They work real cases, with the records behind them and your own policy base. Haven records the route to every answer: what they searched, what they read, what they decided to ignore, and whether the reasoning holds against your bar.
An interview hears what they would do.
A take-home shows you their best day.
A reference confirms they were there.
None of it is the work.
There is no new process to run. You send one link, they sit down once, and the evidence arrives attached to the answer.
A real inbound message, the account behind it, the order history, and a searchable policy base. You set the case count and the clock. One attempt, and it keeps running through a refresh while answers save as they go.
Every search, every article opened, and how long each one was held. The policy a candidate relied on is read off what they did, not off a box they ticked.
Score it against a framework built from your own standard, or disagree with it. Either way the reasoning is on the page, and it is still there a year later when someone asks why.
A customer who is owed an answer, the record that decides it, and a policy base nobody has pointed them at. Nothing on this screen is labelled.
Priya Raman · 16:42 · Order 20918
The driver came at eleven and nobody was in, so he left it round the side by the gate. I have only just got back. The ice packs are completely liquid and the chicken is room temperature.
I have got both kids' dinners planned around this box for the week. Can I still cook it tonight, or am I binning the lot?
The console already knows. The article held longest, past a real reading threshold, is the one they relied on. A three second glance is navigation, not reliance, and it does not count.
Timestamps are relative to the case opening.
A self-report collapses three different candidates into one wrong answer. The record keeps them apart.
Searched the wrong terms, or stopped searching. A training problem, and a cheap one to fix.
Opened the plausible neighbouring article and settled there. They stopped reading one step early.
Held the governing article, then answered against it. That is the one you most need to know about.
The trail explains a score. It does not set one. A candidate who never opened the policy base and still got the case right has shown judgement, and is scored on the reply.
Every article is tagged. A RULE has one correct outcome. A GUIDE invites judgement, and a defensible judgement is the correct answer. Applying a rule where the article asks you to think, or improvising where a rule binds, is the same mistake in two directions.
The brief never explains the difference. Recognising which one you are holding is the thing being measured.
The box is already packed and on a pallet. The cut-off is a fact about the kitchen, not a policy anyone gets to bend, and there is one correct answer here.
Stop, escalate the same day, and never assess the risk yourself. A candidate who reassures the customer has failed the case no matter how warmly they wrote it.
There is a starting point and a ceiling, and the space between them is the candidate's. Any figure inside it with reasoning that holds is correct. Freezing is not.
Nothing here has one right answer. What the reply has to protect is the next thirty weeks, not this one, and the article says so without saying how.
| What you need to know | Interviewstructured | Take-homeuntimed | Referencespast roles | Aptitude testoff the shelf | Haventhe job, under a clock |
|---|---|---|---|---|---|
| Every candidate gets the same task | |||||
| Judgement under a real clock | |||||
| What they consulted to decide | |||||
| Evidence rather than self-report | |||||
| Holds up when the decision is questioned |
A structured interview is a good instrument, and Haven does not replace it. It just cannot see the twenty minutes where someone finds the governing policy and answers against it anyway.
Harvard Business Review puts the cost of replacing a service agent between $10,000 and $20,000, once hiring, onboarding and ramp are counted. The figure is not the painful part. Every new hire resets the clock, and one who looked right in the room resets it twice.
The candidate who speaks fluently about putting the customer first is not always the one who finds the governing policy with nineteen minutes left on the clock.
It surfaces later as a refund that should not have been issued, an escalation that should have stopped two replies ago, and a customer who quietly stops ordering.
That cuts both ways. Rejecting a strong candidate on a feeling costs you the hire; keeping a weak one on a feeling costs you the year.
Pick a standard role and set it up yourself, or tell us the role and we build the cases, the policy base and the scoring framework with you. Either way you keep the record.
Start your first assessment