This is the continued elaboration of my “Tiny Models and the Power of Your Data” session at MAICON, the annual not-to-miss conference produced by the Marketing Artificial Intelligence Institute. Continuing from PART 1/4, The Limitations of Public Models, posted on September 25 in which I concluded with “Before your AI Strategy, You Need a Data Strategy,” here we go!
Data: The Foundation of Effective AI
There are dozens of ways to go about data warehousing and leveraging your CRM, public and private cloud environments, and a rising new class of data exchanges, clean rooms, and data centers. If you think this will be uniform, standardized, or executed by industry with “best practices,” please hold my 🍺.
“The future is already here — it’s just not very evenly distributed,” William Gibson (or someone else) originally stated. This has never been more true.
Below, I will provide guidance for outcomes and use case-based planning for responsibly aligned data management and governance. This may lead your organization to artificial intelligence (AI) that your organization controls and scales with accurate data (intelligence) to train and operate custom language models (CLMs), or “tiny models,” which we will address later in this series.
AI agents are on the way, but for now let’s meet the prompt engineering crowd where they are comfortable.
This involves several key steps and considerations:
1. Establishing a Data Intelligence Task Force: Bring together practitioners from marketing/CX operations, analytics, IT, HR, and business strategy, not C-level leadership alone. The hype to hire a “Chief AI Officer” today often overlooks the importance of your current state of operations and workflows. Forming a task force or council that builds “data” or digital literacy across the organization can also save you millions in regulatory fines and security breaches.
Jessica Hreha has been a constant beacon for enterprise adoption of AI. I’ve always appreciated her leadership in assembling the VMWare AI Marketing Council, and at MAICON this year she provided a deeper and more prescriptive way to vertically work across disciplines and corporate management structures to align use case prioritization and organizational planning for AI-assisted operations. At BrainTrust Partners, we consider “data intelligence” and data literacy as the fundamental priorities for business today, as without a business’s secure and unique data (intelligence), any AI effort may be short-lived and irrelevant to employees and customers alike.
Define clear objectives and key performance indicators (KPIs), and prioritize use cases for data management initiatives in specific business units and operations. Too often marketing teams and contact center/CX teams forget that they are engaging the same customers and audiences and they could mix measurements like lifetime value (LTV), return on ad spend (ROAS), net promoter scores (NPS), and customer satisfaction calculations (CSAT). Often the same audiences and same data, surely redundant and costly in separate systems, are also the origin source for “user” feedback on mobile applications, post-purchase surveys, and all the digital products/interfaces your brand puts in front of customers today.
Establish a data dictionary and governance protocols for data usage and AI implementation. Leveraging a task force or council comprised of leadership across all business units, you need to have everyone aligned and acutely monitoring data security, which includes customer privacy and company intellectual property (IP).
Regularly assess and report on data quality and AI performance. Quarterly business reviews (QBRs) are smart to keep vendors and your team aligned, and for spotting conflict between automated analytics across legacy SaaS tools, CRMs, marketing, and other operational systems. Every legacy vendor is claiming they have AI of some flavor, and we need to be sure these new-fangled automated wares don’t conflict with one another… and conflict, they will. Especially if your systems are not integrated with each other and disparately managed.
2. Implement Master Data Management (MDM): Create a single source of truth for your organization’s data, across all operations and divisional business units. We work closely with Reltio for enterprise data unification that vastly improves operational efficiency, builds trust, and ensures consistency and accuracy across all systems, not just marketing alone. Customer data platforms (CDPs), which have largely moved to composable deployments, and can be integrated with an MDM such as Reltio, a data management system like Databricks, or public/private cloud environments managed by Snowflake (each a partner of BrainTrust Partners) may also greatly assist with data organization and AI readiness. We also work closely with Axel Automotive to provide master data management to auto OEMs and dealers.
