From weeks to minutes: How Components 1® makes use of agentic AI on AWS to speed up information operations

Components 1^® (F1) engages an viewers of over 800 million followers globally throughout digital platforms, F1 TV, social media, ticketing, and merchandise year-round. Races occur each two weeks. Fan engagement home windows are measured in minutes and industrial choices want to maneuver on the velocity of the grid. Behind the scenes, F1’s advertising and marketing know-how (MarTech) platform, Buyer 360, captures interactions throughout all of those touchpoints to energy personalization, segmentation, and industrial technique.

Nevertheless, the platform confronted a big operational problem. Based on Chris Roberts, Director of IT at Components 1, “Our MarTech platform is the nervous system of F1’s fan engagement. However each new information supply required 6 to eight weeks of guide engineering. We had an 18-month backlog simply to combine 12 new sources.” The enterprise was producing information quicker than the engineering staff may wire it up. Consequently, Matt Kemp, F1 Head of Knowledge Operations, set to enhance efficiencies and information high quality. “Manually ingesting information sources is time consuming, creates answer variances, and finally ends in information integrity points. I wished an answer that was repeatable, sturdy and dependable. AWS labored backwards from our must implement an agentic answer that labored finish to finish, making use of enterprise logic at every step.”

In early 2026, F1 and AWS labored collectively to construct the Knowledge Accelerator, an answer that makes use of agentic AI on Amazon Bedrock AgentCore to rework F1’s MarTech information platform from a manually maintained system right into a self-managed, observable, and unified information property. On this submit, we present how the Knowledge Accelerator diminished information supply onboarding from as much as 8 weeks to roughly 40 minutes of code technology plus hours of deployment. It additionally recognized and glued information supply anomalies in manufacturing, tracked information platform operations and agent lineage in a single window, and opened a gateway for analysts, engineers, and scientists to collaborate. “For the primary time, we now have end-to-end visibility throughout your entire MarTech platform with information lineage and root trigger evaluation, not simply dashboards filled with alerts,” says Roberts.

The problem

F1’s Buyer 360 platform ingests information from ticketing companions, streaming integrations, sponsor activation feeds, social media, and merchandise methods. Working an information property of this breadth and velocity surfaced three areas of friction the staff got down to clear up. First, onboarding every new information supply was a closely guide effort: engineers wrote schema mappings, constructed ingestion pipelines, configured information high quality checks, outlined Normal Knowledge Safety Regulation (GDPR) classifications, and set governance insurance policies by hand. This course of took 6 to eight weeks per supply. Second, the platform needed to maintain tempo with always evolving upstream feeds. Suppliers regularly modified column names, added fields, or restructured and rescheduled payloads with out discover. These modifications usually surfaced on the worst attainable second, corresponding to mid race-weekend or throughout a mission-critical marketing campaign launch. Third, visibility was fragmented. Logs had been scattered throughout companies with no unified information lineage. When a stakeholder questioned a metric, engineers spent hours manually tracing the problem throughout Amazon Easy Storage Service (Amazon S3) paths, Amazon Redshift management tables, Airflow logs, and DBT outputs.

Answer overview

The Knowledge Accelerator addressed these challenges by way of 5 workstreams delivered concurrently:

• Agentic information supply onboarding utilizing Amazon Bedrock AgentCore, internet hosting brokers in its runtime containers.
• Automated schema evolution detection and remediation.
• Unified information entry by way of Amazon SageMaker Unified Studio.
• Finish-to-end observability with root trigger evaluation instrument (RCA) and context graph.
• Automated identification of a failure in observability dashboard and agentic operation in the event that they could possibly be mounted with code modifications.

A sixth workstream optimized the client identification decision algorithms that unify fan touchpoints throughout channels. The next sections describe every workstream intimately.

