TapCub is live — analytics, insights and live chat, free on one platform
Glossary

What Is Data Discovery? Four Steps, an Example and Common Mistakes

Data discovery is finding out what data you have, where it lives, how good it is and who can use it. How it differs from a catalog, four steps and a worked inventory of six sources.

Ethan Brooks Reviewed by Su He Also called data discovery process Last updated Nov 18, 2026 Glossary

Quick answer

Data discovery is finding out what data you have, where it lives, how good it is and who can use it, before you analyze it. In practice it is four steps: inventory, classify, assess quality and record an owner and definition for each item.

What data discovery means

Data discovery is the work of finding out what data you have, where it lives, how good it is and who can use it, before you try to analyze any of it. IBM describes it as collecting, evaluating and exploring data from multiple, often disparate, sources. The first step of an analysis is often not analysis at all: it is finding the right data and deciding whether to trust it.

The phrase gets used for two different jobs, and the difference matters when you read other pages:

  • Discovering your data assets (this entry): taking stock of datasets, event streams, tables and files, and recording what each one is. This is what data teams, privacy teams and new analysts mean.
  • Visual data discovery: exploring a dataset you already have, by dragging fields into charts to look for patterns. Some BI tools market this under the same name. It starts after the inventory work and does not replace it.

A short test: if the question is "which data do we have?", you need discovery. If the question is "what does this data show?", you need analysis. Teams that skip the first step spend the second one arguing about whose numbers are right.

Discovery, catalog and governance: how they differ

Three terms get mixed up. They describe different things: an action, a record, and a set of rules.

TermWhat it isTypical outputQuestion it answers
Data discoveryAn activity: finding and assessing dataAn inventory with a quality note per itemWhat data do we have, and can we trust it?
Data catalogA record: where discovered data is organized and searchableA searchable index with descriptions, owners and tagsWhere is it, and what does it mean?
Data governanceRules and roles: who may do what with dataPolicies, owners, access and retention rulesWho decides, and what are the limits?

IBM's description of the sequence is a useful guide: data is discovered and profiled first, then organized in a catalog. A catalog tool without ongoing discovery goes stale; discovery without a catalog gets repeated by every new hire. A data dictionary is a smaller thing again: the definitions of the fields inside one dataset.

The four steps of data discovery

Whatever tool you use, the work comes down to four steps. Each one has an output you can show someone.

  1. Inventory. List every place data is created or stored: event streams, databases, exports, spreadsheets, vendor tools. Count rows and note when each was last updated. Include the awkward ones, such as the marketing spreadsheet that has never been in a system.
  2. Classify. Say what kind of data each item holds (behavior, customer, financial, content) and how sensitive it is. A simple three-way split is enough to start: directly identifies a person, could identify someone in combination with other fields, or is not personal.
  3. Assess quality. Check a few measurable things on the fields that matter. The UK Government Data Quality Framework, which follows DAMA UK, names six dimensions: completeness, uniqueness, consistency, timeliness, validity and accuracy. Completeness (the share of records that have the field) and timeliness (is the latest data recent enough?) are the cheapest to compute and catch most problems.
  4. Record. Write down for each item an owner, a one-line definition, how often it refreshes and who can use it. This is the step teams skip, and it is the one that stops the next person starting from zero. If you also want to trace where each dataset comes from, that is data lineage.

AWS's analytics guidance frames the same idea as an early, interactive phase of working with stakeholders to learn what sources exist and how they will be used, which is why step 4 needs a person's name beside each item and not just a scan result.

A worked example: an inventory table

Sample data: an online store runs the four steps on six data sources. Completeness is the share of rows that have the field the data is meant to carry.

DatasetWhereRowsOwnerRequired field presentCompletenessLatest dataVerdict
checkout_completedEvent stream4,820Growth engineeringorder_id on 4,6754,675 ÷ 4,820 = 97.0%TodayUse
signup_submittedEvent stream2,310Growth engineeringplan on 2,3102,310 ÷ 2,310 = 100.0%TodayUse
purchase (old name)Event stream1,120Noneamount on 1,0081,008 ÷ 1,120 = 90.0%11 months agoRetire, map to checkout_completed
button_clickAutocapture51,400Nonelabel on 38,55038,550 ÷ 51,400 = 75.0%TodayName it or ignore it
crm_contactsCRM export8,900Sales operationscountry on 8,0108,010 ÷ 8,900 = 90.0%2 weeks agoFix, then use
newsletter_list.csvSpreadsheet12,600Nonevalid email on 9,4509,450 ÷ 12,600 = 75.0%5 months agoFind an owner first

The verdict rule is explicit so it can be repeated: an item is usable now if it has a named owner, a written definition, required-field completeness of at least 95% and data newer than its expected refresh. Here two of six sources pass (2 ÷ 6 = 33%). The 95% line is our own rule of thumb; set yours from what a wrong number would cost.

