All conversations read, not a small sample
Manual QA reaches a few percent of what your team handles. Core is built to evaluate every finished conversation it receives.
Emberfield is the support QA service we are building. It will read the finished conversations you hand over, check every reply, and show you what to fix first.
Sign-ups for early access are not open yet.
Nothing is collected on this page. Core has not opened, and we promise no date.
One made-up conversation. Change the reply, or use a ready-made one, and see which demo rules pass, with the words behind each result.
I want to cancel before it renews and get this month back.
Try this: use the reply that passes, then delete the sentence about this month’s money. Watch Answered the question flip to fail, with the customer’s own words marked.
These are a few plain rules that run in this page, so you can feel how it works. They are not Emberfield’s checks, and this is not their score. Nothing you type leaves your browser.
Manual QA reaches a few percent of what your team handles. Core is built to evaluate every finished conversation it receives.
We will write and run the checks. You will not design a rubric or label a single conversation.
Each result will open to the conversation and the exact words it rests on, so you can show your team which check failed.
Topics ranked by the checks they fail most, with risk flags one click from the conversation. Scores are for conversations, never people, so you coach from real examples.
See how each check moves over time, by topic. A weekly report with figures only, no conversation text, that can be forwarded as it is. An alert when a check starts failing more than usual.
Where conversation text is stored, how long it is kept and whether any of it trains an AI model: we state each here once it is in place. None is stated yet, so none is claimed.
Built to cover what a sample misses, with the evidence behind every score.
Quality checks on every reply, six risk flags on what the customer says. We will define and run them, so nobody on your team builds a rubric or labels a conversation first.
A result will open to the conversation, the exact words it rests on, and what stands behind the check. We will say so when your own reviewers have not checked it.
Two kinds of result, each with the words behind it. Quality checks pass or fail. Risk flags are found or not found.
Each conversation will get one quality score from its own checks, counted equally. Flags never change it. There is no score for your whole support and none for a person. Checks read the conversation: they cannot know whether an answer was true or fits your policy. Handoff quality applies only where an AI handed over to a person and the export marks the AI’s replies.
Topics ranked by the checks they fail most, what keeps repeating, and risk flags with the conversation one click behind each. A weekly report and alerts will cover the days you are not looking.
Helpdesks we plan to readIntercomFreshdeskZendesk
Your work email, and if you like, your helpdesk and how many conversations a month. Free, no card. We write once, when Core opens.
A helpdesk admin hands over an export of finished conversations. You will see what is read and kept before you agree.
Scores, topics and risk flags for the conversations you handed over, and a weekly report with figures only.
Emberfield will only read: nothing is installed in your helpdesk.
Support teams of three to thirty agents, with a head of support who wants to know what is really happening.
Under 500 conversations a month, or in a regulated industry such as health or banking? Emberfield is not built for you yet.
No price per seat, conversation or resolution. No set‑up fee.
These are planned prices, confirmed when Core opens. Nothing can be bought on this page.
For up to 2,000 conversations a month. Higher volumes are planned to cost more, in flat steps.
A resolution rate says the AI closed a conversation. It does not say the answer was right, which check failed, or which topic drags it down. Emberfield is built to read the same conversations from outside and report on them check by check.
Conversations answered by the helpdesk’s AI come through where the helpdesk exports them. On Zendesk, an export leaves out the tickets its AI agents handled.
You leave your work email and confirm it from the email we send. If you like, you also tell us your helpdesk and how many conversations you have a month. It costs nothing, asks for no card, and we promise no date. We write once, when Core opens.
No. We define the scorecards and run them. Using your own labels and criteria is an option we are planning, not something Core will ask of you.
Conversations your agents answered will be covered on each helpdesk we read. Whether AI conversations come through depends on what the helpdesk exports. On Zendesk, an export leaves out the tickets its AI agents handled.
No. Checks read what was said: was the question addressed, was it clear and polite, did the customer show a risk. They cannot know your policy or whether a fact was right, so a polite reply that breaks policy can still pass.
No. Emberfield only reads finished conversations you hand over. It does not write into your helpdesk, and nothing it does reaches your customers.
A helpdesk admin runs an export and hands it over, as a file or through a cloud bucket. On Zendesk the account owner first asks Zendesk Support to switch exports on, and Team plans have none.
No. A single score hides which check failed. You will get counts and a fail share for each check. A quality score exists on one conversation only.
No. An alert will say a check failed more than it did in your own four weeks before. It names no cause, because a rule can count but cannot know why.
No. Nothing is shown by person: no list, no filter, no email. Emberfield scores conversations, never people.
Not yet. Core has not opened, so there is nothing to try today apart from the demo on this page. Join early access and we will tell you when it opens.
Sign-ups for early access are not open yet.
Nothing is collected on this page. Core has not opened, and we promise no date.