AI and Email in 2026: I'm sitting on the fence

Last week, I spoke at Litmus Live for the second year running. This time, I was part of a panel titled “Embrace AI for Your Email Campaigns in 2026.” I was joined by Larry Kim, founder of Customers.ai; Leah Miranda, Lifecycle Marketing Manager at Zapier; and Rafael Viana, Senior Email Strategist at Validity.

I left that session feeling two very distinct things at the same time.

I felt inspired, and I felt uneasy.

Inspired, because I saw use cases that genuinely extend what’s possible in email marketing. Uneasy, because I’m not convinced the majority of businesses are remotely ready to implement what we were discussing.

This isn’t an anti-AI piece. It’s not a hype piece either. It’s a reflection, and a practical take on where I think AI actually fits in email, and where we may be in danger of over-engineering something that is already fragile.

The part that REALLY impressed me

There’s no denying the sophistication of what some teams are building.

We talked about everything from AI-assisted copywriting and end-to-end “magic button” tools that generate design, content and code in one go, through to AI agents embedded inside CRMs that continuously scan behavioural data to identify emerging intent signals.

The latter is where my interest sharpened. Using AI to analyse patterns at scale — identifying which combinations of behaviour correlate with conversion, churn, or activation — is not a gimmick or low level tool. It is commercially valuable. Humans are not good at spotting complex, multi-variable patterns across thousands or millions of data points. AI is.

When used this way, AI becomes an intelligence layer. It helps answer questions like:

• What signals typically precede a sale?

• How long does it take for a new subscriber to meaningfully engage?

• Which behaviours indicate someone is warming up versus drifting away?

• Where are we missing activation opportunities?

That’s powerful. And it aligns with something Larry said on the panel that I strongly agree with: the majority of email revenue tends to come from the smallest portion of emails sent.

In other words, it’s rarely the weekly batch campaign doing the heavy lifting. It’s the intent-based flows — the abandoned basket, the replenishment reminder, the high-intent page view trigger, the trial expiry nudge. These moments of relevance consistently outperform generic sends.

AI can absolutely help us identify and prioritise those moments better.

But here’s where my caution kicks in.

We have to be honest about the state of email

Before we start layering advanced AI orchestration over our programmes, we need to acknowledge something uncomfortable.

Email marketing, as an industry, is not operating from a position of structural strength. Over the past decade, particularly during and after COVID,  email volume accelerated dramatically. Businesses leaned into automation. Cadences increased. Performance expectations rose. Email was framed, repeatedly, as a revenue machine.

And in many cases, we burnt the channel.

We now operate in an environment characterised by:

• Information overload rather than information scarcity

• Deliverability instability

• Inbox fatigue

• Overlapping automations

• Weak lifecycle prioritisation

• Inflated expectations about what email “should” deliver

Many programmes have been inherited rather than intentionally designed. Email has often been treated as a performance lever instead of a trust channel. When revenue dips, the instinct is to send more.

That is the backdrop against which we are now introducing AI. So when someone says, “AI will fix this,” my immediate question is: fix what, exactly? If the underlying infrastructure is unstable, AI will not fix it. It will amplify it.

The danger of automating chaos

One of the most exciting concepts discussed was runtime generation — emails written or assembled in real time, based on current behavioural signals. In theory, this represents a significant leap forward in personalisation.

In practice, it introduces complexity that most organisations are not currently equipped to manage.

Imagine a single customer who, within a short time frame:

• Abandons a basket

• Revisits a product page multiple times

• Downloads a guide

• Subscribes to a newsletter

• Engages with a support ticket

Each of those actions is a signal. Each could theoretically trigger a highly relevant message. Without strict prioritisation rules and lifecycle clarity, you don’t end up with a beautifully orchestrated experience. You end up with message collision.

Multiple automations fire, tone shifts, urgency clashes with reassurance, marketing speaks while customer service is mid-conversation. The brand/business appears disjointed.

When I raised this concern, the response was essentially that you can instruct the AI not to create these overlaps. But that assumes the organisation has already mapped lifecycle stages, defined hierarchy between triggers, aligned sales, service and marketing communication, and cleaned its data.

Most have not.

AI cannot compensate for missing strategic thinking. It will execute whatever logic you give it. If that logic is flawed, the output will scale the flaw.

The black-and-white gap in our industry

Another tension I felt during the panel was the difference in scale between what is possible in enterprise environments and what is realistic for most businesses.

Large organisations with dedicated lifecycle teams, engineering support, strong data infrastructure and substantial budgets can build internal AI agents and advanced orchestration frameworks. That is real.

But most marketers do not operate in that environment.

