inyourroots® Fairness, Inclusion & AI Statement

Some things about you should have nothing to do with the work you are shown. We maintain a formal classification of the profile information held by the service and whether it is permitted for matching.

Blocked from matching — 13 attributes, including:

• Your age range
• Where you live
• Your pronouns
• Whether you have told us you have a disability
• Your accessibility notes
• What support helps you work at your best
• Your work situation and the kinds of experience you have had

Allowed into matching — 12 attributes, all about how you like to work. These include your superpower, what excites you, whether you prefer improving or creating, whether you are drawn to individuals or systems, the problems you like, your appetite for technology, and the subjects and industries you are curious about.

The privacy gateway is designed to read that classification, use only permitted information and record blocked attributes by name. Its coverage across all live calculation routes and its audit behaviour remain to be confirmed. That is a design control rather than a guarantee about outcomes — section 5 explains the difference and what we still cannot show.

The matching calculation is one thing. A conversation is another, and we would rather be clear about the difference.

Budi may be given your name, age range and general area so the conversation can feel relevant and mention things near you.

If you tell us that something makes a type of work difficult, Budi may take that into account in conversation. For example, if you describe a mobility barrier, Budi may discuss related roles involving coordination, planning, analysis or communication rather than repeatedly leading with frontline physical roles. Budi is instructed not to make disability the headline or assume a limit you have not described.

That means disability or accessibility information may shape the order and framing of a conversation. It does not enter the arithmetic described in section 1, it is not automatically shared with an employer, and it does not remove a role from your results.

If you would rather Budi did not use that information conversationally, tell Budi or leave that part of your profile empty. This does not remove your access to the rest of the service.

Budi currently runs on Google Gemini 2.5 Flash through Google Cloud Vertex AI.

• Budi can be wrong, including confidently wrong. Check anything important and talk it through with someone you trust before making a significant decision.
• General-purpose AI models reflect biases in their training material. We can reduce and monitor those effects, but we cannot promise to remove them completely.
• Safety filters are enabled. Harassment, hate speech, sexually explicit content and dangerous content are filtered using Google’s configured medium-and-above safety threshold.
• Budi uses inyourroots® data and tools. When Budi discusses a stored strength, role or progress result, the service is designed to retrieve it from our maintained data rather than rely only on the model’s memory.

This conversational part of Budi has fewer guarantees than the deterministic matching calculation. We think it matters to say that plainly rather than let the word “AI” cover both.

We cannot yet show how results differ between groups. We can explain how the system is designed. We cannot yet produce evidence that it operates fairly across different groups because we have not completed that analysis. Before collecting any equality information for outcome monitoring, we will assess whether it is necessary and proportionate and seek appropriate data-protection advice.

Our strengths map was written by people. The links between strengths, industries and roles reflect human judgement. Consistency is not the same as fairness: a map can apply the same assumption to everyone and still contain a poor assumption. We have not yet completed and recorded a formal fairness review of that mapping.

Allowed signals may still act as proxies for blocked information. Removing postcode from matching does not remove every effect of background if permitted answers correlate with it. We have not fully solved that risk.

Audit verification remains outstanding. The service is designed to record the information used, attributes withheld, scoring version and time. Current audit entries and coverage across calculation routes need verification; we do not infer the present production state from an earlier testing snapshot.

Your match suggestions are a starting point, not a verdict.

If you think your suggestions do not reflect you — because something feels off, a result seems unlikely or you want to understand why a particular industry or role came up — you can ask us to take a look.

Email support@inyourroots.com and tell us:

• What you were expecting or hoping to see
• What came up that surprised you or felt wrong
• Anything about how you answered that you think might have affected it

You do not need to disclose health information or another sensitive detail to ask for a review.

We will review your results manually, explain how they were produced and, where appropriate, help you adjust your answers or flag a problem with the underlying data. We will consider your point of view and whether the result or underlying process needs correction; review is not simply a restatement of the automated output.

The acknowledgement and response aims in section 13 apply. Where your request exercises a statutory data-protection right, the applicable legal deadline also applies; our internal aims do not replace it.

We have never used anyone’s conversations to train or fine-tune an AI model. No training export has been run.

The technical ability currently exists. The system contains a filtered conversation view and an admin-only export function created for a possible future in which we might train our own model.

Our position for the pilot: we will not use identifiable conversations, recordings, reflections or uploads to train or fine-tune a model. If this position ever changes, it will require a separate assessment, a clear notice and an appropriate explicit opt-in. It will not happen quietly.

We are reviewing whether to remove the unused export function and view entirely.

Sometimes an industry you are drawn to may have very few current openings near you. Budi may say so and suggest a related path with more activity.

Our design intention is to make that visible whenever it happens, so you can see that you are being pointed sideways and understand why. Until that labelling is verified across the live product, treat it as an intention rather than a guarantee. You can ignore the suggestion. Current vacancy numbers describe the market now, not the limit of your future.

We are building Budi so that people can understand what an activity requires before starting, including reading, typing, timing, camera and microphone expectations.

Our intended standards are:

• Camera use is optional, with audio treated equally.
• Choosing not to show your face does not lower a match score.
• Timed practice counts up as guidance rather than cutting someone off.
• Retakes are unlimited and are not presented as a score or failure count.
• People can pause and return unless there is a genuine reason an activity must be completed in one sitting.

These standards are being checked against the production experience. We will not describe an individual behaviour as live until it has been verified.

An optional part of the journey may ask about focus, sensory needs, how you process information and what helps you work at your best.

• It is optional and can be skipped.
• Skipping does not restrict access to the rest of Budi.
• The answers are blocked from the arithmetic that generates matches.
• They are not automatically included in anything an employer sees.
• Because the answers may reveal disability or health information, we treat them as special category data and will use an appropriate separate consent process before this feature is live.
• Where consent applies, you can withdraw it by emailing support@inyourroots.com.

The final consent wording, Article 9 condition and deletion process will be aligned with our Privacy Policy before this feature is launched.

Employers see only information you have chosen to share through the controls available to you. They do not see internal match scores and must not use inyourroots® as an automated filter on people.

We have considered internal concepts such as filtering people by whether a named strength appears in their highest-ranked results. That type of exclusionary filter will not launch unless its purpose, lawful basis, equality impact, human oversight, user explanation and challenge route have been assessed. We are also considering inclusive, user-controlled evidence views instead.

We aim to meet WCAG 2.2 Level AA. Where we fall short, we want to know.

If you need information or support in a different format — such as larger text, simpler wording or another way to access something — please email support@inyourroots.com and we will do what we reasonably can. See our Accessibility Statement for more detail.

Tell us at support@inyourroots.com.

We aim to:

1. Log the report within 1 working day.
2. Acknowledge it and assign someone within 2 working days.
3. Explain what we found and what we did within 10 working days.

If an investigation needs longer, we will explain why and provide an updated timeframe.

You can also complain to the Information Commissioner’s Office at ico.org.uk.

1. We use permitted strengths and work-preference signals — not personal background attributes — to calculate matches.
2. We distinguish between what is built, what is intended and what we can prove.
3. We will publish meaningful fairness-review findings once we have evidence to report, including identified problems and actions.
4. We will not use your conversations to train AI without an appropriate assessment, clear notice and explicit opt-in.
5. We aim to fix problems faster than we defend them.

Version history