Customer 360
Your customer shows up in your website analytics, your app, your order system, your email tool and your support inbox, usually as five different people. We bring those records together so you can see one customer and treat them as one.
Signs you could use this
- You can't say how many unique customers you really have.
- Marketing emails go to people who just bought, or have just complained.
- Segments are built by hand in spreadsheets each time.
- You'd like to personalise but don't have a reliable customer table to base it on.
What we do
- Map the sources
- We list every system that holds customer data, what identifies a person in each one and how often it updates.
- Match the records
- Rules for deciding that two records are the same person, tested on real data, with the uncertain cases flagged rather than guessed.
- Build the customer table
- A clean, documented table in your warehouse with one row per customer: first and latest activity, orders, value, channels, support history.
- Put it to work
- Recency, frequency and value segments, churn signals, cohorts and audience lists that your marketing and product teams can use straight away.
What you receive
Usually six to ten weeks for a first version covering your main sources. Further sources can be added later.
- A documented customer table in your warehouse
- Identity matching rules with accuracy checks
- A starter set of segments, such as RFM and lifecycle stage
- Access and consent notes so you use the data responsibly
Tools we use: BigQuery, AWS Redshift, PostgreSQL, Airflow, Airbyte, Python, SQL, Power BI.
Common questions
Do we need a customer data platform?
Not necessarily. For many businesses a well-built customer table in an ordinary warehouse is enough, and far cheaper. We'll tell you if a platform would genuinely help.
What about privacy?
We design with consent and minimal data in mind, and we only combine data in ways your privacy policy allows. See our note on data ethics in the Insights section.
How accurate is the matching?
We measure it on a sample, show you the results and flag uncertain matches for review. Perfect matching doesn't exist, but good-enough, measured matching does.