Choosing Your Platform

Choosing an email platform is one of those decisions that feels reversible but rarely is. The switching costs are real. Not just in dollars, but in time, rebuilt automations, re-warmed domains, and the inevitable dip in deliverability during migration. I’ve watched companies spend six months recovering from a poorly planned platform switch.

The good news is that the market in 2026 offers genuine choice across every price point and use case. The bad news is that the abundance of options makes the decision harder, not easier. There are several hundred email marketing tools on the market today. Most of them are mediocre. Perhaps a dozen or two are genuinely worth your attention, and the right one for you depends entirely on your business model, technical capabilities, and growth trajectory.

How to Choose

Six factors should drive your platform decision. I’d rank them in this order.

Deliverability comes first because nothing else matters if your emails don’t reach inboxes. Ask the platform for their aggregate deliverability data. Ask to speak with customers in your industry. Check third-party deliverability monitoring reports from companies like EmailToolTester or InboxReady. Some platforms invest heavily in deliverability infrastructure and sender reputation management. Others treat it as an afterthought.

The difference between a platform with 95% inbox placement and one with 85% inbox placement might not sound dramatic. But on a list of 50,000 subscribers, that’s 5,000 people who simply never see your email. Over a year of weekly sends, that’s 260,000 missed impressions. At a $0.50 revenue per email, that’s $130,000 in lost revenue. Treat deliverability as a revenue line, not a technical detail you can defer to later.

Automation capabilities come second. The gap between a platform with basic autoresponders and one with full behavioural automation is enormous. You need event-triggered flows, conditional branching, time delays, A/B splits within flows, and the ability to build complex multi-step sequences without engineering help. If your automations can’t respond to what subscribers actually do, you’re leaving money on the table.

I’d specifically test for: can you branch a flow based on whether someone clicked a specific link? Can you add a time delay that waits until a specific day of the week? Can you A/B test different paths within an automation, not just different subject lines? Can you trigger a flow from a custom event sent via API? These capabilities separate serious automation platforms from glorified autoresponders.

Segmentation depth determines how precisely you can target. Can you segment by purchase behaviour, email engagement, website activity, custom properties, and combinations of all four? Can you create dynamic segments that update in real-time? The difference between ‘purchased in last 30 days’ and ‘purchased Product X in last 30 days AND opened 3+ emails AND hasn’t visited the site in 7 days’ is the difference between generic and genuinely useful marketing.

Test the speed of segment calculation too. Some platforms take minutes to compute a complex segment. Others do it in seconds. When you’re building a last-minute campaign, that difference matters more than you’d expect.

Integrations matter more than most people realise during evaluation. Your email platform needs to talk to your ecommerce platform, your CRM, your analytics, your customer support tool, your payment processor, and increasingly your data warehouse. Check the native integration quality, not just whether it exists. A Shopify integration that syncs order data in real-time is fundamentally different from one that batch-syncs daily.

Pay attention to webhook support and API quality. Even if you don’t need custom integrations today, you will eventually. A platform with a well-documented, responsive API and flexible webhooks gives you room to grow. A platform with a limited API constrains your future options.

Pricing model is where things get interesting. Most platforms charge based on subscriber count, which creates a perverse incentive to keep unengaged subscribers on your list. Some charge by email volume instead, which is better because it encourages you to clean your list (fewer subscribers, same cost) and rewards you for list hygiene. Braze uses value-based pricing anchored to revenue, which means higher absolute cost but typically stronger ROI because the pricing structure aligns the platform’s incentives with your business outcomes. At the enterprise level, you might pay $50,000 to $200,000 per year for Braze, but if you’re generating $10M+ through email, that cost is a rounding error.

Watch out for hidden costs. Some platforms charge extra for SMS, for additional users, for priority support, for advanced reporting, or for removing their branding from your emails. The advertised price is often the starting price, not the real price. Calculate the total cost at your expected usage level, including all the features you actually need.

AI and programmatic interface is the factor that didn’t exist when I built SmartrMail and now belongs near the top of the list. The whole market shipped agentic email in the first half of 2026, so “can you run this from your AI agent?” is a real question with a real answer. The first thing to check is whether the platform has a native app or an MCP server you can point Claude, ChatGPT or another tool at, or whether everything still happens in a dashboard a human has to click through. The second is whether the AI is genuinely native to the product or bolted on top of a campaign-first architecture that was built in 2015. The two feel completely different in practice.