Agentic information supply onboarding

The centerpiece of the Knowledge Accelerator is a set of platform brokers that take a Enterprise Necessities Doc (BRD) with restricted details about the information supply and produce a totally production-ready onboarding pipeline. This consists of infrastructure code, information transformations, governance insurance policies, and GDPR classification with out a human writing a single line of boilerplate. The brokers work in two phases:

Part 1: Configuration technology

When a brand new information supply wants onboarding, a staff member uploads a BRD to an Amazon S3 bucket. The add triggers an AWS Lambda perform, which invokes Amazon Bedrock AgentCore Runtime, a functionality of Amazon Bedrock AgentCore. The agent reads the BRD and generates a set of configuration information. It then accesses GitHub by way of a GitHub App to push these information as a pull request to the standardized Git repository, and accesses Jira by way of its REST API to create a ticket referencing the PR. All agent conversations and actions are traced in Amazon CloudWatch by way of built-in AgentCore observability. The assigned engineer opinions, adjusts if vital, and approves.

Part 2: Full pipeline technology

As soon as the configuration information are accredited, a human triggers the subsequent stage. The agent takes the accredited configuration and generates three separate Pull Requests:

1. AWS Glue software and infrastructure code.
2. DBT transformation framework.
3. Governance insurance policies together with GDPR tagging.

All three PRs hyperlink to a single Jira ticket for traceability. Engineers assessment every one throughout the Infrastructure, DBT, and Governance repositories and approve.

Automated GDPR classification

What distinguishes this from a primary code generator is the built-in GDPR classification. The agent proactively analyzes each information column, determines whether or not it accommodates private information, delicate private information, or pseudonymized information, and tags it with the suitable GDPR class. These tags publish on to the governance registry in SageMaker Unified Studio, giving the compliance staff speedy visibility with out guide assessment cycles.

Modular talent structure

The system will not be a tightly coupled agent graph. A single agent operates with modular talent definitions, every encapsulating a definite functionality: schema mapping and information sort inference, information high quality validation, governance enforcement, and delicate information classification. At runtime, the agent evaluates incoming necessities and prompts the related abilities, composing them by way of a multi-pass reasoning course of. Move-0 handles token administration by way of scrubbing, Move-1 summarizes instrument outputs, and Move-2 rolls up an general evaluation, refining accuracy and completeness progressively moderately than counting on a one-shot response. New capabilities ship as new talent modules with out altering the core agent loop, preserving the structure maintainable and composable because the platform grows.

The result’s onboarding time dropped from 6 to eight weeks to roughly 40 minutes of code technology plus hours of deployment and assessment. AI brokers now deal with 95% of the work autonomously.

Automated schema evolution

Onboarding new information sources is one problem, however preserving present integrations wholesome is one other. Upstream suppliers regularly modify their information buildings, from renaming a column to creating a brand new discipline. Beforehand, the F1 staff found these modifications when a pipeline failed, usually throughout a dwell race weekend. The identical agent structure that handles onboarding now repeatedly displays for upstream schema modifications. When a supplier modifies their information construction, the agent detects it by way of event-driven triggers utilizing AWS Lambda and Amazon EventBridge. It assesses the downstream affect, figuring out which pipelines are affected, and which shoppers rely on the modified fields. It then generates the required code updates throughout all affected repositories and creates a Jira ticket with full context and linked PRs. Engineers obtain a notification that explains what modified, describes the affect, and presents a proposed repair for assessment. Finish-to-end decision now takes hours as a substitute of days.

Unified information entry with Amazon SageMaker Unified Studio

Earlier than the Knowledge Accelerator, working with Buyer 360 information required navigating a number of disconnected environments. Knowledge engineers curated pipelines in a single account. Knowledge scientists who wished to mannequin fan habits wanted entry to a separate account, and analysts operated in a 3rd world fully. No one shared tooling or context, and getting from a query to a solution took days of coordination earlier than any evaluation may start.

The answer makes use of Amazon SageMaker Unified Studio as the inspiration for an information mesh framework the place a central governance account brokers information discovery and entry throughout a number of producer groups. The important thing enabler: governance is codified as declarative configuration, not guide console operations. A single information supply definition concurrently publishes information to the catalog and provisions the entry management wanted for shoppers to subscribe. This implies brokers can safely onboard new information merchandise end-to-end, from storage to catalog to ruled entry, as a result of the framework enforces safety constraints by building. No human must assessment IAM insurance policies or AWS Lake Formation grants. The platform ensures correctness structurally. That is what makes the “one entrance door” attainable.

