“Single customer view” has become a strikingly popular phrase for anyone selling enterprise software. CRM vendors promise it. So do CDP providers, data cloud suppliers, and increasingly the companies looking after your warehouse. You could spend a fortune comparing them and still end up with agents who can’t tell whether a refund has cleared.
Customer data management (CDM) is the work of keeping customer information accurate, connected, secure, and available where it’s needed: deciding which records to trust, sorting out duplicate identities, and getting information to the people using it. A CDP might help enormously. So might fixing an integration you’ve been working around for three years. As always, AI has raised the stakes, as Carlie Idoine, VP Analyst at Gartner, put it,
“AI needs to be tightly aligned with data, analytics and governance.”
Despite this, Salesforce’s 2026 survey of 3,450 commerce professionals found that only 27% said their customer data was fully unified across sales, service, marketing, and commerce.
It’s time to fix the customer data problem in the contact center once and for all.
Key Takeaways
-
A disappointing CRM doesn’t automatically justify buying a CDP. Find out what’s failing first.
-
CDPs can connect customer identities, although incorrect matching can create problems worse than the duplicates they replace.
-
Data warehouses and composable CDPs deserve serious consideration before you agree to another separate customer database.
-
“Data cloud” tells you remarkably little about the capabilities included in a contract. Make suppliers show you what’s actually available.
-
Check identity accuracy, data freshness, and consumption charges using your own workloads. Vendor demonstrations rarely expose the expensive surprises.
If an agent can’t find a customer’s refund, buying a CDP is an absurdly expensive answer. If millions of records genuinely need stitching together across channels, it could be money well spent.
CRM, CDP, Data Warehouse and Data Cloud: What Does Each Do?
A data warehouse can hold more about a customer than anyone in the business, while the agent taking their call knows next to nothing about them. Neither system is failing; they were built for different jobs. The issue is that companies expect one product to handle everything, which isn’t surprising when vendors keep using the same language for different capabilities. It’s also why customer context has become CRM’s next competitive battleground, especially now AI agents need those records mid-interaction.
Some of the capabilities you’re shopping for may already be sitting in your existing technology. Databricks, for instance, describes using Hightouch for activation while customer information stays in its lakehouse, a sensible starting point for businesses already invested in a warehouse.
The CDP Institute’s 2026 research also gives buyers reason to challenge the “everything under one roof” pitch. Of five major platforms examined, none documented behavioral collection, identity resolution, and consent/data-rights management together as generally available first-party capabilities.
How Does Customer Data Management Work?
A customer changes their phone number in the banking app. Six months later, the fraud team is still sending verification codes to the old one. Both systems have accurate records of what they were told. Nobody’s made sure the newer information takes precedence.
That’s the sort of problem customer data management (CDM) is supposed to catch. There are five parts to the process, and it’s definitely worth paying attention to what happens in between them:
-
Collect first-party data: Everything customers have told you or generated, from purchases and support requests to website activity and account preferences.
-
Standardize and clean data: Check what records mean before combining them. One system’s “inactive” could be another’s “temporarily suspended,” and that causes a surprising amount of trouble.
-
Unify customer identity: Three email addresses might belong to one customer; one phone number might belong to a whole household. Matching rules need to handle both, especially where money is involved.
-
Activate data across teams and tools: Service agents should see what marketing and sales already know, where they’re permitted to, which means CRM and contact center software talking to the other systems through pipelines that don’t need constant rescuing.
-
Govern and protect customer data: Someone needs the authority to settle disputes between records, with corrections tracked and the original source traceable. If bad data has affected someone’s account, “we don’t know where that came from” is a dreadful answer.
Also, borrow one rule from master data management: agree which system has authority over each customer field. The returns platform should be able to correct a return status without an older CRM record reversing it overnight.
CRM, CDP or Data Cloud: Which Do You Actually Need?
It’s usually worth exploring a few cheaper options before putting another customer data platform on the shopping list. Particularly if you’re already paying for a CRM and a warehouse that nobody seems to be using properly.
When Is Your CRM the Problem?
Sit with an agent for an hour and count how often they leave the CRM for information it should already display. Then check 50 closed cases: are orders and billing details linked, or pasted into the notes? Those findings can save an unnecessary purchase, because single customer view projects fail despite plenty of CRM investment. If you’re actually replacing the sales CRM, our 2026 Gartner Magic Quadrant rundown covers that separate decision.
When Is a CDP Worth Paying For?
When fragmented identities are genuinely preventing recognition or cross-channel activation, beyond what your CRM’s identity features can do. Be ready for it, though: Alium Research’s 2026 buyer interviews describe an enterprise software company that suspended its CDP project because its data wasn’t ready. If nobody owns duplicates, conflicting fields, or preferences, assign that first. A new CDP inherits the arguments your existing systems are already having.
Could You Keep Your Existing Architecture?
