Request a quote
Emergency SectorsCareers About us Blog Get in touch
Close-up of two monitors showing log lines and coloured terminal output

What is DLP, and why data classification comes first

What is DLP? Data loss prevention is software that spots sensitive data about to leave your organisation, and then logs it, warns the user or blocks the action. It only works once you have decided what is sensitive. That is why a DLP project starts with data classification, not with rules: four labels, and an agreement on who applies them.

7 min read Written by the ITproposal team
Short answer

DLP is built for mistakes and carelessness: the email to the wrong recipient, the sharing link open to anyone. It does not stop an attacker who is already inside.

Without data classification, DLP only recognises numbers such as a National Insurance number or a sort code. Four labels, a default label and automatic labels per folder let it recognise a proposal or a contract too.

Start by logging, then warn, and only then block, and only for the top label. Block on day one and within a month the rule will be switched off.

What is DLP, and what it does not stop

DLP looks at data in motion: an attachment emailed outside the organisation, a file made public through a sharing link, a folder copied to a USB stick, text pasted into a web form. If the software recognises something sensitive, it takes the action you set in advance: log only, warn the user, or stop the action.

There are three places this happens, and a product rarely covers all three equally well:

  • On the device (endpoint): copying to USB, printing, uploading through the browser, pasting into another app.
  • In email and cloud: Exchange, OneDrive, SharePoint and Teams, and with some products other cloud services too.
  • On the network: traffic leaving through the proxy or firewall. That is the oldest form and the weakest for home working, because the traffic never passes the office.

What DLP does not do: stop an attacker who is already in. Someone with stolen admin rights switches the rules off or takes a route nobody watches. DLP is built for mistakes and carelessness, and it is good at that. Against intrusion you need other measures: multi-factor authentication, permissions that follow the role, and someone watching.

Data classification first: four labels are enough

DLP has to know what it is looking for. It can match on a pattern, a string that looks like a sort code and account number or a National Insurance number, or on a label: a marker attached to the document itself that travels with it when it is copied or emailed. Patterns catch part of it. Labels let a rule work for a proposal, a contract or a design, which carry no recognisable number.

Data classification means agreeing those labels. The temptation is to create seven. Do not: a colleague who has to choose between “confidential” and “internal confidential” on every document picks whichever comes first. Four works:

  • Public: can go outside without harm. The website, a brochure, a published press release.
  • Internal: the default for almost everything. Not a disaster if it leaks, but awkward.
  • Confidential: causes harm outside its circle. Customer data, HR files, proposals, contracts, financial figures before publication.
  • Highly confidential: a small, named set. Acquisition plans, medical data, special category personal data, the password vault. This is where DLP may block.

Describe each label with examples from your own organisation rather than a definition. “Anything that shows a colleague’s salary” reads faster than “data with a high confidentiality level”. ISO/IEC 27001 asks in Annex A for classification of information (5.12) and labelling of it (5.13), and lists data leakage prevention separately (8.12). If you supply UK government, its Government Security Classifications are a separate scheme you follow for that work; your own four labels sit alongside it.

Who applies the label

A classification nobody applies is a document in a folder. There are three ways to get labels onto files, and you need all three.

  • A default label. Every new document gets “Internal” unless someone chooses otherwise. That covers most of the estate at once and costs the user nothing.
  • The user, at the moment it matters. Mandatory labelling on every file is too much. Asking for a short reason when someone lowers a label is exactly enough.
  • Automatically, by location or content. A folder, a Teams channel or a SharePoint site gets a label and everything in it inherits it. On top of that, the software can suggest or set a label on a pattern: a document full of account numbers or National Insurance numbers becomes “Confidential”.

Decide who may lower a label. Anyone may raise one. Lowering from “Highly confidential” is for an owner, and it is logged. That is where the leak nobody intends sits: the file that briefly went to “Internal” because otherwise it would not go to the accountant.

In Microsoft 365 these are called sensitivity labels and they live, together with the DLP policies, in Microsoft Purview. Which features you have, such as automatic labelling or DLP on the device itself, depends on your licence. Check that before you draw up a plan that relies on a feature you do not have.

Rolling out a data loss prevention policy: log, warn, then block

The quickest way to make DLP fail is to block on day one. In the first week purchasing, payroll and the accountant all get stuck, and within a month the rule is off. Run it in three steps.

  • Log only. A few weeks in test mode: the rules run, nobody notices, and you see how often they fire, for whom and on which files. This is when you discover that a supplier receives a list full of account numbers every week and that this is simply the job.
  • Warn with a way out. The user sees a policy tip (“this message contains confidential data and is going to an external address”) and may continue with a short justification. Read those justifications back: they tell you where the rule is too broad, or where the process is wrong.
  • Block for a narrow set. Only once the number of alerts has fallen to something a person can follow, and only for the top label or a few clear patterns. For everything else, warning remains the end point, and that is not a shortcoming.

