Quick answer
Data processing is the series of steps that turns raw data into usable results: collecting it, cleaning and checking it, transforming it, storing it and presenting the output. In analytics it is what turns raw hits into visitors, sessions and reports.
What is data processing?
Data processing is the series of steps that turns raw data into something you can use: collect it, check and clean it, transform it, store it, and present the result. A click on a button is raw data. "Mobile visitors converted at 5.0% last week" is the processed result, and everything in between is data processing.
The word "processing" does not imply a computer or a particular scale. A person tallying survey forms by hand is processing data; so is a pipeline handling millions of events a day. What defines it is the direction: from messy, unorganized input to organized output that answers a question.
A note on a second meaning: in privacy law, "processing" is much broader. Under the EU GDPR (Article 4(2)) it covers any operation on personal data, including collecting, storing, using or deleting it. When someone says "data processing agreement", they mean that legal sense. This page is about the analytics sense; the legal sense is explained in the GDPR text itself.
What are the steps of data processing?
| Step | What happens | Typical problems it fixes | Example (web analytics) |
|---|---|---|---|
| 1. Collect | Capture raw data from its source | Missing sources, duplicated sends | Script or SDK sends a page view |
| 2. Prepare (clean) | Remove, correct or flag bad records | Bots, test traffic, malformed or duplicate rows | Drop crawler traffic; reject a payload with no event name |
| 3. Input | Load the cleaned data into the processing system | Format mismatches | Parse the record into time, event, visitor, properties |
| 4. Process | Transform, join, group and calculate | Raw rows are not answers | Group page views into sessions; count visitors per day |
| 5. Output | Present results as tables, charts or exports | Numbers nobody can read | Dashboard, scheduled report, export file |
| 6. Store | Keep raw and processed data under a retention rule | Cost, privacy, auditability | Keep daily totals; delete raw rows after the retention period |
The order is not always strict. Storage happens at several points (raw data is usually kept before cleaning), and many systems loop: an error found in output sends you back to collection. Detail on the preparation step is in data preparation; moving data between systems is covered in ETL and data pipeline.
A worked example: from 10,000 hits to a visitor count
Sample data: one day of traffic reaching a collection endpoint.
| Stage | Records | What changed |
|---|---|---|
| Received | 10,000 | Raw hits |
| Minus recognised bots | − 1,400 (14.0%) | Crawlers kept apart from people |
| Minus invalid payloads | − 90 (0.9%) | Missing fields, wrong domain, rate-limited |
| Clean human hits | 8,510 | 10,000 − 1,400 − 90 |
| Grouped into sessions | 3,404 | 8,510 ÷ 2.5 pages per session |
- 1,400 ÷ 10,000 = 14.0% of raw hits were bots, so counting raw hits as visitors would have overstated human traffic by 10,000 ÷ 8,510 − 1 = 17.5%.
- 90 ÷ 10,000 = 0.9% were unusable. Small, but worth watching: a jump to 9% would signal a broken tracker, not a change in audience.
- 8,510 human hits at 2.5 pages per session give 3,404 sessions. Change the rule that defines a session and this number changes, which is why the rule must be written down.
The point: the same raw feed gives 10,000, 8,510 or 3,404 depending on which processing steps ran. A metric without its processing rules is not comparable.
What is the difference between batch and real-time processing?
| Batch | Stream (real time) | |
|---|---|---|
| How it runs | Collect for a period, then process everything at once | Process each record as it arrives |
| Typical delay | Minutes to a day or more | Seconds |
| Strength | Efficient; easy to re-run and correct | Immediate; good for monitoring |
| Weakness | Stale between runs | Harder to correct late-arriving or out-of-order data |
| Use for | Daily reports, billing, reconciliation | Live dashboards, alerts, fraud checks |
Many systems do both: a fast path for "what is happening now" and a slower, more complete path for the report you will rely on tomorrow. The Dataflow Model paper by Akidau and colleagues (2015) is the standard reference for how to reason about both in one design. The delay between an event happening and appearing in a report is data latency.
What are the types of data processing?
- Manual: people do the work by hand (a spreadsheet tally). Slow, error-prone, fine for tiny volumes.
- Mechanical / electronic: software or machines do fixed steps automatically.
- Batch and stream: see above.
- Distributed: the work is split across many machines; MapReduce (Dean and Ghemawat, 2004) is the classic description.
- Transactional (OLTP) vs analytical (OLAP): the first records individual events reliably (an order); the second summarizes many of them (orders per week).
What goes wrong in data processing?
- Cleaning without a record. If you drop rows and do not log how many, nobody can explain why this week's total changed.
- Mixing raw and processed counts. "Hits", "sessions" and "visitors" are different stages; labelling matters.
- Silent rule changes. Changing the bot list or the session rule shifts history; note the date.
- Processing before deciding the question. Heavy aggregation can throw away the detail you later need. Keep raw data as long as policy allows.
- Late data ignored. Mobile events sent hours later land on the wrong day unless the pipeline handles event time.
- Keeping personal data longer than needed. Retention is a processing step, not an afterthought.
How does TapCub process analytics data?
Following the 10,000-hit example, here is where each step happens in TapCub:
- Collect. The website script, an iOS, Android, Flutter, React Native or Harmony SDK, a mini-program SDK, or the server HTTP interface sends events. The tracking script stores no cookie by default, and the address is truncated before storage.
- Filter. Recognised bots and AI crawlers are counted only in Web analytics › Bots, so they never enter human visitor counts. All-platform analytics › Data › Data quality lists collector rejects by reason, such as domain mismatch, exclusion rule, quota, rate limit, invalid payload and bots.
- Validate. Events marked Blocked are dropped before storage. Validation rules in the Property dictionary can warn, drop a property, block the event or quarantine it. The Tracking plan with live verification (24 h) shows whether planned events arrive complete.
- Aggregate. Reports read the processed events, with visitor, device and person counting modes. The daily totals do not hold personal data and are not deleted on the raw-data retention schedule.
- Output and store. Use charts and reports for answers, All-platform analytics › Data › Data export to create an asynchronous raw export, and the project's privacy and retention settings for how long raw rows stay. Web analytics › Realtime shows people active in the last 5 minutes within seconds, while search-engine data connected in Monitoring › Search performance usually arrives one to three days late, so do not mix the two timelines in one comparison.
TapCub does not offer a pipeline editor for custom transformations; its processing rules are the ones in these settings.
Sources
- European Union, Regulation (EU) 2016/679 (GDPR), Article 4(2) "processing".
- Akidau, T. et al. (2015). The Dataflow Model: A Practical Approach to Balancing Correctness, Latency, and Cost in Massive-Scale, Unbounded, Out-of-Order Data Processing. Proceedings of the VLDB Endowment, 8(12), 1792–1803.
- Dean, J., & Ghemawat, S. (2004). MapReduce: Simplified Data Processing on Large Clusters. OSDI '04.
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.