There is a model of the AI modernization story that goes: construct the platform, then determine the use circumstances. Ankur Jain would let you know that is backwards — and that almost all organizations are studying that the onerous approach.
Ankur is Chief Cloud and Information Modernization Officer at Acxiom, the linked knowledge and know-how basis that helps world manufacturers resolve buyer identification throughout channels, enrich buyer profiles with greater than 10,000 attributes, and ship outcomes throughout buyer acquisition, retention and personalization.
Ankur leads each product engineering and client-facing options engineering — that means he’s accountable not only for what Acxiom builds, however for the way these capabilities get embedded contained in the environments the place shoppers really function.
After becoming a member of the corporate lower than two years in the past, Ankur led the modernization of Acxiom’s core infrastructure, knowledge pipelines, legacy structure and underlying tech-stack. Right now, Acxiom is actively constructing agentic workflows that automate the total advertising worth chain.
Why the Basis Has to Come First
Aly McGue: Plenty of organizations wish to transfer to agentic AI however are nonetheless operating core workloads on legacy infrastructure. What’s the threat of attempting to construct intelligence on prime of a basis that wasn’t designed for it?
Ankur Jain: The chance is that you simply hit a ceiling nearly instantly. Once I joined Acxiom, each merchandise and shopper options had been hosted largely on-premises. When your merchandise and options are constrained to a knowledge middle, they’ve restricted scalability. Efficiency was less than par for the real-time use circumstances shoppers had been asking for. After which there was quite a lot of legacy tech — the stack wanted a refresh, a reimagining of what cloud-native structure might seem like.
What we additionally noticed was quite a lot of handbook pipelines, quite a lot of knowledge redundancy, copies of the identical knowledge in a number of locations. The method itself was not very environment friendly. Any group attempting to construct agentic capabilities on a fragmented or legacy basis goes to spend extra time managing infrastructure than constructing merchandise.
For us, the strategic imaginative and prescient comes down to 2 north stars: knowledge modernization and agentic advertising. They’re sequential, not parallel. You can’t construct an agentic advertising ecosystem on a legacy basis.
How an information warehouse migration shifted the main focus from upkeep to enterprise outcomes
Aly: You moved from on-premises Hadoop to Databricks. What did that shift make potential that wasn’t potential earlier than?
Ankur: By way of efficiency, now we have seen enchancment throughout the board, throughout various kinds of workloads and various kinds of pipelines, nearly 80 to 90 p.c sooner run occasions. Workloads that used to take 50+ hours, generally 90+ hours — and I am speaking hours, so actually days, generally as much as per week — at the moment are getting carried out inside 2-3 hours. Those self same workloads, in 2-3 hours.
It has additionally freed up our individuals. In some circumstances now we have been in a position to liberate a number of full-time roles to focus extra on value-added outcomes relatively than managing infrastructure. The primary factor it enabled was for the engineering group to focus extra on enterprise outcomes relatively than worrying in regards to the infrastructure beneath. Which may sound like a tender win, however when your engineers are spending their time constructing merchandise and delivering shopper options relatively than retaining the lights on, it modifications what you possibly can even try.
What the Agentic Advertising Worth Chain Truly Appears Like
Aly: The place are you seeing agentic AI reshape precise advertising workflows as we speak, and the place does that imaginative and prescient prolong?
Ankur: Acxiom’s core operation could be very data-centric. We herald advertising knowledge from a number of platforms — CRM, e-commerce, Adobe Analytics, Google Analytics — and assist manufacturers construct a holistic buyer view, enrich it, and ship outcomes. Historically, that required a group of information engineers and knowledge architects who would mannequin the whole lot and construct pipelines manually. ETL is at all times the longest pole within the tent, and it might take months.
By way of AI, that whole cycle compresses. Code technology by means of prompts, automated testing of outputs, accelerated CI/CD pipelines. On the advertising aspect, producing completely different variations of an advert used to take artistic companies months. Now you possibly can analyze adverts at scale by means of machine studying, feed these outcomes into an AI engine and generate extremely personalized variations in minutes.
The place now we have seen the largest actual shift is on execution. Take viewers planning — a marketer passes a immediate describing a marketing campaign goal and goal profile, and the agent builds the viewers segments with pattern personas utilizing Acxiom knowledge, surfaces completely different demographic and behavioral dimensions and lets the marketer refine from there. What used to take effort from a number of individuals with diverse ability units and quite a lot of lead time is now carried out agentically in minutes. We’ve demonstrated the identical sample for media shopping for: an agent queries obtainable stock, evaluates it, makes a shopping for resolution and prompts the audiences throughout channels.
The aim is to attach the whole pipeline — from viewers design by means of media shopping for, activation and efficiency analytics — into an agentic framework. That complete AI for BI functionality that Databricks is constructing by means of the Genie and agentic ecosystem is strictly the place advertising workloads like ours are heading. It could actually all be put to work end-to-end.
