The Wrong AI Question Marketers Keep Asking
One of the biggest questions I get from marketers and CMOs is this:
Which AI tool should I use for this job?
It could be content creation, media monitoring, analysis, thought leadership, account-based marketing, reporting, planning, or any number of workflows across the marketing organisation.
It is a fair question.
But it is usually the wrong place to start.
The real issue is ’t which tool to buy first; that’s where the friction is.
Where is the workflow breaking down?
Where is the business challenge most acute?
Where are teams losing time?
Where are my team's skills inconsistent?
Where are metrics and business outcomes weak?
Where is judgment still needed, but capacity is thin?
Where would a better human-plus-AI model create the biggest lift?
That is the right starting point.
Not the tool.
Marketers effectively have three choices.
Once you identify the workflow with the highest friction, marketers usually have three broad AI options.
Use a standard LLM and configure it around the workflow
This means using tools like ChatGPT, Claude, or Gemini, then building projects, GPTs, gems, or custom instruction layers around the task.Use a wrapper solution built for that workflow.
These are AI applications designed for a specific job, often with templates, collaboration features, approval flows, integrations, and a built-in workflow structure.Build or commission a more customised AI solution.
This is the route when the workflow is strategic, highly specific, deeply integrated, or too differentiated for either a standard LLM setup or an off-the-shelf wrapper.
To make this practical, let’s take the ‘Content workflow’ as an example.
It is one of the most common areas where marketers are deciding which AI route to take, and it makes the trade-offs very clear.
Start with friction, not fascination.
Too many AI decisions in marketing begin with fascination.
A team sees a demo. A vendor makes a big promise. Someone hears about a new wrapper app, a new model, a new copilot, a new agent. The conversation jumps quickly to features, speed, and output.
But marketing leaders should resist that temptation.
The smarter approach is to work backwards from operational friction and business need.
Start by identifying the workflows with the highest pain and the greatest potential impact from improvement. In some cases, the problem is speed. In others, it is inconsistency. In others, it is a lack of strategic thinking, poor data interpretation, content bottlenecks, weak collaboration, or weak governance.
Then ask a more useful question:
Should this workflow remain human-led, become human-plus-AI, or move toward greater automation?
Only after that should you decide which category of AI solution makes sense.
That sequence matters. Because if you start with tools, you often end up buying technology in search of a workflow. If you start with workflow friction, you are much more likely to find technology that actually solves something.
Option one, standard LLMs plus Projects, GPTs, or Gems
The first approach is to work directly within major LLM environments and shape them to fit the workflow.
In a content workflow, that means creating a project, GPT, or gem with the right instruction set and knowledge base. That could include brand voice and tone guidance, audience personas, messaging priorities, product facts, examples of strong past content, performance signals, content rules, and hard stops for legal or reputational safety.
This approach gives the marketer flexibility and control.
You are not buying a pre-structured workflow. You are effectively building your own workspace inside the LLM. That can be a major advantage when the work is brand-sensitive, when tone of voice matters, or when the team wants to experiment with and refine outputs in a highly customised way.
The upside is clear. You get flexibility, lower entry cost, closer control over how the model behaves, and the ability to create highly brand-specific outputs.
But this route also asks more from the team.
Someone has to write good instructions. Someone has to maintain the knowledge base. Someone has to know what quality looks like. And someone has to turn prompting into a repeatable process rather than a one-off experiment.
There is another layer to this decision that is often underestimated.
Choosing Option One is not just about using an LLM. It is also about deciding which LLM to use for the workflow and, sometimes, when to switch between them. One model may be stronger for writing, another for reasoning, another for analysis, another for cost efficiency, and another for speed. That gives marketers flexibility, but it also adds complexity.
So while this path may begin as the lower-entry-cost option, it does not always remain the lower-cost option in practice. The hidden cost can show up in model testing, switching, retraining users, inconsistent outputs, duplicated effort, and the time required to keep evaluating which model is best suited for which task.
Done well, this can be very effective.
Done badly, it turns into inconsistency disguised as innovation.
Option two, wrapper solutions for the workflow
The second route is to use a wrapper solution that is purpose-built for the workflow.