Two things stand out. First, many of the problems are about ownership, not quality: three of the six sources have no owner, which a scan of the data itself would never have told you. Second, the old purchase event and checkout_completed describe the same thing under two names, a duplicate that only an inventory reveals.

Common mistakes in data discovery

  • Treating it as a one-off project. New events appear with every release and old ones go silent. Re-run the inventory on a schedule, such as monthly or after each major release.
  • Buying a catalog before doing the inventory. A tool shows you what it can scan. The spreadsheet on someone's laptop is still missing, so list it by hand.
  • Mistaking a scan for understanding. A scanner can find a column called status; only an owner can say that 3 means "refunded". Step 4 exists for this reason.
  • Measuring volume, not fit. "We found 4,000 tables" says nothing about whether any of them answers your question. Report how many are usable, as the example does.
  • Skipping sensitivity. Classifying in step 2 is what lets you mask identifiers by default and answer a deletion request later.
  • Documenting without an owner. A definition nobody maintains is wrong within a quarter.

How to run data discovery in TapCub

Take the store from the example, which has been tracking events for a year and now asks "which events do we have, and which can we trust?" TapCub covers the data that flows into a TapCub project. It does not scan your warehouse, CRM or spreadsheets; for those, build the inventory by hand as in the table above.

  1. Inventory the events. All-platform analytics › Data › Events & properties lists each event with how often it fired, how many visitors triggered it, its share and a seven-day trend. Sort by trigger count to see the big ones and the long tail.
  2. Clear the unnamed. All-platform analytics › Data › Event inbox shows two things that have no definition yet: events your code sent under a name nobody registered (Unplanned) and autocapture candidates, such as the button_click row. For each one, accept it with a proper name, merge it into an existing event or ignore it. Accepting a candidate backfills the last 90 days. See handle unplanned events for the steps.
  3. Compare plan with reality. All-platform analytics › Data › Tracking plan records the events you intended to collect, with owner and status. Turn on Live verification (24 h) to see hits per event and completeness of required properties, your completeness check from step 3 of the process. Low completeness is a warning; it does not block events.
  4. Classify the properties. All-platform analytics › Data › Property dictionary scans user, event and account properties, guesses each type and sorts each key into direct identifier, indirect identifier or non-personal. Direct identifiers are masked by default; if you change a type or class and save, later scans do not overwrite it.
  5. Check the sources. All-platform analytics › Data › Data sources lists the registered domains and apps with their Last event time, which is the timeliness check from step 3.

TapCub does not keep a catalog of your external tables or decide who owns an event. The owner column in the tracking plan is where you record that.

Sources

Was this explanation clear?

We use your answers to fix and expand this page.

Ethan Brooks

Solutions Engineer

Covers SDKs, API integrations, and data implementation, with an emphasis on validation, system interoperability, and troubleshooting.

FAQ

Questions people ask

Short answers to the follow-up questions that come up most.

Still have a question?

Chat with the team behind TapCub — we usually reply within a few hours on business days.

What is data discovery in simple terms?

Data discovery is taking stock of your data: what exists, where it is stored, what each item means, how complete it is and who owns it. It happens before analysis, and it answers the question of which data you can trust and use, not what the data shows.

What is the difference between data discovery and a data catalog?

Discovery is the activity of finding and assessing data; a catalog is the searchable record where the results are kept. Discovery without a catalog gets repeated by every new person, and a catalog without ongoing discovery goes stale. The two work best together, with discovery first.

What are the steps of the data discovery process?

Four steps cover most cases: inventory every place data is created or stored, classify each item by type and sensitivity, assess quality on measurable dimensions such as completeness and timeliness, and record an owner, a definition, a refresh rate and access for each item. Repeat on a schedule.

How often should you run data discovery?

Run it on a schedule that matches how fast your data changes, such as monthly or after each major release. New events appear with every release and old ones go quiet, so a one-time inventory is out of date within a few months.

See this number on your own site

Add one script tag for web and apps, and the first reports fill in within minutes.

See your data clearly. Find your growth.

Every click, backed by data. Install one line of code and see your first numbers in a minute.

No credit card · Free plans for analytics and chat · Cookieless analytics