step 1. Replace it with:

   Route here only when the user reports a crash, an error, or something that doesn't work. This includes broken billing pages.

   Click Save. Then click the pencil icon on the Billing row, replace its rule with this one, and click Save again:

   Questions and complaints about charges, refunds, and plans. Not for a billing page that is broken.

   Click Test Routing and run the same test again. Routed to should now name Bug Reports and not Billing, and the reasoning should mention that something is broken. If the result still names Billing, reread the Billing rule and check that it says what the buffer excludes.

   If you'd rather see this feedback in both buffers, leave Not for a billing page that is broken out of the Billing rule. The AI can then put feedback like this in both buffers, and the test will usually name both.

Working with buffers

The rest of this page covers what you'll do once you're running buffers of your own.

Writing routing instructions that work

The two rules above work because each one uses "only when" or "not for" to say what its buffer excludes. Compare these two rules for Bug Reports:

• Route here only when the user reports a crash, an error, or something that doesn't work. This rule works, because feature requests and pricing questions are clearly outside it.
• Anything about problems. This rule doesn't work, because a complaint about a price is a problem too.

When you write your own, use the words your users write, such as "crash", "refund", or "can't log in", and keep each rule to one to three sentences.

The AI routes on a buffer's routing instructions alone. It never sees the description, and we tell it not to decide from the name. We copy the description into the routing instructions once, when you create the buffer, so if you later edit the Billing description to mention invoices, routing doesn't change. Make routing changes with the pencil icon.

A buffer you create with no description starts with empty routing instructions, and we tell the AI to put nothing in a buffer like that.

How the AI handles each piece of feedback

Every new piece of feedback goes through the same steps as the test in step 2. Let's use the invoice feedback again as an example. First the AI reads its title and description. Next it compares them against the routing instructions of each buffer. Finally it picks every buffer the feedback belongs in, and gives one confidence score for the whole list.

The confidence score decides what happens next:

• At 80% or higher, we assign the feedback to the buffers the AI picked. You don't have to do anything.
• Between 50% and 80%, we assign the feedback to those buffers and mark it Needs review, so you can check the AI's choice.
• Below 50%, the feedback stays out of every buffer. You can place it yourself, as the next section describes.

You can change both numbers in Settings → AI, under Auto-apply Threshold (%) and Review Threshold (%).

Reviewing feedback the AI wasn't sure about

Feedback marked Needs review collects on the Review page. Click Review in the left sidebar to see it. Here's the Review page with two pieces of feedback waiting.

Each row shows the feedback, the buffer or buffers the AI chose, its reasoning, and its confidence. What you do next depends on whether the AI was right:

• The AI chose the right buffer. Click Mark reviewed. The feedback stays where the AI put it and leaves the Review page.
• The AI chose the wrong buffer. Click the row to open the feedback, then use + Add in the Buffers section to put it in the right one. You can remove the AI's choice in the same section.

The same + Add control is how you place feedback that scored below 50% and has no buffer.

When you add a buffer the AI didn't choose, we record that as a correction. The AI includes recent corrections when it classifies new feedback, so it tends to put similar feedback in the buffer you chose.

If the same kind of feedback keeps landing on the Review page, add a sentence about it to that buffer's routing instructions.

Approving buffers the AI suggests

Some feedback doesn't fit any buffer you have. If Auto-create Buffers is turned on in Settings → AI, the AI can suggest a new buffer for that feedback. It makes a suggestion only when its confidence in the new buffer reaches the Buffer Suggestion Floor (%), which is 40% unless you change it.

Suggested buffers appear at the top of the Buffers page under an AI suggested heading, as cards with a dashed border. Each card shows the suggested name, the AI's reason for suggesting it, and the number of feedback items already in it. The same cards appear on the Review page.

Each card has two buttons:

• Approve turns the suggestion into a regular buffer, and it starts receiving new feedback.
• Delete rejects the suggestion. The feedback in it goes back to having no buffer, and the AI won't suggest that name again.

Rejecting a suggestion is not permanent. The Buffers page shows a line such as 1 rejected suggestion below the suggestions. Click it to expand the list, then click Restore next to a name to bring that suggestion back.

Changing a buffer

There are two places to change a buffer, and they change different things:

• To change the routing instructions, click the pencil icon on the buffer's row on the Buffers page.
• To change the name, description, or icon, click the buffer's name to open it, then click Edit. The routing instructions can be changed there too.

Opening a buffer also shows the feedback routed into it, newest first. Here's Bug Reports after it has been in use for a while, with its routing instructions under the name.

A change applies to new feedback right away. Feedback that is already in the buffer stays where it is.

Next steps

You now have two buffers with routing instructions, a way to test where a piece of feedback will go, and a Review page that catches whatever the AI wasn't sure about. Next, add the feedback widget to your site, so real feedback starts arriving.

• Feedback Widget: collect feedback from your web app or docs site, with no backend required
• API getting started: send feedback from your own backend, a support tool, or a script