Then test the things that actually break. How good is prompt-to-campaign? Brief it to build a welcome flow and see whether you get something close to launch-ready or a pile of placeholder blocks you have to rebuild by hand. Does it send finished HTML as-is, and does the dashboard render every send the way it’ll land in the inbox? Raw-HTML parity sounds like a technical footnote, but if the platform silently mangles the HTML your agent emits, or if you can’t preview what an AI-built campaign actually looks like before it goes out, it’s a migration blocker and a trust problem. And if you run more than one brand, ask about multi-brand or agency mode: one agent login that can switch between client brands is a distinct buying criterion, not a nice-to-have, the moment you’re managing email for more than yourself.

The category split worth holding in your head is this. On one side, an email tool with AI features added. On the other, email automation that lives inside your AI agent. Most incumbents are the former and racing to look like the latter. A handful of newer platforms are built the second way from the ground up. Neither is automatically right for you, but you should know which one you’re buying, because the day-to-day experience of running your programme through a chat prompt versus a dashboard is not a small difference.

Make a shortlist of three platforms, running real campaigns through free trials or demos, and making your decision based on actual experience rather than feature comparison spreadsheets. The platform that feels right when you’re building your third automation is usually the right choice.

Platform Comparison

Here’s an expanded comparison of the major platforms worth considering in 2026.

Starting prices are as of mid-2026 and free tiers move around, so treat the numbers below as a snapshot to verify on the vendor’s pricing page, not a fixed quote.

Platform	Best For	Starting Price	Key Strength
Mailchimp	Small businesses	Free (500 contacts)	Ease of use, brand recognition
Klaviyo	Ecommerce (Shopify)	Free (250 contacts)	Deep ecommerce data, flows
Constant Contact	Small biz, nonprofits	$12/mo (500 contacts)	300+ integrations
Brevo	Multi-channel	Free (300 emails/day)	All-in-one (email + SMS + chat)
Braze	Enterprise	Custom pricing	Real-time stream processing
Loops	SaaS, PLG companies	Free (1,000 contacts)	Product email focus
Resend	Developers	Free (3,000 emails/mo)	React Email, dev-first
HubSpot	B2B, inbound	Free (2,000 emails/mo)	CRM integration, full suite
ActiveCampaign	Automation-heavy	$15/mo	135+ triggers and actions
Kit (ConvertKit)	Creators	Free (10,000 subscribers)	Creator-focused, simplicity
Postmark	Transactional	Free (100 emails/mo)	99%+ delivery, sub-1s median
SmartrMail	Ecommerce (Shopify)	$14/mo	ML product recs, fastest setup
Bento	Dev-first	$30/mo	API-first, SOC 2, MCP integration
beehiiv	Newsletters	Free (2,500 subscribers)	Growth tools, ad network
Sendlane	Ecommerce	Custom pricing	Deep ecommerce, SMS
Omnisend	Ecommerce	Free (250 contacts)	Email + SMS + push combined, MCP + ChatGPT app
Postup	Publishers, enterprise	Custom pricing	Publisher-grade deliverability, audience data
Vero	SaaS, product-led	$54/mo (5K profiles)	Event-driven, data warehouse native
nitrosend	AI-native teams	Free tier + paid plans	MCP-first, AI chat, API-driven

That’s nineteen platforms, and I could list another fifteen. The market is that crowded. A few things worth noting.

Mailchimp’s free tier dropped from 2,000 to 500 contacts, pushing many small businesses toward Kit or Brevo. Kit’s free tier at 10,000 subscribers was the most generous in the market as of mid-2026. Free tiers change often and usually downward, so check the current numbers before you commit. beehiiv has emerged as the newsletter platform of choice since 2022, because Tyler Denk built it to solve the problems he encountered growing Morning Brew.

Platform Performance Data

Klaviyo’s top 10% of users convert at 0.44% (5.5x the average of 0.08%). Automations generate 30x more revenue per recipient than campaigns. If you’re spending 80% of your effort on campaigns and 20% on automations, flip it. The gap between average users and top performers is entirely about how they use the tools: tighter segmentation, more granular triggers, better copy, more testing. Klaviyo is also the reference architecture for the AI-bolted-on side of the split: its Composer agent builds launch-ready email and SMS campaigns and flows from a single prompt, with human approval before anything sends, and there’s an official MCP server that brings the same workflows into Claude.

HubSpot reports 129% more leads and 36% more closed deals after one year of use. CRM-integrated email outperforms standalone email, especially in B2B. The flywheel effect (email feeds CRM, CRM makes email smarter) is HubSpot’s real value proposition.

