Give AI the Right Work, Not All the Work

Rachel McLeay

Strategy & Innovation Lead

10th November 2026

Share
this article

At some point, someone decided that exporting data, copying and pasting it into a spreadsheet, and manually reformatting it for a report was just part of the digital marketer’s job. It got added to the to-do list, the to-do list became the workflow, and the workflow became the standard. Now here we are in 2026, with more tools at our disposal than ever before, still doing a lot of it by hand.

Digital marketing teams work with enormous volumes of data by necessity, from keyword sets and prompt banks to backlink profiles and social listening reports, across half a dozen platforms. That volume is structural, but what isn’t structural is the assumption that assembling, formatting, and moving that data around is sacred, human-only work… in most cases, it’s not – we just haven’t made a deliberate decision about what to do with it.

The admin tax: The work behind the work

The problem isn’t unique to digital marketing (though I’d argue SEOs deal with inordinate amounts of data compared to the average marketer!), or even to marketing, but applies to most modern ‘knowledge work’.

Research by ProcessMaker found that the average enterprise employee performs over 1,000 copy-paste operations every week, and spends more than half of their working time creating or updating documents – spreadsheets, PDFs, Word files, etc.

…over half.

…of a working week.

…on document admin.

So how do we address this so-called “admin tax”? McKinsey’s 2024 Superagency in the Workplace report outlined that organisations using AI effectively are reclaiming 20 to 30% of working hours for higher-value activity. The caveat is “how?” – because we also know that nearly half of LLM users think that models are smarter than they actually are. I would argue that, though substantial time can be saved, defaulting to AI as the main means of automation creates unnecessary risk – saving time at the expense of data integrity – and good data is the bedrock of informed decision-making.

The gap we have right now is in deciding, deliberately and systematically, which tasks should be delegated, and to what tool or to whom. These decisions are often overlooked, and so the admin tax keeps compounding.

Before we get into the how, it’s worth naming the three main options available when you look at any task on your plate:

Automate it – fully, whether through rules-based logic or AI.
Streamline it – automating the process-driven tasks while keeping a human involved where judgement, context, or expertise is needed.
Keep it human – not because a tool can’t do it, but because the human doing it is precisely what makes it valuable.

First, let’s talk about what happens inside the automation decision – because we’ve got to step away from equating “we can automate this” with “let’s give it to AI”.

Within automation, there’s still a choice

Most conversations about automation treat it as a binary; something is either automated or it isn’t. However, within automation, there’s a decision that most teams aren’t making explicitly – and skipping it is where things can quickly, but oh so quietly, go wrong.

There are two fundamentally different types of automation available to us:

Rules-based (deterministic) automation
AI-assisted (probabilistic) automation

What is rules-based vs AI marketing automation?

Rules-based automation follows a fixed set of instructions and always produces the same output, like a calculator. AI-assisted automation makes judgement calls based on context, more like asking a knowledgeable colleague – useful, but not always guaranteed to be right.

When should you use rules-based vs AI-assisted automation?

If a task has a clear right answer (i.e. something you could fact-check) keep it rules-based. If the task involves messy or varied information that requires interpretation or synthesis, that’s where AI earns its place.

What are the operational implications of each automation type?

Rules-based automation is cheaper to run, consistently accurate, and low maintenance once built, while AI-assisted automation trades some of that certainty for greater flexibility, which means higher running costs – and a human still needs to check the work.

What type of automation is right for your workflow?

First, break your workflow down into each step or ‘task’. Then, the question to ask before automating any part of it is: “Does this task have a correct output I could verify against a ground truth?”

If the answer is “yes”, keep it rules-based. If the task requires inference, synthesis, or interpretation of unstructured inputs, then that’s where either AI earns its place, or we keep it human.

This is the logic behind two internal tools we’ve built at Kaizen. In both, deterministic automation handles the structured work – things like keyword movement analysis, backlink extraction, and relevance scoring – while AI is introduced only where it adds value, such as surfacing themes from cleaned datasets or generating search terms from press releases and outreach copy. The sequencing is deliberate: AI doesn’t make decisions that have clear, rules-based answers, and it doesn’t interpret messy data. Instead, it operates within a controlled workflow, with humans owning the final analysis and recommendations.

The tools themselves aren’t the point; it’s the decision-making behind them. Every time a new workflow lands on your plate, the same question applies: which parts of this have a correct answer, and which require inference? The former stays rules-based. The latter gets AI, with guardrails.

McKinsey’s research is telling here: the largest productivity gains from AI consistently appear in language-heavy workflows, dealing with unstructured data – reporting, documentation, and communication. To mitigate unnecessary risk and make the most of what AI has to offer you, try to consider the nature of your tasks before defaulting to AI when automation is needed.

The delegation framework: automate, streamline, or keep it human

So, when rethinking a workflow, I look at the tasks involved and make a deliberate choice between three choices – landing on any of these by default is where things start going wrong.

