Responsible use of assessments

1. Keep a person in the loop

Up&Up scores multiple choice, typing and game questions deterministically, and uses AI to suggest a score for open text and voice answers. AI-suggested scores are advisory. The review screen shows you the candidate's actual answer next to the suggested score, you can change any score by hand, and the submission is marked as reviewed once you save. Totals are always recomputed on the server from the per-question scores, so what you see is what was actually counted.

We ask that a person reads the answers before a rejection is sent, particularly where AI scoring or proctoring flags are involved. Nothing in Up&Up automatically rejects a candidate, and we do not recommend wiring it up to do so.

2. Tell candidates what they will be assessed on

Before a candidate starts, the platform shows them your organisation name, the test title and description, the estimated duration, and a notice listing any invigilation you have switched on. That is the floor, not the goal. Use the test description and welcome message to tell them:

• Which skills or attributes the assessment covers.
• How the results will be used, and at what stage of your process.
• Whether anything is timed, and whether they can pause or come back.
• Who to contact if something goes wrong or they need an adjustment.

A candidate who knows what is coming performs closer to their real ability. Surprise mostly measures nerve.

3. Offer adjustments before you are asked

Disabled candidates, neurodivergent candidates, candidates taking the assessment in a second language, and candidates with caring responsibilities may need something different. Say in your invitation that adjustments are available and give an address to ask. Do not require a diagnosis or medical evidence to grant a small change.

Practical things you can change in Up&Up:

• Time limits are set per question, and clearing the field leaves that question untimed. If a candidate needs longer, build a version of the assessment with the limits raised or removed and send them that link instead.
• Webcam recording, screen capture, fullscreen enforcement and ID verification are per-test settings and are off unless you turn them on. Turn them off for a candidate for whom being recorded is a barrier rather than a risk control.
• Typing tests measure keyboard speed. If keyboard speed is not part of the job, do not include one — and offer an alternative if a candidate cannot use a keyboard in the usual way.
• AI voice interviews require a microphone and comfortable spoken English. Offer a text equivalent or a human conversation where that is not a fair test of the role.

An adjustment that changes the conditions is not cheating. Scoring the adjusted attempt as though nothing changed is the mistake to avoid.

4. Personality profiles describe working style

Up&Up includes personality questionnaires built around the five-factor model commonly known as the Big Five — openness, conscientiousness, extraversion, agreeableness and emotional stability. The five-factor model is one of the most widely used frameworks in work and organisational psychology, and it is the model our items follow.

Being straight with you about what that means here:

• These items are not scored and never affect a percentage. Personality answers are excluded from the score total and stored as a separate trait profile. A test made up only of personality items produces no score at all, by design.
• Our specific questionnaire has not been independently validated. It has not been through a published norming study, and we make no claim about its predictive validity for any particular job. Treat the output as a structured prompt for conversation, not as a psychometric instrument.
• There is no pass mark, and you should not invent one. Do not filter, rank or reject on a trait score. Use it to decide what to explore in an interview: "you came out as someone who prefers settled routines — tell me about a time the plan changed on you."
• Self-report is self-report. Candidates can and do answer as they think you want. That is another reason not to build a gate on it.

5. Make the assessment the same for everybody

• Send the same assessment to every candidate for a role, at the same stage.
• Decide what a good score looks like before you look at any results, and write it down. A bar set after the fact is a bar set around a person.
• Assess things the job actually needs. Every extra question is a reason for a good candidate to drop out, and an extra opportunity to measure something irrelevant.
• Keep the assessment as short as it can be while still being informative. If you would not spend that long reading the results, do not ask a candidate to spend it answering.
• Review the questions themselves. AI-generated questions are drafts: read them for accuracy, for assumptions about background or culture, and for anything that would disadvantage a capable person who took a different route into the work.

6. Use proctoring proportionately

Recording someone's camera and screen while they work is intrusive. It is reasonable for a high-stakes, high-volume assessment where a cheated result would put the wrong person in a job. It is heavy-handed for a fifteen-minute screening exercise. The settings are per test, so choose deliberately rather than switching everything on by default.

Where you do use it: tell candidates in the invitation as well as on the start screen, explain what is recorded and who can see it, and review flags yourself before acting on them. Automated flags — a second face, a missing candidate, a tab switch — are signals, not proof. People take assessments in shared houses, in libraries, and with children in the room. Ask before you conclude.

7. Give something back

Candidates give you an hour of unpaid work. Tell them the outcome either way, within the timeframe you promised. Where you can, tell them something they can use — which areas were strong, which were not. You do not have to publish the answer key to be useful.

8. Your legal obligations are yours

For candidate assessment data, you are the data controller and Up&Up is the processor acting on your instructions. That means candidate requests — access, correction, deletion — come to you, and we support you in answering them. Keep assessment data only as long as you need it, and delete tests you are finished with.

Two things worth checking with your own advisers before you roll out any assessment tool, including this one:

• In the EU and UK, decisions based solely on automated processing that have a legal or similarly significant effect on someone carry specific obligations under data protection law. Up&Up is not designed to make hiring decisions on its own, and we ask you not to use it that way.
• Some jurisdictions regulate automated employment decision tools directly — New York City, for example, requires bias auditing and candidate notice for tools that substantially assist hiring decisions. Rules differ by location and change often; check what applies where you hire.

We have not commissioned an independent bias audit of our question generation or scoring, and we will say so rather than imply otherwise. If a bias audit is a requirement in your jurisdiction, factor that into how you use any assessment tool, including ours.

9. If we get this wrong, tell us

If you find a generated question that is unfair, a scoring behaviour that looks wrong, or a part of the product that pushes employers toward a bad decision, email hello@upandup.work. Candidates are welcome to write to us too, though for anything about a specific assessment you have taken, the employer who invited you holds the data and makes the decision.

Related: Security · Privacy Policy · Contact