Quite possibly. Pets at Home uses separate Salesforce environments because different parts of its business need different things; there’s no advantage in forcing everything into one platform for the sake of it. Give the supplier a journey that’s regularly causing trouble and a sample of your records. Can they recreate the problem and sort it out without an engineer holding everyone’s hand? If your existing technology can handle the fix, ripping everything out is a peculiar way to spend the budget.
What Does Fragmented Customer Data Actually Cost?
Say an agent spends 90 seconds finding an order because the CRM and billing system aren’t talking properly. Multiply that by 400 calls a day, and you’ve lost 220 hours over a 22-day working month. At a loaded labor cost of $35 an hour, that’s $7,700 worth of agent time.
That’s not $7,700 in cash unless you can cut overtime, avoid hiring, or handle more customers with those hours, and the commercial damage (the customer who renewed elsewhere because nobody spotted their complaint) is harder still to pin down. Salesforce’s 2026 State of Commerce research found 37% of B2B respondents said disconnected data affected their ability to respond to customer issues.
As Ben Robinson, Head of Technology at Fair Finance, said during our CRM roundtable:
“Whenever I’ve seen these systems being set up, they’re set up with internal metrics. They’re not actually started with customer outcomes in mind.”
Compare repeat contacts and renewal outcomes for customers affected by missing information against a comparable unaffected group. Finance will find that more convincing than a falling duplicate count.
Identity Resolution and the Single Customer View: How Do You Test Accuracy?
Companies are having a hard time fixing customer identity gaps, whatever the tools promise. Australian retailer Oroton had roughly 34 million customer records in Twilio Segment yet recognized just 8% of shoppers. Consolidating them into 1.34 million profiles lifted its known-customer rate to 42%: a big improvement, and still not a single customer view.
Where Identity Resolution Goes Wrong
Deterministic matching uses verified identifiers; probabilistic matching connects on less certain signals. Two insurance customers sharing an email address is where that goes wrong: combine their records and you could expose sensitive information or interfere with a claim.
As Salesforce technical architect Apoorva Sharma warns:
“In regulated use cases like lending or insurance, a false match is usually more costly than a missed match.”
How to Test a Vendor’s Accuracy
Give the supplier verified records with shared households, changed contact details, and deliberately difficult matches, and demand two figures:
Demand two figures:
Ask to see an incorrect merge reversed, including how the correction reaches connected systems, and who approves uncertain matches. Then put the resulting view in front of an agent: can they find the right account and complaint history without an engineer? A supplier quoting a 95% match rate should be able to explain exactly what happened to the other 5%.
Real-Time vs Batch Customer Data: Where Does Speed Matter?
A payment made at 9:15 shouldn’t still show as overdue when collections starts calling at lunchtime. Yet plenty of data architectures are designed around how often systems can exchange information, with little attention to when someone actually needs it.
The question for real-time vs batch is simple: how old can a record get before it causes trouble? Fraud alerts and payment confirmations have almost no room for delay. A monthly retention report can wait until tomorrow, and processing every event live would be an expensive way to produce something nobody reads until Friday.
Amperity shows why “real time” needs unpacking: its live layer can recognize activity against an existing profile, while Stitch still rebuilds the underlying identity graph in batch. Our Amperity Customer Data Cloud analysis explains the difference.
Pricing is important to consider, too. Pradeep Krishna, Product Management Director at Salesforce, acknowledged the cost difference in October 2026:
“From a cost perspective, streaming ingestion incurs higher credit consumption than batch processing.”
That’s Salesforce Data 360 specifically, but ask any vendor what continuous updates will cost. And watch the performance figures: a payment can reach the CDP in two seconds, then sit in a queue for 40 minutes before the CRM gets it. The advertised ingestion speed was accurate. The customer still gets chased for money they’ve paid.
Make vendors test the 95th-percentile delay from transaction to usable record, including when a connection stalls, and make sure agents can see timestamps. If the payment data is three hours old, they deserve to know before they call.
Be more interested in what happens when an integration breaks than in how many connectors a vendor advertises. Two hundred supported applications won’t help if a failed update goes unnoticed for three days. There’s a growing argument that composable CX will replace monolithic CRM platforms, and I see the appeal of swapping specialist tools without replacing the CRM every time the business changes direction.
Someone still has to maintain the connections, though. Cordial’s November 2026 headless CDP launch, promising access to customer data through ChatGPT and Microsoft Copilot without working inside the CDP, is the direction of travel; I’d still want to see it working with real permissions and real records.
Be careful with “zero-copy” claims, too. Salesforce’s October 2026 architecture guidance explains:
“Query federation executes compute on both Data 360 and the external database layer.”
So you may be paying for processing in two places.
Before signing, have the supplier disconnect an integration and show how failed updates are detected and recovered, and check API rate limits and who maintains connectors. Then try leaving. Export customer profiles, consent histories and identity mappings, including the rules used to connect them. Be suspicious of any platform that lets you take your records but keeps everything that makes them useful.
How Are CDPs and Data Clouds Priced, and What Causes Overages?
CDP pricing is awkward because suppliers charge for different things, which is how companies end up with a retention CDP whose contract doesn’t cover enough customers to run the campaigns.