Automate

These are the fully process-driven workflows – predictable inputs, predictable outputs, no human interpretation required. When the task has a correct answer, rules-based automation should be your first instinct. AI-assisted automation applies where deterministic methods genuinely can’t handle the complexity: synthesising unstructured inputs, identifying patterns across large datasets, and generating outputs that require intelligent inference rather than pure rule execution.

The monthly reporting pipeline that still involves someone manually pulling data from four platforms into a spreadsheet every Friday morning? That’s an ‘automate’. The rank tracking comparison that takes three hours because the deduplication is being done by hand? Also an ‘automate’. The task is about getting from A to B.

Streamline

This is where most of the interesting work lives, and where most workflows fall – workflows where some steps are process-driven and automatable, but others genuinely require a human in the loop. Humans aren’t used as a checkbox here, but rather because their judgement, context, or expertise is what makes the output worth anything.

The goal of streamlining is to bring in your experts where they’re actually needed, and free up time for the other strategic or creative tasks on their to-do list. For example, an LLM can draw solid conclusions from reporting data, but your account lead might have gained critical context surrounding that data following a recent client call, or from their deep, historic knowledge of the account – you need that human to apply their subject matter expertise that the tool does not have access to. We don’t want to automate these steps away, because they’re the steps that define the quality of the work.

Keep it human

Sometimes, keeping a task firmly within the hands of your experts is the decision that protects the very thing your clients are actually paying for.

Let’s consider campaign ideation in digital PR. The value of a successful campaign lives almost entirely in the unexpected angles – yes, there will be data behind the final campaign, but it’s the idea nobody saw coming, the story that’s genuinely interesting rather than just topically relevant, that gets journalists excited and earns coverage. A model trained on what has worked before will, by design, trend toward what has already been done. It can surface patterns, process existing coverage, and support the research leg of the process, but it can’t bring something genuinely new to market. Outside-the-box thinking that makes a campaign land is integral to campaign ideation.

The same logic applies to strategy. Automation can approximate parts of it – auditing competitors, identifying gaps, and flagging priority areas. However, building, visualising, and communicating a strategy that’s tailored to a specific client, with a point of view that’s ahead of where the market is now rather than reactive to where competitors and ‘best-in-class’ currently sit, requires commercial judgement and contextual knowledge that comes from your people.

The skill behind the framework

A framework is only as useful as the people applying it, and applying this one well requires enough foundational knowledge to make a confident delegation decision.

It’s pivotal that we continue developing our future experts and leaders. How will we decide whether a task belongs to rules-based automation or AI-assisted automation if our team doesn’t understand what the task actually involves? How will we spot a flawed AI output if no one knows what the correct answer looks like? Who will action the ‘keep human’ work if the experience and judgement that makes it valuable has been gradually compressed out of the team by over-reliance on tools?

To support our teams further when it comes to AI vs human delegation, which can feel slightly blurrier, every member of the team at Kaizen now holds an accreditation in AI Fluency – part of Anthropic’s course covers delegation to AI tools through three core lenses: problem awareness, platform awareness, and task delegation. After running the first training cohort, we saw a +20.6% uplift in our team’s confidence when delegating tasks between themselves and LLMs, with decidedly less reliance on AI for highly creative tasks. This matters to the clients we work with because there is a meaningful difference between an agency that uses AI and an agency that uses AI deliberately.

HubSpot’s 2025 State of Marketing report found that 92% of marketers say AI has already impacted their roles. This isn’t surprising; AI is a game-changing tool, just like Microsoft Office was when it first landed on the scene, but impact and intentional adoption are not the same thing. Yes, most teams are using AI, but there’s so much to gain from asking the right questions before doing so.

Key Takeaways

The cost of process-driven work in digital marketing depends on a series of decisions that most teams haven’t made yet
Before automating anything, ask: Does this task have a correct answer I could verify? If yes, keep it rules-based
Rules-based and AI-assisted automation are not interchangeable – defaulting to AI where deterministic logic would suffice introduces an unnecessary risk
Keeping some types of work human is an active decision that protects the quality, originality, and impact of your work
Delegation is a skill that requires the same intentional judgement as any other strategic decision

The ‘admin tax’ only reduces when teams start treating process-driven work as a series of decisions rather than an inevitable part of their day-to-day; they decide what would be faster using rules-based automation, what would be safely augmented with AI-assisted automation, what requires non-negotiable expert input or handling, and what would function best with all three working in sequence.

The three-layer framework is a starting point for that conversation, but not the complete answer. Where the lines fall will look different for every team, every brand, and every workflow. However, the question: “What does this task actually require?” is one worth asking every time – the alternative is another year of copy-pasting data into spreadsheets, and that’s no one’s idea of fulfilling work.

Things worth reading

Superagency in the Workplace – McKinsey, 2024

McKinsey’s research on how organisations are reclaiming 20-30% of working hours for higher-value activity – with the emphasis firmly on workflow redesign over tool adoption.

2025 State of Marketing – HubSpot, 2025

HubSpot’s report, framing AI as no longer experimental but embedded in everyday marketing workflows – shifting the focus from adoption to application.