Brevo’s pricing model charges by email volume rather than subscriber count. You can have 100,000 subscribers and only pay for what you send. This rewards list hygiene rather than penalising list growth.

SmartrMail is the Shopify pick for stores that want results without complexity. Now part of Relay Commerce and led by product manager Nicolas Vibert, its machine learning product recommendation engine can have automated flows running in under 15 minutes, the fastest setup-to-revenue time of any ecommerce ESP.

Bento is the developer-first pick: clean REST API, webhook-driven, SOC 2, a native MCP server, and a flat $30 per month regardless of list size, which makes it cost-effective at scale.

Vero is the pick for product-led teams whose email needs to respond to what users do inside the product, not what they buy in a store. The reason to look past the quieter brand is native data warehouse connectivity: it connects straight to Snowflake, BigQuery, Postgres and Redshift, so you build segments from production data without routing everything through a CDP first. No free tier, no ecommerce feature set, but over 12 years and 5 billion messages mean the infrastructure is proven.

Omnisend is the bootstrapped ecommerce pick, combining email, SMS and push in one workflow builder, and across the 12 billion emails it analyses, automated messages generate 29.6% of orders off just 2.2% of sends, which is why its pre-built flows get stores to automated revenue fast. Its annual benchmark reports (billions of messages across 150,000+ brands) are some of the most cited in the industry, and it now runs an official MCP server and a ChatGPT app, with a Claude connector flagged as coming, so you can build campaigns and query performance in plain language.

nitrosend is worth a look because it shows what an AI-native ESP actually looks like today, rather than an email tool with AI bolted on. Where the incumbents added AI on top of dashboard-first products built years ago, this one was designed from the assumption that the primary interface is conversational: a 23-tool MCP surface, an official ChatGPT app, native connectors for Claude, Gemini, Codex and Cursor, plus a REST API, all available down to the free tier. It runs the AI-native primitives the rest of the chapter argues for: brand-kit-driven generation so output isn’t generic, send-safety holds that block agent-initiated bulk sends without a human confirming, raw-HTML parity so finished HTML lands and renders as sent, and multi-brand mode for agencies running several clients from one login. Full disclosure: this guide and nitrosend share the same founder, so weigh that and test it against the others on your own shortlist rather than taking my word for it.

Loops thinks in product events and user journeys, not campaigns and audiences. Purpose-built for SaaS onboarding and product email. Resend treats email as a first-class engineering concern with React Email components, an official MCP server, and an AI editor that drafts in your brand voice from a URL, which is why it has become the default sender that AI agents reach for. Postmark delivers transactional emails in under one second. They deliberately don’t offer bulk marketing sends because it would compromise their transactional deliverability.

On the AI front specifically, the rest of the incumbents have moved fast enough that a snapshot is worth recording. As of mid-2026, the Mailchimp app is live inside both Claude and ChatGPT, so you can draft, refine and launch a campaign without leaving the AI client. Customer.io ships an AI Agent that builds a campaign from a prompt plus LLM Actions that call a model mid-journey to generate per-recipient content, which is a step beyond the build-once-then-send pattern everyone else markets. beehiiv runs an official MCP server that exposes content, subscribers, analytics and automations to Claude, ChatGPT, Gemini and Perplexity, for free. ActiveCampaign, Brevo, HubSpot and Kit all shipped their own agents and assistants in the same window. The takeaway isn’t that any one of these is the answer. It’s that “does it have AI” is no longer a differentiator. The question is whether the AI is the way you run the tool or a feature panel inside it.

Budget ESP Comparison for Small Businesses

If you’re just starting out or running a small operation, the free tiers are genuinely useful. Don’t spend money you don’t need to spend.

Under 500 subscribers: Mailchimp Free, Kit Free, Brevo Free, and MailerLite Free all work. At this stage, the platform barely matters. Pick whichever one feels most intuitive to you and start sending. You can always switch later when the stakes are higher. The important thing at this stage is to start building the habit of regular email communication, not to optimise your tech stack.

500 to 5,000 subscribers: Brevo at around $25 per month, MailerLite at $10 per month, and Kit’s free tier (which covers up to 10,000 subscribers) are your best value options. Kit’s free tier is remarkable at this level, giving you the core functionality of a platform that creators pay $49 per month for, with the tradeoff being limited automation features.