How governance accelerates agentic workflows
Aly: Acxiom operates in extremely regulated industries, and deploying brokers requires a excessive stage of belief. How does that form the best way you design governance into agentic workflows?
Ankur: The info we deal with spans PII, so each agentic workflow we construct begins with privateness as an architectural precept.
In observe, meaning AI-generated content material by no means goes immediately right into a dwell marketing campaign. It routes by means of an approval workflow the place authorized critiques artistic and messaging earlier than something reaches a buyer. The brokers function inside outlined boundaries, with safety and privateness controls baked into the pipeline, and people keep within the loop at each resolution level that carries regulatory or model threat. The aim is to not gradual issues down. It’s to ensure pace doesn’t come at the price of belief — for the shopper, the model or Acxiom.
Embedding AI into advertising merchandise and workflows
Aly: What does it imply for Acxiom’s merchandise to be AI-native, and the way does that change what shoppers really expertise?
Ankur: AI-native means intelligence is embedded throughout the whole advertising worth chain: ingesting first-party knowledge, resolving buyer identification, enriching profiles with Acxiom’s knowledge property, constructing viewers segments, planning media buys, activating campaigns throughout channels and feeding efficiency analytics again into the following cycle. Every of these steps can now be AI-driven relatively than manually orchestrated.
For shoppers, the largest change is transparency. Historically, quite a lot of what we offered operated as a black field. Manufacturers despatched knowledge in, outcomes got here again, and the logic in between was opaque. Now those self same capabilities may be delivered collaboratively, contained in the platforms shoppers already use, with full visibility into how selections are being made. That’s what shoppers are asking for: meet them the place they’re, function of their setting and make the method clear.
And it’s a forcing perform that comes not solely from throughout the group, however from our shoppers immediately. They’re asking us: how are you going to make it more cost effective? How are you going to make it extra performant? How are you going to make it sooner? If you wish to reply these questions truthfully, you must herald AI.
Proprietary Information because the Aggressive Moat
Aly: Your knowledge property are core to what Acxiom sells. How is the best way you ship that knowledge to shoppers evolving, and what does that unlock?
Ankur: Acxiom helps shoppers benefit from their buyer knowledge. We assist them put it to work and monetize it. We offer knowledge property that manufacturers in any other case wouldn’t have, throughout automotive, retail, healthcare and pharmaceutical. Traditionally, delivering that knowledge was by means of conventional means — by means of SFTP. A model would request enrichment, we’d enter right into a contract and ship the recordsdata. That was the outdated approach.
Now we’re embedding our knowledge in an agentic style, both in our personal platforms or immediately within the shopper’s setting. We companion with main martech platforms the place our knowledge property are natively obtainable. If a shopper is constructing their very own AI platform, we will combine agentically to allow them to make a name to our property and serve them up immediately. We’re additionally growing clear room options in partnership with Databricks, the place shoppers can combine with Acxiom knowledge in a privacy-safe method inside their very own ecosystem.
The manufacturers we work with perceive that first-party knowledge is their most respected asset. Information privateness performs a vital function whereas dealing with and processing this knowledge. Manufacturers wish to train higher management and are always in-housing the advertising capabilities. The expectation is shifting for companies to work inside manufacturers’ platforms and governance frameworks. The companies that may function and ship outcomes natively into that setting will likely be indispensable.
Deal with It as a Basis Downside, Not a Instruments Downside
Aly: In the event you had been chatting with a C-suite peer simply starting to scale their AI efforts, what is the one factor you’d need them to listen to?
Ankur: Ensure the muse is stable. There may be quite a lot of AI buzz, which is not a buzz anymore; it is actuality. However what makes or breaks the entire AI initiative is the muse that it wants to take a seat on. In our case, shifting from on-premises to the cloud was not solely an ambition. Conserving the long run in thoughts made it a necessity in order that we may very well be an actual participant within the AI journey. Stable knowledge basis, cloud-native structure, knowledge governance and safety — these are the important thing elements. Any group that skips that step goes to search out out ultimately that it wasn’t elective.
The sample at Acxiom is a helpful body for any govt evaluating the place to place their power. Modernizing the muse and pursuing agentic AI aren’t two separate applications competing for price range and a focus. They’re the identical wager, made in sequence. Get the information layer proper, show worth by means of centered pilots, then embed your differentiated capabilities the place shoppers really need them.
The shift Ankur describes — from delivering knowledge by means of file transfers to embedding intelligence natively inside shopper environments — is not simply an architectural improve. It modifications what sort of firm Acxiom is. That form of repositioning would not occur by bolting AI onto an on-premises stack. It requires the muse to come back first.
Discover how over 25 trade consultants and 1,200+ leadership-level survey respondents are paving the best way for profitable AI deployment by accessing the “Making AI Ship” report from Economist Enterprise, created in partnership with Databricks.
