Front Line
AI Customer Support Agent
Resolves customer questions across chat and email using your knowledge base and hands off complex cases.
Extensions
PII Redaction Layer
Names, contact details and other personal data are masked before your AI model or your logs ever see them.
Customers share names, emails, phone numbers and account details in conversations, and most of that never needs to reach an AI model or sit in a log. This add-on masks the personal data types you agree before they leave your product for the model or land in your logs. A test report shows exactly what still reached the model.
- Masks personal data before it reaches the AI model
- Keeps names and contact details out of your logs
- Proves the result with a test report
$2,500
10–20 working days
What it is
PII Redaction Layer is an add-on that strips personal data out of the text an Upmark Labs product sends to its AI model and writes to its logs. You agree which data types count as personal, and they are masked on the way through. A test report shows what still reached the model.
Why businesses buy it
Every piece of personal data sent to a model provider or kept in logs adds privacy risk and paperwork. Many businesses hold back from AI, or lose larger customers, because they cannot answer where personal data goes. A redaction layer reduces what leaves your product in the first place. The written scope agrees the data types, applies masking to model calls and logs, and ends with a test report showing what reached the model, so you can answer reviewers with evidence.
Problems this solves
Personal details reach the AI model and its logs unfiltered
The PII Redaction Layer strips personal data from the text the product sends to its AI model. Data is masked on the way through, before it reaches the model or the logs.
Teams disagree on which details count as personal data
You agree which data types count as personal, such as names, emails or account numbers. The updated documentation records that list so everyone works from the same definition.
You cannot show what personal data still gets through
A test report shows what still reached the model during testing. You get a clear picture rather than an assumption, and can decide whether to adjust the rules.
A good fit if
Customers type phone numbers, emails or account details into your agent, and you would rather the model provider never received them.
Your logs and transcripts are read by staff who do not need to see personal details.
Your security or privacy reviewer has asked what personal data leaves your systems.
You already run an Upmark Labs product, or are adding this to an order now.
Not the right fit if
You want to watch for sensitive data leaving across many tools, not one product. AI Data Leakage Monitor covers that.
You need stored records removed after a set period. Data Retention and Deletion Configuration handles retention and deletion.
You need a record of who changed what and when. Audit Logging Add-on adds that trail.
You want policies and controls across every AI system you run. AI Governance System is the wider build.
How it works
- Personal data types are listed. Together we agree what to mask for your product, such as names, email addresses, phone numbers or account references.
- The redaction step is placed in the path. It sits between the incoming text and the model call, and again before anything is written to logs.
- Masked values stay usable. Where the product needs to act on a detail, such as booking against a phone number, we agree how that is handled at intake.
- Redaction is tested with the parent product. We run sample conversations containing each data type and record what reached the model and the logs.
- You review the test report, request one round of changes, and receive a written walkthrough and updated documentation.
What you provide within 72 hours
Your portal lists these the moment you order. Delivery starts the working day after everything is in and your access is checked.
- The order reference of the product being extended
- A list of the personal data types you want masked, and any you must keep visible
- Sample messages or transcripts that show where personal data appears today
- Admin access to the product's model account and log storage, shared securely as agreed at intake
- One person who signs off what counts as personal data
The 72-hour rule. Send everything on your checklist, including access, within 72 hours of ordering so your build starts on time.
Timeline
Start
Data types and sample messages collected
Week 1
Redaction rules agreed and mapped to the data flow
Week 2
Masking added before model calls and log writes
Week 2
Sample conversations run, test report written
Week 2
Your review and one revision round
Go-live
Redaction live, walkthrough recorded
| Option | Delivery time | Price |
|---|---|---|
| Standard | 10–20 working days | $2,500 |
What you receive at handover
- Redaction of agreed personal data types before text reaches AI models or logs
- The agreed list of data types treated as personal
- A test report showing what still reached the model
- Updated documentation, a handoff note and a short written walkthrough
Expected benefits and how to measure them
- Helps reduce the personal data held in model requests and logs
- Makes it easier to answer privacy questions with evidence
- Can support your own data protection reviews
- Personal data found in logs
- Sample log entries before launch and monthly after, and count any unmasked personal details.
- Redaction misses
- Run the agreed test inputs each quarter and record any data types that slipped through.
- Privacy queries answered
- Note how long it takes to answer a customer privacy question, before and after.
Before you start: Sample current logs and model requests and note which personal details appear. Keep that sample for comparison.
These are things to track, not promised results. Outcomes depend on your data, your team and your market.
Running costs
Running costs stay as they are for the parent product. The redaction step runs inside the same setup, and any model or hosting usage it adds is billed to your own accounts, not through us.
Ask about running costsCustom requirements
Redacting files, images or audio, or masking data inside systems the product does not touch, sits outside this add-on. Describe it in the custom requirements note at checkout and you get a written Change Quote. Larger or bespoke projects go through the Custom AI request form instead.
Request a custom versionKeep it running
Changes to the redaction rules, such as a new data type, are covered by the parent product's maintenance, which grows in line with the add-on's price. Adding markets with very different privacy rules may need a Change Quote.
Adds $125/mo to maintenance
Covered by the main product's maintenance, which grows by this amount.
How maintenance worksHow this compares
| Do it yourself with a SaaS tool | Hire a custom agency | Upmark Labs | |
|---|---|---|---|
| Price | A privacy tool usually bills monthly per seat or per volume | An agency quotes after a data review | Ours is one fixed price, shown before you buy |
| Where it applies | A generic filter, set up by you | Depends on the brief | Ours sits in your product's own model calls and logs |
| Proof | A settings screen | Set per project | Ours ends with a test report showing what reached the model |
Questions about this product
Which personal data does the PII Redaction Layer mask?
The data types you agree at intake. Common choices are names, email addresses, phone numbers, postal addresses and account or order references. You can also mark data that must stay visible because the product needs it. The agreed list is written into the documentation, so your reviewer can see exactly what is covered.
Does redaction stop the agent doing its job?
No, when it is planned properly. Some details are needed to complete a task, such as a phone number for a callback. We agree how each one is handled during intake, and the sample conversations in testing confirm the parent product still completes its work with redaction switched on.
Is personal data removed from my existing logs?
No. The layer masks data from the day it goes live onward. Cleaning historical logs or transcripts is not part of this add-on. If old records need removing, Data Retention and Deletion Configuration sets deletion rules, or you can request a Change Quote for a one-off clean-up.
How do I know redaction is working?
The test report shows it. We send sample conversations containing each agreed data type through the parent product and record what arrived at the model and what was written to the logs. You review that report before launch and can ask for one round of changes.
Can the PII Redaction Layer be added to a product we already run?
Yes. It can be bought alongside a new Upmark Labs product or added to one you already run. Masking applies from the day it goes live. You choose the delivery speed at checkout, and each speed has a fixed price and date shown before you pay. Bring a list of the personal data types you care about to intake.
Does this make my product compliant with privacy law?
No add-on can promise that on its own. The layer reduces how much personal data reaches the model and the logs, and documents what it masks. Whether that meets your obligations is a judgment for your own privacy or legal adviser, who can use the documentation in their review.
Keep personal details out of prompts and logs.
Add the PII Redaction Layer and the data types you choose are masked before they reach the model or the logs, with a test report to prove it.
Works well with
AI Features & Bots
AI Website Chatbot
Adds a branded AI assistant to your website trained on your content.
Call Intelligence
AI Call Summarization
Summarizes calls into notes, outcomes and next steps in your CRM.