They are managing email alongside SEO, paid media, website updates, content production, reporting and more. They are working with inherited ESP setups, limited data governance and stretched time.

When we talk about predictive orchestration engines and runtime content generation without acknowledging that context, we risk widening the gap between what sounds impressive and what is actually achievable.

The majority of the market is not choosing between “AI at scale” and “no AI.” They are choosing between “clean up what we already have” and “add another layer of complexity.”

For most small to mid-sized teams, the higher return often comes from tightening segmentation, improving onboarding, cleaning data, reducing volume and protecting deliverability — not from implementing advanced AI-driven orchestration.

Where AI makes sense in B2B

In B2B, the clearest and most practical application of AI is predictive intent modelling inside the CRM.

If a business has a content engine — blogs, webinars, whitepapers, events — and tracks meaningful behavioural signals, AI can analyse patterns across accounts and surface likely buying windows.

It can identify combinations of engagement that typically precede a booked call. It can flag accounts showing early churn indicators. It can help prioritise outreach and forecast pipeline movement.

This is not about generating prettier emails. It is about timing and prioritisation.

When AI functions as an analytical layer, helping marketing and sales align around real signals rather than gut feel, it strengthens the system rather than complicating it.

That is a use case I can get behind.

Where AI makes sense in B2C and D2C

In B2C and D2C, predictive retention and replenishment modelling is particularly compelling.

Consider a subscription or refill-based product. Retention hinges on usage behaviour, timing and memory as much as satisfaction. AI can analyse historic retained customers to identify patterns: average reorder windows, behavioural spikes before purchase, signals of lapse risk.

This allows brands to shift from reactive discounting to proactive timing.

Instead of sending generic promotional reminders, you can intervene when predictive signals indicate a drop in probability. That is materially different from blasting offers and hoping.

AI also has enormous value in attribution analysis. Many businesses still rely on surface-level metrics. AI can help quantify assisted conversions, lifecycle benchmarks and realistic performance expectations within the context of their own data, rather than generic internet advice.

Again, the value lies in insight and prediction — not novelty.

The time-saving trap

I understand why so many marketers are using AI to save time. Modern marketing roles are overloaded. If a tool can draft copy or assemble a campaign in seconds, the appeal is obvious.

But there is a subtle risk in framing AI primarily as a time-saving mechanism.

If the goal is speed, quality tends to erode. Email becomes faster, but not necessarily better. More content is produced, but not necessarily more relevance.

AI is most powerful when used to enhance judgement, not replace it. When it surfaces insight, clarifies patterns and sharpens targeting, it elevates the human role. When it becomes a shortcut for thinking, it degrades trust.

Email marketing is not just execution. It is psychology, timing, exclusion, restraint and narrative design. Those elements cannot be delegated entirely to automation.

Deliverability: alignment versus evasion

Another discussion point on the panel was the idea of “Sender AI versus Inbox AI.” As inbox providers become more sophisticated in filtering and risk modelling, some are building tools to navigate or work around those systems.

For me, that is a short-term game.

The long-term strategy is alignment. Sending fewer, higher-signal emails. Reducing negative engagement patterns. Protecting domain reputation. Designing experiences that encourage positive interaction.

If AI is used to amplify healthy engagement patterns and suppress risk, it strengthens deliverability. If it is used to outsmart filters without addressing underlying relevance, it becomes fragile.

Email itself is not changing — our approach is

One thing I feel strongly about: email as a channel is not fundamentally changing. It remains a utility and task environment. People scan, prioritise and triage. That behaviour has been shaped over decades.

The innovation is not the inbox. It is how we design around it.

AI can absolutely support that shift — if it is layered onto strong foundations.

But if those foundations are missing — if deliverability is unstable, segmentation unclear, lifecycle unstructured and data messy — AI will not produce a step-change in performance. It will produce more activity.

For small teams and solo marketers in particular, the priority should not be building runtime AI orchestration engines. The priority should be:

• Understanding deliverability health

• Clarifying lifecycle stages

• Tightening exclusions

• Defining engagement benchmarks

• Protecting reputation

Once those are stable, AI becomes leverage.

Where I’ve landed

I am not anti-AI. I am not dismissing its potential. I am cautious about skipping steps.

AI in email should amplify intent modelling, improve prediction, surface insight and strengthen decision-making. It should not be used as a substitute for strategy. If you do not understand your deliverability, lifecycle, segmentation and data infrastructure, implementing advanced AI tools will not transform your results. It may make your processes more complex without increasing impact.

The future of email is not AI replacing marketers. It is marketers who understand the system using AI intelligently.

For now, I’m choosing to focus on foundations first — and then using AI to level up what already works, rather than to fix what doesn’t.

That’s the version of 2026 I’m willing to embrace.