Quick answer
Data unification is the process of matching, deduplicating and reconciling records about the same person, account or thing from several systems so each appears once, with consistent fields and one stable identifier, for example one profile instead of separate web, app and chat records.
What is data unification?
Data unification is the process of taking records about the same people, accounts or things from several systems and turning them into one consistent set: the same entity appears once, its fields are reconciled, and it carries one stable identifier. A web visit, an app session and a support chat from one person become one profile instead of three unrelated rows.
It is more than moving data. A research paper on the Data Tamer system describes data curation as discovering sources, cleaning and transforming the new data, integrating it with other local sources, and then deduplicating the combined result. The last step is what unification adds: after copying, the same person is still in your data three times until you match the records.
What are the steps of data unification?
One vendor's documentation for a customer-data product (linked in Sources) describes unification as four steps. The names are theirs, but the sequence is the general one:
| Step | Question it answers | Example |
|---|---|---|
| 1. Select and map source data | Which tables describe the entity, and which column is the email, the phone, the name? | Map mail, email_addr and e-mail to one "email" field |
| 2. Deduplicate within a source | Are there several rows for one person in the same table, and which row is best? | Two web profiles for one login: keep the most recent |
| 3. Match across sources | Which rows in different tables are the same entity? | App user 4127 and billing customer 88 share a login id |
| 4. Build the unified view | Which columns are kept, merged or dropped? | Six email columns merge into one |
Two further points from the same documentation matter in practice. Activity data, such as purchases, is kept out of the profile because it has a one-to-many relationship to the customer. And the unified profile gets an identifier that stays the same between runs except when profiles merge (two profiles become one) or split (one becomes two), which is recorded as a "previous ID". Plan for both, because every report keyed on the old ID changes when it happens.
What does unification look like for one user?
Sample data: one person, Ada, uses a product on the web, in a mobile app and through live chat, and pays by invoice.
Before: four records, four identifiers
| System | Identifier | What it holds |
|---|---|---|
| Website analytics | Visitor id 3FA2C1 | 2 sessions from a newsletter link, pricing page viewed 3 times |
| Live chat | The same visitor id 3FA2C1 | One conversation asking about annual billing |
| Mobile app | Device id d_41 | App opened 5 times, checkout started, not completed |
| Order system | Login id u_123 | One paid order, $240 |
After: one profile keyed by u_123
| Layer | What the profile now shows |
|---|---|
| Identity | Visitor 3FA2C1 and device d_41 both bound to u_123 |
| Timeline | Newsletter click → pricing views → chat about billing → app checkout started → order |
| Answer it enables | "She asked about annual billing, started checkout in the app and paid on the web" is one story instead of four |
Chat and web already shared an id because they come from one script; the app device and the order record joined only when a shared key (the login id) was present. That is the general rule: unification can only be as good as the keys the sources share.
The effect on numbers is just as concrete. Sample data: 1,000 web visitors, 400 app devices and 150 chat visitors. The chat visitors are a subset of the web visitors. 180 of the app devices belong to people who are also among the web visitors, matched by login.
- Counting every system's "users" and adding: 1,000 + 400 + 150 = 1,550.
- After unification: 1,000 + 400 − 180 = 1,220 distinct people. The naive count overstates by 27.0%.
- With 61 buyers, the conversion rate is 61 ÷ 1,550 = 3.9% on the naive count and 61 ÷ 1,220 = 5.0% on the unified count.
The business did not change. The denominator did.
How are records matched?
There are two families of rules, and most real projects use both in order.
- Deterministic matching joins on a stable key that two systems both hold: a login id, a hashed email address, an order id. It is simple and explainable, and it is the rule you should try first.
- Probabilistic (fuzzy) matching scores how similar two records are on several fields (name, address, device details) and merges above a threshold. It finds links that no shared key reveals, but it will sometimes be wrong.