Instead of starting with a blank LLM environment, you start with a more structured application layer. In content, that may include built-in templates, campaign inputs, SEO prompts, approval flows, collaboration tools, publishing logic, and integration with content management or asset systems.
This route tends to work well when the priority is speed, structure, and scalability.
It is often easier for larger teams. It lowers the skill threshold for adoption. It can create more consistency across users. And it usually fits better within broader operational workflows involving multiple stakeholders.
It can also simplify the model decision itself. Many wrapper solutions now span multiple models, and some can automatically route tasks to the model they believe is best suited for the task. That means the marketer does not always need to decide which LLM is better for drafting, summarising, analysing, or optimising. The platform may abstract that complexity away.
That is one of the real advantages of the wrapper model. It is not just simplifying the workflow. It is also simplifying model choice.
The trade-off is that you are now working within the platform's structure.
That is not necessarily a bad thing. In fact, for many organisations, that is exactly the point. But it does mean you may have less flexibility, less visibility into how the logic is being applied, and less freedom than you would have if you built the workflow directly inside an LLM.
Wrapper solutions are often a very good answer when the workflow is repeated often, when teams need fast onboarding, and when governance or collaboration matters as much as generation.
Option three, customised AI solutions
The third route comes into play when neither of the first two options is enough.
This is where marketers work with an ISV, a specialist vendor, or an internal team to create something more tailored to their business. In content workflow, that could mean a solution that combines brand governance, internal content performance data, channel logic, approval rules, regional nuances, rights management, or links into existing planning and publishing systems.
This route makes sense when the workflow is too strategic or too differentiated to be handled well by a standard setup.
It usually requires more investment, more planning, and more technical support. But it can be the right choice when the workflow is central to competitive advantage or when it needs to fit tightly with how the organisation already operates.
Customised solutions are not always necessary. But in some environments, especially complex enterprise settings, they are the only route that properly reflects the realities of scale, governance, and organisational nuance.
In many cases, the answer is not either or
One of the biggest mistakes marketers make is assuming they have to choose just one path.
In reality, many teams will need a mix.
There are cases where a standard LLM with well-built projects is more than enough. There are cases where a wrapper solution is the smarter route because it adds structure and speed. And there are cases where a custom solution is necessary because the workflow is too important or too specific to leave to generic tools.
There are also cases where the best answer is a combination.
A team may use LLM projects for voice-critical thought leadership and messaging work, a wrapper solution for high-volume production and collaboration, and a customised layer for approvals, governance, or deeper integration with internal systems.
That is often what maturity looks like.
Not a single tool, but a deliberate architecture based on workflow needs.
The same decision logic applies across marketing
Content is just one example.
The same decision process can be applied across a much broader list (but not exhaustive, also not listed by priority) of marketing workflows, including:
Customer Segmentation, Audience Research and Insights
Customer journey orchestration
Brand governance and approvals
Brand Strategy
Creativity Strategy
Content Production
Social media management
Media monitoring
Campaign planning
Reporting and dashboard summaries
Thought leadership development
Email and CRM programs
Account-based marketing
Sales enablement content
SEO, GEO and AEO content support
Localisation and adaptation
Competitive intelligence
Lead management and nurture workflows
For each of these, the questions are broadly the same.
Where is the friction highest?
Where is the business impact greatest?
What skills are missing or uneven?
What part of the workflow should stay human-led?
What part should become human plus AI?
What part could be more automated?
And only then, what kind of AI solution is most appropriate, a standard LLM setup, a wrapper solution, a customised build, or a combination of the three?
That is the real takeaway.
The decision on AI in marketing should not start with a tool comparison.
It should start with workflow diagnosis.
Once that is clear, marketers can make smarter, more grounded decisions on where LLMs are enough, where wrappers add value, and where custom solutions are worth the investment.
That is how better AI choices get made, workflow by workflow.
Final closing line options
Content may be the example, but the decision logic applies across the entire marketing organisation. Diagnose the workflow first, then choose the AI path.
The future of AI in marketing will not be decided tool by tool. It will be decided on a workflow-by-workflow basis.
The smartest marketers will not ask, “What is the best AI tool?” They will ask, “What is the right AI approach for this workflow?”