Knowledge engineers curate and govern datasets in a single place, and information scientists discover those self same datasets in the identical surroundings: ruled, documented, and able to mannequin. A knowledge scientist constructing a fan segmentation mannequin or optimizing the client identification algorithm doesn’t must know the place the information lives, who owns the pipeline, or which S3 prefix to make use of. They open Unified Studio, discover the curated Buyer 360 datasets, and begin modeling. They get shared notebooks, constant tooling, and ruled entry, as a result of declarative governance made secure self-service attainable with out sacrificing management. The curation and the consumption lastly dwell facet by facet.

Finish-to-end observability with RCA and context graph

A knowledge platform is barely as reliable because the staff’s means to reply one query: is the information appropriate proper now? Earlier than the Knowledge Accelerator, answering that query meant logging into Apache Airflow, checking Amazon S3 paths, querying Amazon Redshift management tables, and studying DBT logs. “No one had the complete view. When a stakeholder requested, ‘why does this quantity look fallacious?’ the reply was at all times, ‘give us a number of hours.’ The observability dashboard modifications that fully,” provides Roberts.

The observability layer presents full information lineage from S3 Uncooked ingestion by way of Processed layers into Amazon Redshift DBT phases as a single interactive graph, color-coded for well being. Customers click on on any node to drill all the way down to particular person sources and tables, every exhibiting go/fail standing, final run time, and length. If a pipeline fails, the lineage visualization exhibits precisely the place the break occurred, and which downstream information is affected.

Root trigger evaluation (RCA) is an agentic instrument inside F1’s platform that reads system logs and identifies failure factors throughout the information property. By itself, RCA can inform you what failed. We increase the RCA instrument by passing by way of enterprise context and system topology, codified as JSON. A lacking file in S3 may be the error, however with the context graph, RCA tells you that the upstream supplier rescheduled their supply window, which is why the file wasn’t there when the pipeline ran. That’s the distinction between realizing what failed and understanding why.

For the primary time, F1 has full lineage, causal root trigger evaluation, and enterprise context definitions in a single place, and the dashboard auto-refreshes each quarter-hour.

Buyer identification decision

The ultimate workstream optimized the algorithms that resolve buyer identification throughout F1’s fan touchpoints. A single fan would possibly work together by way of the app, purchase tickets on the web site, watch on F1 TV, and have interaction on social media. Unifying these interactions right into a single identification with out false merges or missed matches is what makes efficient personalization attainable inside the Fan Personalization Platform (FPP).

F1 already had a working identification decision course of, nevertheless it was sluggish and struggled to scale with the rising quantity of fan interactions throughout channels. Moderately than re-architecting the pipeline or changing elements, the staff centered on optimizing the present decision algorithm’s computational efficiency. By profiling execution bottlenecks and tuning the matching logic, the engagement diminished processing time by 50%, whereas preserving your entire decision pipeline and its downstream integrations absolutely intact. No processes had been modified, no accuracy trade-offs had been made: the identical algorithm now runs in half the time at F1’s manufacturing scale.

With quicker decision, F1 can onboard any new information supply and collect new buyer information in half the present time. Sooner decision means more energizing unified profiles, which in flip means extra well timed and related personalization throughout each advertising and marketing channel.

“The entire level is to ship the suitable message to the suitable fan on the proper time, whether or not that’s by way of e mail, the F1 app, ticketing, or social. Now that we will onboard sources in hours and resolve identities quicker, we will really ship the customized experiences our followers count on throughout each advertising and marketing channel,” says Kemp.

