The most important thing Treasure AI owns is not an artificial-intelligence model. It is memory. More precisely, it is the patient, unglamorous machinery required to remember that the person who browsed a jacket on a phone, opened an email on a laptop and bought the jacket in a shop is probably the same customer - and that the company is allowed to act on that knowledge.
That problem has occupied Treasure Data since 2011. The Mountain View company built pipes for collecting data, storage for keeping it, identity tools for matching it and controls for using it. In April 2026, it changed its business name to Treasure AI and widened the pitch. The old customer data platform waited for a marketer to ask a question. The new “agentic experience platform” is meant to notice, decide and prepare an action continuously, with a person setting the goal and approving what goes live.
A rebrand can be the corporate equivalent of changing a restaurant menu while keeping the same freezer. This one is more revealing. Treasure Data remains the legal name, and the CDP remains the foundation. But Treasure AI Studio now presents that machinery through a conversation across web, desktop, mobile and command line. A marketer can upload a retention brief and ask for segments, decision points, content and a journey. An engineer can use Treasure Code to describe a workflow, inspect the generated code, put it through peer review and roll it back if necessary.
The customer is hiding in the joins
Enterprise customer data is rarely neat. A carmaker may have dealership records, configurator visits, app activity, test-drive forms, service histories and television response data, all created by different teams. A bank may have loans, transactions and marketing responses in separate systems. Treasure AI ingests batch and streaming data, resolves identities, builds a governed profile and makes that profile available for analysis and activation.
This is the problem its customers are paying to solve: not a shortage of data, but an excess of disconnected evidence. The buyers are large organizations in automotive, retail, consumer goods, finance, entertainment, healthcare, technology and travel. Publicly discussed users include Subaru, AB InBev, Nestlé, Shiseido, Universal Music Group, Six Flags, Michaels and Extraco Banks. Marketing teams want usable audiences. Data teams want consistent definitions. Security teams want permissions and audit trails. Finance wants fewer overlapping tools.
Subaru offers the most vivid example. Its team describes modern car shopping as an “invisible tournament”: much of the contest is over before a buyer visits a dealer. By connecting online behavior, dealer information, apps and campaign activity around a Subaru ID, the automaker could see more of that hidden contest. It reported making the same television advertising budget 14.5 times more effective by learning which time slots and creative work produced response, rather than assuming audience size alone predicted value.
“AI without trusted data is just noise.”Rafa Flores · Chief Product and Growth Officer
One loop, no weekend babysitting
Treasure AI organizes its product story around a loop: collect, unify, understand, decide, engage, then collect the response and begin again. A conventional CDP moves through that cycle at human speed. Someone requests a segment, someone else checks the query, a campaign is scheduled and the results arrive for next month’s meeting. Agents can keep parts of the loop turning between meetings.
The response becomes the next signal. Agents increase the clock speed; people still choose the destination and the guardrails.
The product family maps onto this cycle. The Intelligent CDP handles ingestion, storage, identity resolution, segmentation, predictive models and governance. AI Agent Foundry lets teams build agents grounded in approved data and expose them through the console, a webhook or Slack. Treasure AI Studio gives nontechnical users a conversational surface. Engagement, Personalization, Creative, Paid Media and Service suites carry decisions into email, SMS, mobile push, LINE, websites and other channels.
Marketing Super Agent sits above specialist agents for research, sentiment, personas, campaign concepts, audiences and execution. The name sounds like a cape is included; the practical idea is an orchestrator that breaks a brief into smaller jobs. Each job passes context to the next, while outputs remain visible for review. Treasure Code brings the same approach to technical operators, who can manage workflows and CDP configuration as code instead of clicking through a console.
The moat is permissioned context
The CDP market is crowded. Adobe and Salesforce bundle customer data into broad enterprise suites. Tealium, Twilio Segment, Amperity, mParticle and BlueConic specialize in customer data. Hightouch and other warehouse-native tools argue that businesses should activate data where it already lives. Braze and Iterable approach the problem from customer engagement.
Complete when needed
Treasure AI can supply ingestion, storage, identity, segmentation, decisioning and activation as one managed platform.
Composable when useful
Its control plane can also work with data in Snowflake, Databricks, BigQuery or an existing warehouse.
Treasure AI’s distinction is the combination of a mature, large-scale data foundation with an agent interface and its own activation products. A generic assistant can draft an email. It cannot safely decide which customers should receive it unless it has current identities, consent, business rules and channel access. Treasure AI is selling that permissioned context, plus the wiring to do something with it.
The company says every agent can be limited to particular attributes, segments or tables, with role-based access and auditability. Its AI Agent Foundry is built with Amazon Bedrock, and the wider platform runs on AWS. Treasure AI has also earned a responsible-AI certification from TrustArc. Those facts do not make automated decisions infallible. They show where the company thinks enterprise buyers will draw the line: agents may move quickly, but they must leave footprints.
A business built for large commitments
Treasure AI is enterprise SaaS. It sells subscriptions, usually through negotiated and multi-year contracts, with add-on activation suites, implementation and professional services. Treasure AI Studio is available to customers, while conversations consume usage-based AI Credits. The product portfolio is also available through AWS Marketplace, allowing buyers to use existing cloud procurement and billing relationships.
That model favors organizations with complicated data and enough potential return to justify integration work. It also explains the “trade-up” offers aimed at replacing a CDP, email provider or engagement platform. The company is not merely competing for a line in the AI budget. It is trying to consolidate pieces of the marketing stack and own the control layer connecting customer memory to customer contact.
Its corporate history prepared it for long enterprise cycles. Kazuki Ohta, Hironobu Yoshikawa and Sadayuki Furuhashi founded Treasure Data in 2011. Furuhashi created Fluentd, the open-source data collector that became a widely used part of cloud infrastructure. Arm acquired the company in 2018 for a reported $600 million and folded its technology into an internet-of-things platform. Treasure Data later spun out under SoftBank, then raised $234 million in 2021 from SoftBank Corp. and Carbide Ventures.
The current company is therefore both a veteran and a restart. It has the scars of databases, integrations and global privacy requirements, but it is presenting itself in a category whose vocabulary is still unsettled. In July 2026, IDC named Treasure AI a Leader in evaluations for both B2C and B2B users of AI-enabled customer data platforms. That places it among established marketing clouds and independent CDPs while the whole field races to convert passive records into supervised action.
What a marketer can steal
The useful idea here is not “add AI.” It is to shorten the distance between evidence and action without removing review. Start with a narrow loop: a churn signal, an audience definition, a message, an approval and a measured response. Make the data and permissions explicit. Let software repeat the tedious parts. Keep a person responsible for the objective, the exceptions and the moment a customer stops looking like a row in a table. A smaller loop also produces cleaner evidence. Teams can see whether the agent found a useful audience, whether the message worked and where human judgment changed the result before attempting to automate an entire department.
Treasure AI’s future depends on whether it can make that loop faster without making brands careless. Two billion profiles create a lot of opportunity and an equally large obligation. The company’s name now promises intelligence. Its older name is still the reason that promise might hold: before an agent can act for a company, someone has to do the hard work of knowing who the customer is.