Start with two or three rules, not with the templates a product ships with. Those are written for every country and sector at once, and in your organisation they fire on things that do not matter.

Where data actually leaks

Before writing rules, look at where data really leaves. In a small or mid-sized organisation it is usually the same five routes:

  • Email to the wrong recipient, often through address autocomplete. A warning on an external address combined with a confidential label catches many of these.
  • Sharing links open to anyone. “Anyone with the link” as the default setting is the most common silent route. Set the default to people in your organisation or specific people.
  • Personal cloud storage and private email, usually for a good reason: the file was too large or it did not work from home. That tells you something about your own tools.
  • USB sticks and printers. Less than before, but still the route when someone leaves.
  • Pasting into AI tools and translation sites. The newest route and the hardest to see, because it is a browser window and not a file. It touches everything you do not manage; see our piece on shadow IT.

Two of these five, sharing links and the USB port, are better solved with a setting than with a DLP rule. That is the quickest win in this whole subject.

How to stop DLP stopping the work

DLP touches the daily work of people who did not ask for it. Whether it survives depends less on the technology than on three agreements.

  • Every rule has an owner outside IT. Someone from HR for staff data, someone from finance for account numbers. They decide on exceptions, not whichever administrator happens to be on duty.
  • Exceptions have a route and an end date. The accountant, the payroll bureau and the insurer get a standing exception on their domain, recorded and reviewed every year. An exception without an end date is a hole after two years.
  • Someone reads the alerts. A rule that fires hundreds of times a week and that nobody looks at only teaches staff that clicking past it is part of the job.

And explain in advance what happens and why. DLP inspects the content of email and files; staff should know that it does, where the line is and who sees the alerts. The ICO’s guidance on monitoring workers expects exactly that transparency, and an impact assessment where the monitoring is likely to be high risk. Do this before the rules go live, not after.

Where our work stops

What we do. The technical part: setting up the labels in Microsoft 365 or in the product you already have, default labels and automatic labels per folder or site, the DLP policies in test mode and then through the steps above, and summarising the alerts so the owner of a rule can decide. Also the settings that make a rule unnecessary, such as sharing links and external access, and what limits the damage if something still goes wrong: encryption and permissions that follow the role. We are ISO/IEC 27001 certified and apply this to ourselves.

What we do not do. We do not decide which of your organisation’s data is confidential; that is your decision, and we help turn it into labels. We are not lawyers and not a data protection officer: whether data is special category data under UK GDPR, whether using DLP on your staff is lawful and what you tell them is for someone qualified to judge. And we do not assess individual employees: an alert shows which rule fired, and anyone who follows it up is someone from your organisation. What to do when something has leaked anyway is in our article on data breaches.

Frequently asked

Questions we get about this

The five that come up most often.

What is DLP?

DLP stands for data loss prevention: software that recognises sensitive data about to leave the organisation, through email, cloud, USB or the browser, and then logs, warns or blocks. It recognises sensitivity by pattern, such as a National Insurance number, or by a label attached to the document. It is built for mistakes and carelessness, not to stop an attacker who is already inside.

What is data classification?

Agreeing a small number of labels that show how sensitive information is, and who applies them. Four usually suffice: public, internal, confidential and highly confidential. Describe each label with examples from your own organisation, because an abstract definition does not get read.

What is a data loss prevention policy?

A rule that says what DLP looks for, where and what it does when it finds it: for example, confidential documents sent to an external address trigger a warning, and highly confidential ones are blocked. Start each policy in test mode so you can see what it would catch before anyone is stopped.

Is DLP included in Microsoft 365?

Yes, in Microsoft Purview, together with sensitivity labels. Which parts you have, such as automatic labelling or DLP on the device itself, depends on your licence. Check that before you make a plan that relies on a feature you do not have.

Do you decide which of our data is confidential?

No. Which data is confidential is a decision for your own organisation, and whether something is special category data under UK GDPR is a legal judgement. We set up the labels and rules technically and summarise the alerts. We are not lawyers or a data protection officer, and we do not assess individual employees.

Shall we pick your first three DLP rules together?

We run the rules in test mode for a few weeks and show you what they catch, for whom and why. Then you decide which labels you need and what may be blocked.

Practical IT knowledge in your inbox

New guides on management, security and the workplace, written by the people doing the work. No sales talk, and you can unsubscribe in one click.

We use your address for the newsletter only. See our privacy policy.

Request a quote Call