Measure a matching rule instead of trusting it. Sample data: a rule proposes 200 pairs as "same person". A reviewer checks them: 190 are correct and 10 are not. The reviewer also finds 30 true pairs the rule missed.
- Precision = 190 ÷ (190 + 10) = 95%, so 5% of merges are wrong.
- Recall = 190 ÷ (190 + 30) = 86.4%, so 13.6% of real matches were missed.
A false merge is usually worse than a missed one: it mixes two people's histories and is hard to notice, while a missed match only leaves one person split in two. When in doubt, set the threshold so precision stays high, and keep the weaker rule for suggestions a human confirms.
How is data unification different from data integration?
| Term | What it does | Output |
|---|---|---|
| Data integration | Moves and combines data from several systems | Data in one place, still possibly duplicated |
| Data unification | Matches, deduplicates and reconciles records about the same entity | One record per entity with one id |
| Single customer view | The result for people: one profile per customer | A profile you can read |
| Unified data model | One shared structure for events and entities across sources | Same event names and fields everywhere |
Customer data platforms bundle unification with other functions. Whatever tool you use, the unification step is the same four questions above.
What goes wrong in unification?
- False merges. Two people sharing a family tablet or a work computer end up as one profile. Prefer a login id over a device id for the merge key.
- Missed merges. An anonymous visitor who never logs in cannot be linked to anything, and some identifiers expire. If the visitor id rotates daily, today's browser and tomorrow's browser are different visitors.
- Changing ids. A merge or split changes the unified id. Keep the old id so history can be joined.
- No consent. Combining records is processing personal data. Link only what your privacy configuration and the person's consent allow, and keep anonymous aggregates anonymous.
- Unreconciled meanings. Matching rows is not the same as matching fields: "revenue" with and without tax, or two time zones, produces a profile that is wrong in a confident way.
A checklist before you unify
- List the sources and, for each, the identifier it holds (visitor id, device id, login id, email, order id).
- Choose the strongest shared key and write down the order of fallback keys.
- Decide what happens on a merge or split, including how old ids are kept.
- Sample 200 proposed matches by hand and compute precision and recall before applying a rule to everyone.
- Use the unified count, not the sum of systems, as the denominator for conversion and retention rates.
How does this work in TapCub?
Take Ada's four records above. Here is how they come together in TapCub, and where the limits are.
- Web and chat share one visitor id. They run from one script, so a chat conversation already carries the visitor's source, current page and the last two days of browsing, and View in analytics opens the same visitor's profile. All data sources in a project (website, App, mini program) share one identity model and one event table.
- Call
tapcub.login()after sign-in with your own user id. The first time a device binds to a login id, that session and the device's earlier anonymous events receive the user id; the same user on the app joins through the same id. See identify users. - Read the result in People. A person appears in All-platform analytics › Users › People after login. One login id is one person across web, App and mini program, and the profile shows a day-by-day timeline that merges the bound devices.
- Send the order from your server with the same user id, so purchases join the timeline. See send server-side events.
The limits are the ones above. Identified mode needs a plan that includes it, the console has no switch for it yet (contact support to turn it on), and visitors who never log in stay in aggregates and do not get a profile. On the strict privacy preset the visitor id rotates every day, so there is no single profile across days. See identified vs anonymous. TapCub does not match records by name or address; it joins on the ids you send.
Sources
- Microsoft Learn, Data unification overview (four-step process, customer id behaviour on merge and split; page dated November 2026).
- Stonebraker et al., Data Curation at Scale: The Data Tamer System, CIDR 2013 (curation as discovery, cleaning, integration and deduplication; curation cost reduced by about 90% in three enterprise cases).
Was this explanation clear?
We use your answers to fix and expand this page.
Couldn’t send. Please try again in a minute.
Thanks for the feedback. We’ll use it to improve this page.You’ve already sent feedback on this page. Thanks.

Data Analyst
Explores traffic sources, conversion funnels, and retention, using clear metric definitions and concrete examples to explain what the numbers mean.