Customer analytics is usually bought to answer a commercial question: which segments repeat, where the funnel leaks, which campaign pays for itself. Privacy enters late, when someone notices that the tool is collecting device identifiers, location and browsing history across every page, and no one can explain why.
A better sequence is to define the questions, then the minimum data needed to answer them, then the controls that make the collection defensible. This buying guide is written for a manager in Bangladesh who is selecting an analytics platform, a customer data platform or a specialist measurement tool, and who does not want a compliance problem attached to a dashboard.
Start with the decision, not the data
Write the five decisions the business needs to make in the next two quarters. Examples: whether to invest in a retention campaign for a customer segment, whether a delivery option is losing repeat orders, or whether a product bundle should be discontinued. For each decision, name the metric, the population and the freshness required. If a field is not needed for a decision, it is a candidate for exclusion. This is the simplest form of data minimisation and the most effective.
Requirements to put in the brief
- Lawful basis and consent: how the tool records consent, how it handles withdrawal, and whether it can operate on legitimate interest for the intended use.
- Data map: which fields are collected, from which source, where they are stored, who processes them and how long they are kept.
- Minimisation: the ability to disable unnecessary fields, truncate identifiers and avoid free-text collection.
- Access control: role-based access, single sign-on, audit logs and a named administrator on your side.
- Pseudonymisation: whether customer identifiers can be replaced with stable but non-identifying keys.
- Retention and deletion: configurable retention, deletion on request and evidence that deletion happened.
- Sub-processors: a published list, notice of changes and a right to object.
- Cross-border transfer: where data is stored and processed, and which transfer mechanism applies.
- Model training: whether your data is used to train the vendor's models, and how to opt out.
- Export: a complete export in a usable format, not a screenshot or a summary.
Vendor scorecard
Score each vendor from 0 to 3 against the same criteria, and write the evidence next to the score. A vendor that cannot answer a question should score zero, not "to be confirmed".
- Purpose fit (weight 20): can it answer the five decisions with the data you already hold?
- Data governance (20): retention, deletion, access control, audit logs and sub-processor management.
- Security (15): encryption, tenant isolation, vulnerability disclosure and incident history.
- Transfer position (10): storage location, transfer mechanism and notice of changes.
- Integration effort (10): SDKs, server-side events, identity resolution and data quality controls.
- Usability (10): whether a non-specialist can answer a question without exporting to a spreadsheet.
- Commercial model (10): pricing that scales with events or profiles, and a clear renewal path.
- Exit (5): export completeness, deletion attestation and transition support.
Weights are a starting point. Change them to reflect your risk, but keep them fixed across vendors so the comparison is honest.
Pricing questions that reveal the real cost
- What is metered: events, user profiles, data storage, seats or API calls?
- What happens when volume grows by half? Is there a step change or a smooth curve?
- Are identity resolution, data warehouse sync and retention extensions included or separate?
- What does implementation cost, and who does the tagging?
- What is the support model, and is there a named technical contact?
- What is the notice period, and what happens to historical data after termination?
Pilot design
Run a six-week pilot on one customer journey and one decision. Define the baseline before the tool goes live, instrument the journey, and compare the decision quality rather than the number of charts. Include a deletion request and an export during the pilot; these are the two processes most likely to be untested in a demo. Ask the vendor to show the audit log for a specific user action, not a generic screenshot.
Consent and notice in practice
Consent is not a banner. It is a record of what a person agreed to, when, and for which purpose. Check that the tool can store the consent version, the timestamp and the source, and that withdrawal is as easy as consent. If the analytics relies on legitimate interest, document the assessment and make sure the privacy notice names the purpose in plain language. Test the notice on a phone: if a customer cannot find the explanation in two taps, it is not effective notice.
Also decide how the business will handle a request from someone who is not a logged-in user. Device-level data, cookies and offline identifiers often sit outside the customer account, and the support team needs a procedure for finding and deleting them.
In Bangladesh, notice should work in Bangla as well as English, and on a low-end phone. A consent screen designed for a desktop browser may be unreadable on the device most customers use. Test the wording with a small group and keep the version that was shown to each cohort.
Contract terms that matter
- A data processing agreement that names the processing purpose, duration and categories of data.
- Breach notification within a defined period, with a named contact.
- Audit rights or an accepted independent assurance report.
- Change control for sub-processors.
- Deletion and return of data at termination, with a certificate.
- A limitation of liability that reflects the sensitivity of customer data.
Exit plan
Assume you will leave one day. Keep the event taxonomy and the metric definitions in your own repository, not only in the vendor's interface. Export raw events monthly to a store you control. Document the identity resolution rules so a replacement can reproduce them. A vendor that makes export difficult is telling you what the exit will feel like.
Where this fits in our work
We design analytics and data foundations under Business Automation, from event design to reporting. The pricing page explains how a data engagement is scoped. For the foundation layer, read how AI-ready data foundations change year by year.
Sources
- NIST Privacy Framework, accessed 17 November 2026.
- NIST AI Risk Management Framework, accessed 17 November 2026.
- DataGuidance, Bangladesh data protection overview, accessed 17 November 2026.
Responses (0)
No comments yet. Add the first one below.
Leave a response
Your email address will not be published. Required fields are marked *.