Safety and governance by design

The Knowledge Accelerator operates on the precept that AI proposes and people assessment. The brokers run on Amazon Bedrock AgentCore with long-term reminiscence, retaining context throughout invocations. Growth used Kiro for structured spec-driven growth and Amazon Bedrock (Claude Sonnet 4.6) as the inspiration mannequin. The event-driven spine makes use of AWS Lambda for compute, Amazon EventBridge for routing, Amazon Managed Workflows for Apache Airflow (MWAA) for workflow orchestration, and Amazon S3 because the uncooked information layer. All AI mannequin entry is ruled by way of F1’s AI Gateway for unified entry management, value administration, and audit logging. However the structure is barely half the story. The safety posture is what makes this production-ready.

The safety posture consists of:

• Least privilege: fine-grained permissions, short-lived tokens with one-hour expiry, entry restricted to particular repositories and sources.
• Full audit path: each motion is logged and attributed for compliance.
• Human assessment: each generated Pull Request goes by way of engineer approval.
• Automated testing: brokers generate complete exams for their very own modifications.
• Rollback capabilities: points surfaced post-merge will be reverted instantly.
• Community isolation: your entire system runs inside non-public subnets in Amazon Digital Personal Cloud (Amazon VPC) with no direct web entry.
• Encrypted credentials: all secrets and techniques saved at relaxation in AWS Techniques Supervisor Parameter Retailer.

“What gave us confidence to place agentic AI in our manufacturing information pipelines was what we name ‘Human on the helm.’ The brokers do the heavy lifting, however people make the choices. Each change goes by way of the identical assessment course of our engineers already use, so adoption was speedy.” says Roberts.

The affect

The Knowledge Accelerator delivered measurable affect throughout F1’s MarTech operations:

• Knowledge supply onboarding: diminished from 6 to eight weeks to roughly 40 minutes of code technology plus hours of deployment and assessment.
• Autonomous work: AI brokers deal with 95% of onboarding duties with out human intervention.
• Time-to-value: roughly 99% discount.
• Schema evolution: end-to-end decision in hours as a substitute of days.
• Integration backlog: 18-month backlog cleared in weeks.
• Knowledge engineers who beforehand spent their time writing boilerplate ingestion code and chasing schema breaks now concentrate on strategic initiatives that advance the enterprise.
• Implementation velocity: a single developer took the agentic answer from proof of idea to manufacturing launch in 4 months.

The reliability, consistency, and information integrity of the MarTech platform had been improved, whereas the operational overhead was diminished: “The Knowledge Accelerator didn’t simply velocity issues up. It modified how we function. Our information engineers went from writing boilerplate ingestion code to specializing in strategic initiatives. Points will be recognized and glued earlier than our finish customers even discover.” says Kemp.

Conclusion

The Knowledge Accelerator’s success comes down to a few ideas: assembly builders the place they already work, preserving them on the helm, and embedding governance like GDPR classification from day one moderately than bolting it on after. These ideas formed an answer the place F1 partnered with AWS to make use of agentic AI on Amazon Bedrock AgentCore to rework MarTech information operations. By combining automated information supply onboarding, schema evolution detection, and unified information entry by way of Amazon SageMaker Unified Studio, F1 diminished onboarding time by roughly 99% and eradicated an 18-month integration backlog in weeks.

The strategy is intentionally replicable. Any group coping with multi-source information onboarding, schema volatility, and governance necessities can apply the identical structure to their very own surroundings. The brokers are domain-agnostic, they usually know onboard, classify, and monitor. The area is interchangeable.

Getting began

To be taught extra in regards to the AWS companies used on this answer:

Acknowledgments

This final result is the results of years of incremental enhancements to the MarTech platform, delivered by way of an in depth partnership between F1 and AWS. Many contributors throughout each organizations have formed the structure and strengthened the foundations that made the Knowledge Accelerator attainable. We’re grateful to the next thought leaders and builders for his or her dedication and experience: Paula Marenco Aguilar, Nadeen Nilanka, Taye Aduewa, Marton Juhasz, Deepak Gulia, Alex Goff, Nick Morgan, and Seshadri Senthamaraikannan.

---

Concerning the authors