Customer data platform: what it is, who needs one

"CDP" gets used for anything that stores customer data. It has a narrower meaning than that, and knowing it saves you a licence fee.

This page is only about the definition, the comparison and the buying decision. The strategic case for owning your own customer data is covered in our guide to first-party data, and that argument is settled — this is the next question down: does the argument require this category of tool?

What a customer data platform actually is

The category has an author. David Raab named it in 2013 and founded the CDP Institute in 2016. Its current definition, which the Institute says it updated in 2026, is this:

"Software that creates and maintains a persistent, unified customer record that is accessible to other systems."

One word matters. The earlier wording, still quoted across the web and recorded here in 2020, began "packaged software". The 2026 version drops it, and the Institute now says the services may be delivered inside the CDP or "composed from external systems such as data warehouses".

That single edit is this page's whole argument. A CDP is a job, not a purchase.

So three things define one:

1. A persistent, unified record. The same person, stitched across sources, kept over time — not rebuilt for each campaign.
2. Ownership of identity. The Institute's test is that the CDP "assumes primary responsibility for defining and maintaining customer identity".
3. Readable by other systems. Through an API, a query or a file extract. Data that only the tool itself can use fails the definition.

What is not in the definition: being one bought product, real-time anything, or a built-in campaign builder. Vendors sell all three. The definition asks for none of them.

Identity resolution is the job.

What named CDP vendors say they do

Read the documentation, not the homepage. Three examples, each from the vendor's own docs:

• Adobe Real-Time CDP — brings together "known and anonymous data from multiple enterprise sources in order to create customer profiles", groups them into audiences and activates those audiences to third-party destinations. Docs updated 18 June 2026.
• Twilio Segment — puts identity resolution "at the core", merging each customer's history into one profile using cookie IDs, device IDs, emails and custom external IDs. Docs updated 4 June 2026.
• Hightouch — warehouse-first: marketers "work directly with governed warehouse data" and results "write back to your warehouse". Docs updated 25 August 2026.

Same category, three different answers to one question: where does the customer record physically live? That, not the feature grid, is the decision you are making.

CDP vs CRM vs DMP vs data warehouse

Four systems that all "hold customer data" and are good at completely different jobs. Descriptions of the CRM, DMP and warehouse roles below follow the CDP Institute's own comparison.

The warehouse row is the one that changed. If your orders, events and app data already land in BigQuery, you hold the hard part. The Institute's own framing: a warehouse-native CDP "adds those functions on top of the warehouse" — so the question becomes which missing functions you need, not whether to start again.

Two neighbours worth keeping straight: picking the system of record for sales conversations is a CRM decision, and actually sending the message is a marketing automation job. A CDP does neither.

The prerequisites nobody puts in the deck

Three conditions decide whether a CDP works. No CDP fixes any of them for you.

A CDP bought without these becomes an expensive duplicate of the same mess, with a renewal date attached.

Do you need a CDP? The honest test

For most Indian mid-market businesses the answer today is no, not yet. That is not a criticism of the category — it is a sequencing judgement. A CDP earns its cost when most of the following are already true:

• Four or more systems hold customer data that genuinely must agree — app, website, POS, call centre, loyalty.
• The same person reaches you under different identifiers often enough that mis-stitching costs real money.
• Marketing needs to build and change segments weekly, without raising a ticket each time.
• Event data is already clean, and someone owns the data model.
• Lifetime value per retained customer is large enough that sharper targeting pays the licence plus the implementation.

The far more common picture: two or three systems, one analyst, a warehouse that already holds the orders, and a CRM nobody has finished configuring. In that case buy neither a CDP nor a DMP. Land your data in the warehouse, keep segment logic there, and move audiences out with something small. You get most of the personalisation upside and keep the option open.

And be clear what a CDP does not solve. It will not settle which channel deserves credit — that is an attribution problem — and it will not simplify a stack that has no owner.

What we ask a client to fix before the CDP conversation

DH's own view, from our client work: we do not take a CDP evaluation further until four things exist. They cost nothing but decisions.

1. One identifier policy. A single primary key for a person — usually mobile number in India — collected everywhere, with a written rule for records that lack it.
2. One event dictionary. A shared sheet of event names and properties that both marketing and engineering sign off before anything new ships.
3. Consent captured at the point of collection, with the purpose recorded as data — not reconstructed later from memory.
4. A named data-model owner with the authority to say no.

If a brand cannot name the person in point four, we stop there. Not because the tools are poor — because the tool inherits the ambiguity and charges rent on it.

Why DPDP makes the unified profile a design question

This is a design consideration, not legal advice — take legal advice. But any unified profile store in India now has to be designed against India's Digital Personal Data Protection Act, 2023 (gazetted 11 August 2023). Section 6(1) requires consent to be given "for the specified purpose and be limited to such personal data as is necessary for such specified purpose", and section 5 requires a notice stating that purpose.

A profile store's instinct is the opposite: collect everything, keep it indefinitely, use it for whatever comes up. So decide three things at design time, not after procurement.

• Purpose and consent state live on the profile as queryable fields, so an audience can exclude people who never consented to that use.
• Retention is set per field, not per system. Order history and ad-targeting signals do not need the same life.
• The deletion path is tested — including every downstream tool the profile was pushed to.

Our DPDP guide for marketers covers the obligations themselves in more detail.

Build the foundation and the decision gets easy

Everything a CDP needs is useful without a CDP. One identifier policy, one event dictionary, clean collection, consent recorded as data, a warehouse that holds the truth, one named owner.

Do that work and one of two things happens. Either your existing warehouse and CRM start answering the questions you were going to buy a platform to answer — or the gap becomes so specific that you can brief vendors on exactly the missing function instead of buying a category. Both outcomes are wins, and only one of them has a licence fee.