AI Transparency & Disclosure

Last updated: February 2026

1. Our Commitment to AI Transparency

Rates Goblin is built with transparency at its core. We believe you should always know when AI is processing your data, what it does with that data, and what its limitations are.

This page explains exactly how artificial intelligence works within our platform. It is written in plain language so that anyone — not just technical specialists — can understand what is happening behind the scenes.

Our approach to AI transparency has been informed by the RICS Global Standard on Responsible Use of AI in Surveying and the CIOB Code of Practice for Project Management principles. These standards emphasise accountability, human oversight, and clear disclosure whenever AI is used in professional practice.

2. AI Models We Use

The table below lists every AI model used in Rates Goblin, what it does, and who provides it. We will update this table whenever we add, remove, or change an AI model.

Feature	AI Model	Provider	Purpose
Quote Extraction	GPT-4o	OpenAI	Reads uploaded PDF/Excel files and extracts line items (description, qty, unit, rate, amount)
Rate Grouping	GPT-4o	OpenAI	Groups similar line items across multiple quotes for comparison
NRM Classification	GPT-4o	OpenAI	Classifies items against RICS New Rules of Measurement element codes
Column Mapping	GPT-4o	OpenAI	Auto-detects column structure in uploaded cost plan spreadsheets
AI Query Assistant	GPT-4o	OpenAI	Answers natural language questions about your quotes and cost data
Support Chat	GPT-4o Mini	OpenAI	Answers questions about how to use Rates Goblin
Scanned PDF Fallback	GPT-4o Vision	OpenAI	Reads scanned/image-based PDFs when text extraction fails

3. What Data Is Sent to AI

We want you to know exactly what information leaves our servers and is sent to an AI provider for processing. Here is a clear breakdown.

3.1 Data That IS Sent

Extracted text from your uploaded quotes and cost plans
Line item descriptions, quantities, rates, and amounts
Your natural language questions (when using the AI Query Assistant)
Cost plan section descriptions (when using column mapping or classification features)

3.2 Data That Is NOT Sent

Your email address
Your password
Personal identity information
Device fingerprints
Your IP address
Session data or authentication tokens
Other users' data — your data is isolated and never mixed with another user's data

3.3 How Data Is Transmitted

All data is sent via encrypted HTTPS connections. Data is processed by OpenAI's API and, in accordance with OpenAI's API data usage policy (as of 2024), is not used for model training. See Section 9 for more detail on OpenAI's data processing practices.

4. Consent & Control

You are always in control of whether AI processes your data. Here is how consent works in Rates Goblin:

Explicit consent required: You must give explicit consent before any AI feature processes your data. No data is sent to any AI provider until you actively opt in.
Clear consent dialog: Consent is requested via a clear modal dialog that explains what data will be processed, which AI model will be used, and what the output will be.
Withdraw at any time: You can withdraw your AI consent at any time by navigating to Settings > AI. Withdrawing consent immediately stops all AI processing — no further data will be sent to any AI provider.
Use Rates Goblin without AI: You can use Rates Goblin without AI features. Manual data entry, benchmarking, and export all work without AI consent.
Auditable consent records: AI consent is tracked with timestamps and version numbers, so there is a clear record of when consent was given or withdrawn. This supports audit compliance.

5. Confidence Scoring & Human Oversight

AI is not infallible. That is why every AI output in Rates Goblin includes a confidence score and is subject to human review before it is finalised.

5.1 Confidence Scores

Every AI-generated output includes a confidence score expressed as a percentage from 0% to 100%. This score indicates how confident the AI model is in its output.

Green (85%+) — High confidence. The AI is confident in its output, but you should still review it.
Amber (70–84%) — Review recommended. The AI's output may need correction.
Red (below 70%) — Manual check needed. The AI's output should be treated as a starting point only and verified by a qualified professional.

5.2 NRM Classification Threshold

NRM classifications that fall below 85% confidence are automatically flagged as "pending suggestion" and require manual approval by a qualified quantity surveyor before they are applied.

5.3 AI Disclosure Badges

All AI-generated content is visually marked with purple "AI" disclosure badges throughout the interface. This ensures you always know which data was machine-generated and which was entered manually.

5.4 QS Sign-Off Workflow

The QS sign-off workflow ensures that a named professional reviews and accepts all AI outputs before they are finalised. AI outputs remain in a draft state until a qualified professional has reviewed and approved them.