MailerLite deserves special mention in this tier. At $10 per month for up to 500 subscribers (and $17 per month for up to 1,000), it offers a clean interface, decent automation capabilities, and good deliverability at one of the lowest price points in the market. It’s the platform I’d recommend for anyone who doesn’t fall neatly into the ‘creator’ or ‘ecommerce’ categories.

5,000 to 25,000 subscribers: Now the decision matters more. Klaviyo at $60 to $150 per month is the clear choice for ecommerce because the revenue attribution alone pays for itself. ActiveCampaign at $49 per month gives you the most sophisticated automation at this price point, with 135+ triggers and actions that let you build genuinely complex behavioural flows. Kit at $49 per month is ideal if you’re a creator or educator.

Here’s the most important thing: don’t choose based on your current list size. Choose based on where you’ll be in 12 months. Migrating at 500 subscribers is trivial. Migrating at 25,000 with 15 active automations, a year of engagement data, and a warmed sending domain is a genuine project. Think ahead. The platform that’s perfect for your first 500 subscribers might be completely wrong for your first 25,000.

When to Switch

Migration is never fun. I won’t pretend otherwise. But staying on the wrong platform costs more in the long term than the pain of switching. If your platform’s deliverability is declining, if you’ve outgrown its automation capabilities, if the pricing has become unjustifiable relative to alternatives, or if a critical integration doesn’t exist, it’s time to move.

Here are the specific signals that should trigger a platform evaluation:

Your deliverability has dropped below 90% inbox placement and your platform’s support can’t explain why. Your automation needs exceed what the platform can handle without engineering workarounds. You’re paying more than $500 per month and not using half the features, meaning you’re overpaying for capabilities you don’t need. A competitor’s platform has introduced a feature that would materially change your results (like Klaviyo’s predictive analytics for ecommerce, or Bento’s API-first approach to automation building). Your team spends more time fighting the platform than using it productively.

Jimmy Kim, CEO of Sendlane, recommends running both platforms in parallel for two to four weeks during migration. This is excellent advice. It gives you a safety net and lets you validate that the new platform performs as expected before cutting over completely.

Jordie van Rijn, founder of Emailmonday, puts it well: don’t just compare features on paper. Get demos. Test with real data. What looks good in a sales presentation doesn’t always hold up when you’re building your actual workflows. I’d add: talk to customers who have been on the platform for at least a year. New users love everything. Experienced users know the real limitations.

Here’s a step-by-step migration process that minimises risk:

1. Export everything. Subscribers with all custom properties, engagement data (opens, clicks, purchases), flow structures, and templates. Export your suppression list. Export your complaint history. Don’t leave historical data behind. You’ll need it for segmentation on the new platform and you may need it for compliance records.

2. Set up authentication on the new platform. Configure SPF, DKIM, and DMARC records for your sending domain. Verify everything passes authentication checks before sending a single email. Use a tool like MXToolbox or Mail-Tester to validate your DNS records are correct. Getting this wrong means starting your new platform relationship in the spam folder.

3. Rebuild core automations first. Start with your welcome series, abandoned cart flow, and post-purchase sequence. These are your revenue drivers and need to be live before you migrate traffic. Don’t try to rebuild every automation at once. Focus on the flows that generate 80% of your automation revenue.

4. Send from the new platform to your most engaged segment only. Your 60-day active openers and clickers. This builds sending reputation with ISPs who see high engagement from your new IP or sending domain. Start with your best segment because their high engagement signals to Gmail, Yahoo, and Outlook that you’re a legitimate sender.

5. Gradually migrate remaining segments over two to four weeks. Move to 90-day actives, then 120-day, then the rest. Each wave should maintain strong engagement metrics before you expand to the next. If engagement drops significantly after adding a wave, pause and investigate before continuing.

6. Monitor deliverability metrics closely during transition. Watch inbox placement rates, bounce rates, complaint rates, and engagement metrics daily during migration. Any significant degradation means you need to slow down the migration pace. Use a deliverability monitoring tool like GlockApps or Everest to track inbox placement across major ISPs.

7. Keep the old platform active for historical data and as a fallback. Don’t cancel your old account the day you switch. Keep it running for at least 30 days as insurance. If something goes wrong with the new platform, you can fall back to the old one within hours rather than days. The cost of one extra month is trivial compared to the risk of a failed migration.

The whole process typically takes four to eight weeks for a mid-sized list. For enterprise migrations with complex automation architectures, expect two to three months minimum. Budget the time accordingly and don’t rush it. A botched migration can tank your deliverability for months, and recovering from a deliverability hit is far harder than preventing one.

---

The agent surface: can your stack be run by an AI?