Implement data cleansing and enrichment processes (consider how a CDP can continuously capture and resolve customer and media signals) that are consistent across the enterprise and between business units. Your products and company information change over time, almost as frequently as customer devices (MAC and IP addresses), email, phone, address, and names are changed or used differently as each experience unfolds, a form is filled, and life simply altering how customer identities may be reported or captured.
Develop a data taxonomy, dictionary, and catalog to improve discoverability and accurate usage, helping you also define privileged and limited-privileged data access for security and privacy assurance. Everyone in business must know that data do not lie and there must be a unified way such intelligence in discussed, requested, and applied to decision-making and operations.
3. Deploy a Customer Data Platform (CDP): Centralize customer data from multiple sources and create comprehensive customer profiles (a “golden record” for each customer/household). CDPs are available on the market for almost any size business, and I believe they will soon take over a majority of marketing automation operations as AI models and agents are configured to work alongside proven machine-learning and mature data. This will be the major trend for CDP vendors in 2025.
A CDP enables real-time data activation for individualizing communications across inbound (inquiry, support) and outbound (marketing/sales) customer engagements. Facilitate iterative compliance with data privacy and security regulations, determined by where your business operates and where your customers reside, and register their opt-in consent to process their data, and their agreement to receive your communications/media.
Back to “data enrichment and cleansing”: Embrace the opportunity for your CDP to be the source for Retrieval-Augmented Generation (RAG) on customer preferences, needs, and the performance/behavior signals that all of your owned/paid/earned/shared media and metadata are producing. At BrainTrust Partners, we are closely allied and trained to implement Amperity, Treasure Data, Zeta Global, and Orbee (automotive) to enable high-performance marketing operations with all-but-guaranteed higher revenue, lower media/marketing operating costs, and the first-party data you’ll need for brand/customer-centric AI that you own and control.
4. Enhancing Data Security and Privacy Measures: Get fair and complete cybersecurity insurance. If you can use a broker that employs SecondSight in the application process and underwriting, that’s a sure way to ensure your application and the subsequent underwriting are accurate and fair. Also, consider setting up a company workbench with SecondSight to monitor security vulnerabilities and compliance across your business. Here in the Lone Star State, we are entering our third month with the Texas Data Privacy and Security Actin force, and we expect a litigious future for brands that mismanage customer information.
Perhaps as a regularly scheduled Data Intelligence Task Force meeting, host monthly briefings and quarterly training to improve data literacy and familiarity with security and privacy protocols. MDMs and CDPs, and additional security tools will help you implement robust encryption for data at rest and in transit, end-to-end, and for every origin data source.
Putting it all together:
- Considering secure data access, establish role-based controls for sensitive data, and include multi-factor and bio-authentication on every device and app used in your business.
- Conduct regular security audits and vulnerability assessments (AI agents may take care of this for you, and BrainTrust Partners has you covered). This may also be addressed with Systems and Organization Controls 2 (SOC 2) that include security (penetration testing), availability, processing integrity, confidentiality, and privacy monitoring. Still, you want the assurance of regular password hygiene by every employee and the myriad controls mentioned above in this article.
- Develop incident response plans for potential data breaches. This is as much an “if” as a “when” preparation that every business should put into practice. Bad things happen to good people, and good intentions often result in bad outcomes, so let’s address what we may control and protect as part of any “shield” or wall put around your business. This should involve communications executives (a PR firm, if you retain one), and a full operational plan for managing media outreach and customer notifications.
Before you dive into AI, be sure to have a data strategy and plan for protecting what is an asset class that will grow in value as it matures. Much more is required to responsibly manage and care for this asset than what I’ve outlined above, and the steps to master data management, performance marketing, and AI that you own and control are always changing.
By prioritizing these data management initiatives, organizations can build a solid foundation for AI implementation and ensure they’re working with high-quality, secure, and compliant data.
In Part 3/4, we will get into the exciting world of custom and private language models, “The Promise of Tiny Models.”
