The appeal of an “AI-ready customer data platform” is easy to understand. Organizations want more joined-up customer journeys, faster service, better recommendations and a clearer path into agentic AI, and the market has no shortage of platforms promising to deliver them.
But a convincing demonstration is not the same as an enterprise-ready product. Richard Manthorpe, Product Director at Content Guru, explained to CX Today.
“It’s easy for a startup to download a model from Hugging Face and throw it into a demo. The hard part is productizing that.”
The practical work for buyers starts with looking beyond the front-end experience. Can the platform work with existing data and systems? Can it support governance? Can it prove value? Will it remain useful as AI models and business requirements change?
Here are five questions to ask vendors before committing.
1. Can It Prove A Business Outcome, Not Just Demonstrate An AI Capability?
The first question should be the simplest: what is the platform expected to improve?
The early AI market encouraged organizations to buy first and work out the value later, but that approach is becoming less acceptable, Manthorpe noted.
“At the start it was, ‘I’ve got a million dollars to spend on AI to show shareholders we’re innovating. What does it do? It doesn’t matter – just make sure it involves AI .’ Now, a lot more organizations are focusing on the outcomes and the value, and rightly so.”
Buyers should ask vendors to explain the end-to-end operational journey beyond the AI component, including reporting, billing, adoption, governance, and measurable outcomes.
“Do I have the right reporting? Can I prove that return on investment? Do I have the governance around that? Can I prove to my compliance teams that we are doing the right thing and that we’re getting the right outcomes?”
Cost control belongs in the same conversation. AI consumption can quickly outpace expectations if it is not measured and governed.
“Do I have the information on the cost of tokens so that I don’t get bill shock at the end of the month when I’ve used my monthly budget in the first three days?” Manthorpe said.
The technology should support a defined service or business outcome. If the buyer cannot describe that outcome, and measure it, the project risks becoming what Manthorpe calls “a cool pilot that we can make a press release about,” but one that “never delivers the value”.
2. Can It Connect Customer Context Across Existing Systems And Channels?
No enterprise has a perfectly tidy data estate. CRM, case management, billing, web journeys, contact center records, policy systems, operational databases and other data sources all hold part of the customer story.
“Fragmented data is not abnormal,” Manthorpe said, noting that an agent uses 13 different systems on average. “That’s always going to be the case; it’s about how you can tie those up.”
The question is whether a CDP can connect the information needed to improve a live journey without demanding that every system is replaced, or every record is moved to one place.
Buyers should test how the platform integrates with existing systems, whether it supports open standards, and whether it can be adapted as the organization adds new tools. Manthorpe pointed to the importance of data accessibility.
“Do I have those open standards that allow me to connect to the systems I need? I don’t want to be locked into a fixed set of connectors or dependent on specialist resources every time I need to make a change. I want the flexibility to do it myself.”
This is key for omnichannel service, as customers quickly become frustrated if they need to repeat basic information when they move from self-service to a human agent, or from messaging to voice.
The customer data layer should support continuity, forming a well-connected journey in which information gathered before the agent interaction is available at the point it is needed. “That’s making the journey as frictionless as it possibly can be. The data layer underneath makes that happen,” Manthorpe said.
3. Is Our Data Accurate Enough for AI To Use Safely?
AI can only make decisions from the information it can access. If the data is incomplete, stale, duplicated, or poorly connected, the outcome may be irrelevant at best and harmful at worst.
Buyers should ask how the platform handles duplicate identities and conflicting records across systems, to pick up when a customer uses one email address in an app and another on a website.
The aim is to “make sure you’re bringing those two records together and then proceeding with the journey, but doing that in a frictionless way rather than a complicated data cleansing activity,” Manthorpe said.
4. Can We Govern What Humans And AI Are Allowed To See And Do?
Beyond data access, an AI-ready platform needs clear, enforceable rules about what users, systems, and AI services are permitted to see, recommend, and do.
This raises questions around compliance and data sovereignty, Manthorpe said.
“Do I have GDPR controls? Can I make sure that I’m not storing or using data that I shouldn’t? Do I have the right audit logs? Is my data staying where it should be from a data sovereignty perspective?”
The controls need to apply to AI as well as people. “Make sure you’re building the right role-based access controls,” Manthorpe advised. “Having that field– and role– level security, having the ability to define those roles and who should have access to those different data, both human and AI.”
In addition to meeting compliance requirements, this governance provides the basis for new AI services to be approved. “Once that’s in place, security and legal teams have a much better footing to approve those new types of services.”
5. Will The Platform Still Support Us As The AI Market Changes?
Given how quickly AI is developing, a platform that is useful today can become restrictive if it is tied to a single model, fixed connector set, or proprietary architecture.
Buyers should ask whether they can adapt AI services over time, Manthorpe noted.
“Not being tied into a particular AI model, not having controls and proprietary connectors within that, and making sure that it’s an open framework to be able to go in and adapt is key.”
As automation takes on more front-office activity, human teams may spend more time managing exceptions, reviewing AI-led work or handling back-office tasks that now sit closer to the customer journey.
“The tools that they need are changing with them,” Manthorpe said. “The CDP is then providing that information to a traditional back– office worker as if it was a customer interaction.”
The right platform should help coordinate the changing mix of human work, customer interactions, and AI-supported processes.
The key is to test whether the product behind it can deliver in an enterprise environment.
Buyers should look for “proven deliverables” and a track record that can give them confidence “that the business is still going to be there next year,” Manthorpe said.
The best starting point remains the most grounded one that understands the outcome, checks the data and governance foundations, and ensures the platform can adapt.
“Make sure you focus on outcomes,” Manthorpe said. “Make sure you understand the benefits of what you’re trying to do and then build your ROI model and measure against that.”