Through 2025 and into 2026 the major ESPs and CRMs all shipped a marketing agent, and they landed on roughly the same shape: read the signals, decide a next best action, act across channels, with a human setting the guardrails and, in most of them, approving the send. Because they converged on that shape, whether the platform can actually be run by an agent becomes a real selection criterion. “Does it have AI” is a dead question, because every vendor says yes. What earns a place on your shortlist is whether the platform can be operated by an agent, yours or its own, without you babysitting a dashboard. Test it as a buying criterion, don’t believe it as a claim. Almost none of these vendors have published agent-attributed lift that survives a second look, so weigh the plumbing and discount the promises.

Four things make a stack genuinely agent-operable. Score each platform on your shortlist against all four.

An in-app agent that builds, not just suggests

The first is whether the vendor ships its own agent that does real work inside the product. Klaviyo’s K:AI Marketing Agent went generally available in September 2025 and plans and builds campaigns, flows and signup forms, learning brand voice and catalogue from a URL and staging the output for a one-click launch. Its Composer layer, which turns a single prompt into a staged multichannel campaign plus segments, was still in private beta when I wrote this and covered email and SMS only, so check the status before you lean on it. Salesforce’s Agentforce drafts copy, creative, audience and a journey in Flow off a marketer’s brief. HubSpot’s Breeze agents research, prospect and draft. The test that tells you anything: brief one to build a welcome flow and see whether what comes back is close to launch-ready or a pile of placeholder blocks you rebuild by hand. An engine that writes subject lines is a long way short of an agent that assembles a working flow.

An MCP or clean API so an external agent can drive it

The second is whether you can point your own agent, Claude or Codex or Cursor, at the platform and have it do the work. This is where an MCP server, or a genuinely clean and well-documented API, earns its keep. Iterable shipped an open-source Model Context Protocol server in November 2025 that lets those tools drive the ESP in natural language. Omnisend, beehiiv, Resend, Mailchimp and Bento all expose MCP servers or apps now, and nitrosend was built MCP-first, with its surface as the primary interface rather than a panel bolted onto a dashboard. One disclosure: this guide and nitrosend share a founder, so test it against the others rather than taking my word, and there is no independent outcome evidence for any of them yet. With no MCP and no clean API, your agent is stuck clicking through a browser: slow, brittle, and the first thing to break the next time someone ships a redesign.

Safe-by-default permissions

The third is the one vendors skip and the one you should weigh hardest. An agent with send rights and no guardrail is a production incident waiting to happen. Copy Iterable’s pattern: its setup wizard configures the MCP server read-only out of the box. No PII tools, no writes, no sends. Writes sit behind an ENABLE_WRITES flag and sends behind an ENABLE_SENDS flag, both opt-in. That is the right default. An agent should be able to read your data, draft a campaign and stage it for review long before it can dispatch anything to a real audience. When you evaluate a platform, ask exactly what an agent can do on day one with default permissions, and how you escalate from read to write to send. If the honest answer is “an agent can fire a bulk send the moment it connects”, walk. I have fired a campaign at 24,000-odd subscribers by hitting the wrong endpoint while looking for the approve button, so I am not theorising about how fast that goes wrong. Attentive’s AI Journeys sits at the other end: it triggers and personalises per shopper with no per-message approval, a materially looser model than the staged-for-review pattern everyone else ships. Looser is not automatically wrong, but know that is what you are buying.

A unified real-time profile the agent can read

The fourth is the data the agent stands on. An agent is only as good as the customer profile it can see, and a stale or fragmented one gives you stale or fragmented personalisation. The platforms built for this expose a unified, real-time profile: Braze grounds on first-party event data, Twilio’s conversation suite on Segment profiles, Salesforce’s Agentforce on the CRM plus connected sources. You want a single profile the agent can query in real time, behaviour, purchases, engagement, consent state, rather than a nightly batch that leaves it acting on yesterday’s picture. Test it with a real question. Ask the agent to pull a live segment and check how current the data actually is.

A closing dose of realism, because this question attracts more hype than any other. Gartner expects more than 40% of agentic AI projects to be cancelled by the end of 2027, citing unclear value and inadequate risk controls, and reckons a fair slice of what gets marketed as agentic is “agent washing”. MIT’s NANDA study found 95% of enterprise GenAI pilots showed no measurable P&L impact, the root cause an organisational learning gap rather than the model. So score the agent surface, but weight it behind deliverability and automation, and buy the platform whose guardrails you trust over the one with the boldest demo.