Twilio Segment uses monthly tracked users. Salesforce Data 360 charges consumption credits for activities including identity resolution and calculated insights. Even the definition of a billable customer deserves checking.
Salesforce architect Vishal Soni offers a useful warning:
“Identity resolution is the easiest one to underestimate.” “Identity resolution is the easiest one to underestimate.”
Change a matching rule and Salesforce may need to reprocess every existing record; refresh an insight hourly and it runs 24 times a day instead of once. Both eat the credit allowance.
A Magic Quadrant won’t tell you what you’ll spend, so ask vendors for a 36-month forecast at today’s usage, double, and five times, including implementation fees and the cost of exporting if you leave. Then ask what happens when you exceed the allowance: does processing stop, or are you charged automatically? A dashboard warning you about money you’ve already spent isn’t much help.
Governance and Consent: Who Controls Customer Data After Go-Live?
A customer changes their address in the app; six weeks later the replacement bank card goes to the old house because an overnight CRM sync restored it. Every system worked as configured. Nobody had agreed which one to trust.
Who gets the final say when records disagree? Give important fields a named owner, a data entry policy, and an authoritative system. Billing controls confirmed payment status; the CRM shouldn’t overwrite it with last Tuesday’s data.
Consent needs the same treatment. Marketing preferences, permission to record calls, and AI disclosures are different records with different rules, and bundling them into one “consent” field is asking for trouble. The data question is which system is the source of truth for each; the enforcement test across channels belongs in the security guide.
AI raises the same question. As John Cheney, CEO of Workbooks, asked us:
“At what point do we want to hand over control to agents completely versus how much human guidance we want?”
Reading a balance is one permission; changing payment instructions is another, with its own access controls, recorded actions, and approvals. And give someone responsibility for reviewing failed updates, outdated records, and data kept past its retention period. A governance policy nobody checks after deployment isn’t worth much.
Five Fast Wins for Better Customer Data Management
Start with the records behind your most embarrassing complaints. There’s enough work there for 90 days:
-
Reduce duplicates. Find customers with several account numbers and one active contract, and check which record holds the valid entitlement before merging, so nobody loses a warranty.
-
Build a single preference model. A customer asks for paper statements after a fraud incident; make sure that survives the next CRM sync rather than being overwritten by the app.
-
Create a “CX-ready” segment library. Start with a suppression list for customers awaiting compensation. Nobody needs a loyalty campaign mid-argument about a missing delivery.
-
Improve contact center context. Put replacement shipment tracking beside the original complaint, then count how many agents still transfer customers to the warehouse team.
-
Make AI less risky. Test an AI agent with two similar order numbers and an account under fraud review. It should stop the refund and ask a human.
Measure the complaints these changes eliminate. That’s stronger evidence for your next CRM vs CDP vs data cloud investment than another dashboard showing fewer duplicate records.
What Should a Better Customer Data Strategy Actually Deliver?
There’s a particularly frustrating kind of technology investment where the new system works beautifully, yet customers are still complaining about exactly the same things. Before another CDP contract goes through, investigate one of those complaints properly.
Find the missing payment or incorrect delivery record, trace it through the systems involved, and establish who’s responsible when the information goes astray. You might already own everything needed to put it right.
If a new platform earns its place, insist on a demonstration using your records. Keep the matching results and a record of how long updates take. Get the expected consumption into the contract as well, because the renewal discussion will be considerably less pleasant without those figures.
Customers have better things to do than explain a missing payment for the third time. Whether you’ve unified two million profiles won’t interest them much.
FAQs
Do I Need a CDP if I Already Have a CRM?
It depends on how badly your existing systems recognize customers. Someone shopping online and in-store shouldn’t necessarily appear as two separate people, and a CDP could help with that. An agent opening another application to find an invoice is a different headache. Fixing the CRM’s billing connection could be money better spent.
What’s the Difference Between CRM, CDP, MDM, and a Data Warehouse?
CRMs handle the relationships and cases employees work on. CDPs try to recognize customers across different applications. MDM deals with the awkward business of deciding which conflicting records are authoritative. Warehouses store and process information for reporting and analysis. The boundaries aren’t especially tidy anymore, particularly with broad platforms such as Salesforce.
How Can I Test Customer Identity Resolution Accuracy?
Ask what happens when two customers are incorrectly merged. Can the vendor separate them again? Will the correction reach every connected system? Use verified records to test precision and recall. I’d be especially wary of a supplier presenting its profile consolidation rate as proof of accuracy.
Do I Need Real-Time Customer Data for Everything?
Consider a delivery address changed just before an order ships. Yesterday’s address won’t help the courier. Monthly customer retention reporting has a completely different deadline. Paying for continuous processing there deserves some explanation from the vendor.
Yes, and it’s worth trying. A properly connected CRM and warehouse may already have the capabilities you need. Start with the desired outcome, decide which records take priority, sort out identity and data quality, and establish ownership rules. Test everything on a customer journey that regularly goes wrong. If something’s still missing, you’ll have a reason to consider another CDP.
