If you work in martech, marketing operations or related roles, you’ve surely heard colleagues and leadership complaining about data quality and their lack of trust in data.
We often place the blame for data quality on the system, because we’re not willing to fully say the quiet part out loud: The No. 1 factor in data quality is the people, the processes and the level of rigor in those processes.
Today, we stand on the cusp of a significant paradigm shift, in which we must reevaluate how we establish and measure trust in our own platforms. The infusion of AI capabilities in traditional platforms is growing at an unprecedented pace, and new AI-native challengers are here.
Through 2025, I explored the impact of unstructured data on marketing. I also went back to look at the leading trends that started the year.
As 2025 went on, we saw more AI-powered capabilities appearing in leading martech platforms, many of which were turned on by default. Now is the time to consider the adjustments needed for our “trust mindset.” Our historic data governance processes, which apply to both structured and unstructured data, are no longer relevant.
The heart of the issue is this: More of our stack’s processes will be based on probabilistic approaches rather than traditional deterministic rules.
- Deterministic: Based on exact data matches and preset rules and more rigid workflow conditions.
- Probabilistic approaches: Context-based interpretation, based on LLM AI capabilities to infer the meaning of data, as they process both unstructured and structured data.
Dig deeper: The future of the martech stack and MOps is ‘unstructured’
The tipping point is here
In October 2025, I wrote about the benefits and risks of extracting insights from CRM-captured email. During that same timeframe, multiple headlines caught my attention, declaring the CRM is dead.
While I didn’t necessarily agree with the provocative headlines, I think most would agree with the underlying challenges they highlight. Rigid workflows, data silos and manual overhead typically hamper our efforts to drive growth and productivity for organizations.
Perhaps the original CRM mission — to completely structure the unstructured — was an impossible goal to begin with. In other words, our CRM and marketing automation platform (MAP) initiatives relied on a one-size-fits-all playbook that said properly structuring most of the underlying data was the answer.
Even under the best set of rules and processes, we all end up dealing with edge-case exceptions that always result in our favorite fail-safe — the deterministic drop-down menu known as “Other.” Despite all prior efforts, data quality challenges are still front and center.
To address these challenges, Gartner and other industry leaders suggest that new criteria are needed around “AI-ready” data. I recall an old yet trusted framework called CRUD: Create, Retrieve, Update, Delete, which was ingrained in many underlying data management principles. Let’s have some fun by proposing a modern version of CRUD.
Dig deeper: Customer sentiment — and risk — are hidden in the emails in your CRM
CRUD 2.0: A mindset shift for AI-ready data
C = Context
Brinker and Riemersma’s year-end report provided new terminology to help us navigate these changes. For me, the emergence of “context engineering” was the lead story. The key metaphor the report explored was the Goldilocks principle of data: Evaluating the “just right” amount of data needed for our embedded AI agents and workflow processes to be efficient and effective.
Instead of asking whether we can trust the output, we’ll shift to determining if we can trust the underlying context the AI algorithm is operating on. Martech engineers will become data curation and context engineers.
R = Review by the ’human in the loop’
Probabilistic, AI-based algorithms require us to shift the team’s input capacity for coordinated “human-in-the-loop” review processes. We’ll need to retrain teams and introduce new internal fact-checking procedures so that the appropriate subject matter experts are pulled in, depending on the context of the data.
We’ll also need to expand our data-scientist mindsets to introduce new sampling processes and develop data confidence levels. Most importantly, while traditional processes typically relied on project-based milestones to rally cleanup efforts, the “MarTech for 2026” report forecasts a shift to a continuous review mindset, since many of these embedded capabilities will be running around the clock.
U = Upgrade
We’ll shift more attention from updating data fields to analyzing whether the overall process and decision quality are actually improving. Does the inclusion of AI provide significant business value, given the costs of the process?
AI-processing costs will become the rising tide in utilization metrics, not just account and contact data storage. Brinker had already predicted earlier in 2025 that we would need to move beyond counting users and learn new utilization-based systems, as AI/SaaS cloud vendors are already introducing usage-based consumption models to replace the cost-per-seat, tiered-feature models.
D = Declutter
In a strange twist, the potential risk of AI hallucinations — either in inaccurate reporting or mismatched content — from AI agents operating on poor data quality may be the key to unlocking time and capacity for long-overdue cleanup initiatives. More than 60% (62.1%) of respondents in Brinker and Riemersma’s survey indicate they’re already using built-in agents embedded in existing platforms. The time to act is now.
The risk of highly visible, inaccurate or automated campaigns may be the fuel necessary to make the business case to declutter our legacy systems.
Implementations that are active for more than six to 12 months don’t have their setups and workflows revisited as frequently as needed. However, the impact of this tech debt was likely hidden, as long as your teams maintained solid documentation and processes that clearly identified which fields were actively used. In the new probabilistic world operating across the entire CRM/MAP ecosystem, these historical data fields and workflows will introduce noise, impacting the reliability of your output and increasing errors at a higher rate.
Applying CRUD 2.0
Because most teams will start by revisiting their traditional CRM processes, I believe our organizations will need to bridge the gap in a step-wise fashion, by first retrofitting more minor use cases and then scaling those efforts more broadly. To illustrate this, I will revisit my simplified example from March — classifying a key contact correctly into the appropriate persona category when their job title included keywords such as “contract.”
Context
- If based solely on deterministic rules, using the word “contract” in job titles may be properly classified into Legal/Compliance.
- However, based on the contact’s involvement with RFP processes or terms and conditions captured through deal emails and related context, a probabilistic AI agent could determine that a key member of a buying group is functioning instead in a sourcing/procurement role.
Review
- A probabilistic-based workflow could suggest that the human in the loop create a new sourcing/procurement persona.
- If the answer is Yes, the AI agent could set that new persona appropriately
and/or…
- Notify an appropriate CRM operations leader to flag the related use case for a more thorough review (e.g., determining whether to retrospectively re-classify other contacts’ personas).
Upgrade
- Furthermore, if an automated campaign and content operations process were in place, new follow-up to that procurement audience could be drafted and only sent after the human in the loop reviews it.
Declutter
- Outdated job title and persona-based automations could be flagged for cleanup if a review indicates that former persona workflows were no longer needed, or those workflows could be cloned as templates for the new procurement persona.
Taking advantage of this moment
Based on all indicators, the data quality and AI-agent tipping point is here. If these new capabilities are just turned on by default, it would actually be appropriate to say, “I can’t trust the tool!” But with a proactive mindset that accounts for this paradigm shift, we may uncover hidden opportunities to finally address these challenges.
Fuel up with free marketing insights.
Contributing authors are invited to create content for MarTech and are chosen for their expertise and contribution to the martech community. Our contributors work under the oversight of the editorial staff and contributions are checked for quality and relevance to our readers. MarTech is owned by Semrush. Contributor was not asked to make any direct or indirect mentions of Semrush. The opinions they express are their own.