6. AI Accuracy & Limitations

We want to be honest about what AI can and cannot do. Here are the known limitations of AI within Rates Goblin:

Extraction accuracy: Extraction accuracy is typically 80%+ on well-structured documents. This means that approximately 1 in 5 items may need manual correction, even in ideal conditions.
Document quality matters: Accuracy depends heavily on document quality. Poorly scanned PDFs, handwritten notes, unusual formatting, or non-standard layouts may significantly reduce accuracy.
NRM misclassification: The AI may misclassify NRM elements, especially for unusual, specialist, or highly specific items that do not clearly map to standard NRM categories.
Rate grouping errors: Rate grouping may incorrectly merge items that appear similar but are distinct, or split items that should be grouped together.
AI Query Assistant limitations: The AI Query Assistant derives all figures from your uploaded data and published benchmarks — it does not fabricate numbers. However, it may misinterpret the context of your question or return figures from the wrong part of your data.
Professional review is essential: AI outputs should always be reviewed by a qualified professional before any reliance is placed upon them. AI is a tool to assist professionals, not a replacement for professional judgement.

6A. Community Rate Benchmarking Database

Rates Goblin uses contributed rate data to build and improve an industry-wide Community Rate Benchmarking Database. This section explains how AI interacts with the benchmarking pool.

6A.1 How AI Uses Benchmarking Data

When you use Rates Goblin's benchmarking features (e.g. comparing your rates against industry averages), the AI may reference aggregated data from the Community Rate Benchmarking Database to provide more accurate and up-to-date comparisons. This data is:

Anonymised: No personal data, company names, or project-specific details are present in the benchmarking pool.
Aggregated: Individual rates are combined into statistical ranges (e.g. low/median/high) — the AI never exposes another user's specific rates.
Voluntarily contributed: Only users who have given separate, explicit benchmarking consent contribute data to the pool.

6A.2 AI Does Not Train on Your Data

The Community Rate Benchmarking Database is not used to train AI models. Benchmarking data is used as a reference dataset that the AI queries to provide rate comparisons and insights. The AI models themselves (GPT-4o) remain unchanged — only the reference data they compare against is enriched by user contributions.

6A.3 Separate Consent

Contributing to the benchmarking database requires separate, explicit consent that is independent from your AI processing consent. You can use AI features without contributing to benchmarking, and you can contribute to benchmarking without using AI features. Full details are in our Privacy Policy (Section 3A) and Terms & Conditions (Section 6A).

7. Audit Trail & Compliance

Rates Goblin maintains a detailed audit trail of all AI operations to support transparency, accountability, and regulatory compliance.

Operation logging: Every AI operation is logged with the timestamp, the AI model used, the confidence score returned, and the user who triggered the operation.
AI Disclosure Appendix: An AI Disclosure Appendix is available for every quote and cost plan. This appendix lists all AI operations performed on that document, providing a complete record for professional review.
Edit history: Edit history tracks all changes made to AI-generated data, including before/after values, who made the change, and when it was made.
RICS compliance support: This audit trail supports compliance with RICS requirements for AI transparency in surveying, enabling professionals to demonstrate that appropriate oversight was applied.
Data preservation: Audit data is preserved even if user accounts are deleted, ensuring that the compliance record remains intact.

8. RICS & CIOB Context

Rates Goblin has been designed with reference to established professional standards for the use of AI in construction and surveying practice.

The platform's approach to AI transparency, confidence scoring, and human oversight has been informed by the RICS Global Standard on Responsible Use of AI in Surveying.
It follows CIOB Code of Practice for Project Management principles of transparency, accountability, and record-keeping.

Important Disclaimer

Rates Goblin is not endorsed by, affiliated with, or certified by RICS, CIOB, or any professional body. References to these standards are descriptive only — indicating that the platform's design has been informed by publicly available professional guidance.

Users who are RICS or CIOB members remain bound by their own professional obligations and should satisfy themselves that their use of Rates Goblin complies with the requirements of their professional body.

9. OpenAI Data Processing

All AI processing in Rates Goblin is performed by OpenAI. Here is what you need to know about how OpenAI handles your data:

OpenAI processes data under their API Terms of Service.
As of March 2023, OpenAI does not use data submitted through the API to train their models. Your commercial data remains confidential and is not used to improve OpenAI's AI models.
OpenAI may retain API inputs and outputs for up to 30 days for abuse monitoring purposes, after which they are deleted.
OpenAI is a US-based company. Data transfers to OpenAI are protected by appropriate legal safeguards, including the UK Addendum to the EU Standard Contractual Clauses.
For full details, see OpenAI's privacy policy at openai.com/policies/privacy-policy.

10. Changes & Updates

We will update this page whenever our AI usage changes. This includes adding or removing AI models, changing what data is sent to AI providers, or modifying how confidence scoring and human oversight work.

Where we make material changes to our AI usage, we will notify all registered users by email before the changes take effect. The "Last updated" date at the top of this page will also be revised.

We encourage you to review this page periodically to stay informed about how AI is used in Rates Goblin.

11. Questions

If you have any questions about how AI is used in Rates Goblin, or if you would like further detail on any of the information provided on this page, please contact us at hello@ratesgoblin.ai.

We are committed to answering your questions openly and honestly.