| |

What are the challenges for a marketing team now

In 2023 when the original post of this was written the hot topics that my clients wanted to know about and discuss with me were how to attribute leads when ad blockers and browser privacy features rendered between 40-60% of their traffic invisible. They wanted to discuss the cookie phase out, how to produce video assets at scale, influencer marketing and another shiny object that promised to be the new social media channel.

Some of those have faded, remember Clubhouse ? And one reversed itself: Google never removed third-party cookies from Chrome. The question underneath all of them is still the right one. What is genuinely hard for a marketing team running a real stack today?

Most of the list below collapses into one problem. Nobody owns how the tools connect. Consent records, first-party data, AI outputs, attribution, all of it depends on data moving correctly between platforms. Where that ownership is missing or not 100% defined and documented, every other challenge gets harder.

The stack: sprawl, integration, utilisation

MarTech’s 2025 State of Your Stack survey found that data integration is one of the biggest management challenge, cited by 65.7% of respondents. Gartner, reporting via MarTech, puts utilization at about a third: marketers use roughly a third of their stack’s capability, down from more than half in 2020.

Two symptoms that are emerging more and more when I have discovery calls with prospects or consulting calls with my clients. The first is sprawl, tools bought for a job the stack already does, so the same contact exists in multiple systems with slightly different names. The second is the workaround, the manual export, the CSV dropped into another tool by a person at the end of a long week. The export is what teams do instead of owning the connection, and it works until volume or a schema change breaks it.

The fix is naming who owns each piece of data, what feeds it, and what it may be used for, across the systems you already run. That means a document, a living one, that says: the contact’s email address is owned by your CRM, it comes from the form fill, and it may be used for transactional email and the marketing automation platform’s suppression lists. Nothing downstream may change it. When the CRM and the automation platform disagree about who that contact is, the process document tells you which side is wrong.

Consent is a data problem

Pew’s 2023 survey of US adults found that 81% say the information companies collect will be used in ways people are not comfortable with, 70% have little to no trust in companies to make responsible decisions about how they use AI, and 61% are skeptical anything they do will make much difference.

The legal side of that has a compliance team. The operational side sits in the tech stack. Consent is a value per contact. It lives in your CRM or your marketing platform, and it decides what every downstream system may do with that contact. When consent does not travel with the contact record, you cannot tell which people you may email, which you may profile, and which you may hand to a model. That is a data quality problem with a privacy consequence attached.

First-party data after April 2025

My 2023 post discussed the demise of third-party cookies. Google reversed its stance. In April 2025, Google announced it would maintain its current approach to third-party cookie choice in Chrome. Chrome users still choose, and third-party cookies still function there for those who allow them.

First-party data still matters, for reasons that were true in 2023 anyway. Safari, by WebKit’s own policy, has blocked third-party cookies by default since the early versions. Firefox isolates third-party cookies to the site that set them by default, and blocks them outright in Strict mode. Consent limits what you may collect, and the signals you do have keep thinning. The 2023 post gave you a deadline. The deadline never arrived. The pressure to own your own data is real regardless of what Chrome does.

There is a practical path here that does not require choosing between owning your data and using the tools you already run. You can run analytics collection through your own domain, a server-side or first-party tagging setup, where events are sent to a container you control on your own domain and then forwarded to Google for processing. Google’s dashboards and reports still work on the back end. You keep the tool and own the data at the same time.

AI adoption versus AI results

McKinsey’s state of AI survey, published August 2026 with 1,719 respondents across 97 countries, found nearly nine in ten respondents reporting regular AI use in at least one business function. 44% report scaling AI across the enterprise, up from 38% the year before. But only 37% attribute at least some EBIT impact to AI, and 6% qualify as high performers, defined as 5% or more of EBIT. The survey’s own finding: enterprise-level returns concentrate in the small group that redesigns workflows rather than inserting AI into existing processes. The survey is cross-industry and global, with large firms over-represented.

Salesforce’s State of Marketing, 10th edition, 2026, points at the same gap from the marketing side. 75% of marketers have adopted AI. 84% acknowledge running generic campaigns. Personalizing content is the top AI use case, and siloed systems plus poor data quality remain the top barriers: 98% encounter personalization barriers, with data quality the primary one. Only 58% have access to complete service data, 56% to sales data, 51% to commerce data. Salesforce sells the software its report argues for, which is worth keeping in mind when you read the rest.

Adoption is near universal. Results are rare. The teams that get results did the unglamorous work first: they know what data they hold, where it lives, and they rebuilt the workflow around the model instead of bolting the model onto the old one.

What to hand to AI, what to automate, what to keep

Use AI for judgment-shaped work, with a person at the end: drafting the email from call notes, summarising a conversation into the CRM, spotting the anomaly in a campaign, taking the first pass at a segment definition. The person reviews, corrects, and owns the output.

Automate instead of modelling anything rule-shaped: keep the data in sync between systems, route a lead by score, tag contacts, fire the drip when the trigger fires. This work never needed a model. A model adds latency and a new way to fail.

Keep with a person anything that decides about a customer, spends money, or sends in the company’s name. The refund judgment call, the budget shift, the outbound email under your letterhead.

Measurement and attribution

The Content Marketing Institute’s B2B Content Marketing Trends 2025 found that 96% of B2B marketers measure content performance, and 51% say they do it effectively. For 56%, attributing ROI to content is the top measurement challenge.

The attribution problem gets sharper when you consider what a click even means now. Pew’s July 2025 study tracked 68,879 searches by 900 US adults. When a Google AI summary appeared, 8% clicked a traditional result, versus 15% without one. About 1% clicked a link inside the summary. 18% of searches produced a summary at all.

Search engines are increasingly answering the query on the page, which means the click that used to carry a visitor to your content is disappearing from the equation. The work that still pays is being one of the sources the answer draws on, and that only works if your content is structured so the answer can actually find it. On a fragmented stack, even that is hard to verify, because you cannot tell which of your pages the summary drew on.

Where to start

AI on an unintegrated stack automates the mess. It scales whatever is already in the data, including the gaps and the wrong values. Data ownership comes first: who owns each piece, where the value comes from, what it may be used for. No rip-and-replace. Fix the connection on the tools you already run.

Start with one flow. Pick the path from form fill to first email, or from lead creation to first follow-up. Document every piece of data it touches, who owns each one, and what breaks when the integration between two of the systems fails. Do that before buying another tool or another model.

FAQ

Can we just replace the whole stack with something AI-native?

A rip-and-replace re-buys every integration you already paid to fix, resets consent records and history, and the new stack still needs the same data ownership. Teams that rebuild tend to rebuild the same gaps in new software. Fix the connection first. The stack shape can change later, around it.

If AI results concentrate in workflow redesign, how do we pick which workflows to redesign?

Start with work that already has a rule behind it but too much human variation in practice. Lead scoring and routing, campaign post-mortems, segment definitions. Those have enough structure for a model and enough inconsistency that the model’s steadiness is a genuine win. Leave judgment calls alone until the data feeding them is trustworthy.

If fewer searches end in clicks, what do we track instead?

Track the questions your content answers, whether you appear among the sources an AI summary draws on, and which steps still require a click: pricing pages, comparisons, demos. The unit of value moved from the click to the question, and measurement has to follow it.

Sources