At Premium Inboxes we monitor around 360,000 inboxes across roughly 2,000 agencies. I've watched this exact conflation play out hundreds of times. The language is always the same. It's the diversification line. Or it's the spreading-the-risk line. Or it's some variation on "I don't want to be all-in on one provider." And then the same operators take their sending budget, cut it in half, and call the result a real strategy.
What they've actually done is solve nothing. They've worsened the side that was working, added an untested variable on the other side, and bought no actual insurance in the process. The entire problem starts with the conflation. The rest of this article walks through the data behind why one side wins on output, how the budget should actually be allocated, and the harder question of whether you're running an optimisation game or a maximisation game underneath all of it.
The conflation: diversification vs insurance
The car analogy is the cleanest way I've found to explain this.
When you walk into a dealership to buy a car, you don't ask the salesman "does this include insurance." You don't expect the price of the car to bundle in your protection against the day it gets written off. The insurance is the thing you take out on top. Separate decision, separate budget, with a separate company who isn't even in the building. Nobody confuses the car with the insurance policy.
Yet, somewhere along the way in cold email, people started treating "the second provider" as if it was part of the same purchase. As if splitting your sending budget across Google and Microsoft was the same as buying insurance.
It isn't.
Because when you split your sending budget 50/50, here is what you've actually done. You've cut your output in half on the side that was working. You've added an unknown variable on the other side that you haven't tested. You've doubled your operational complexity, meaning two warm-up regimes, two sets of DNS quirks, two completely different deliverability fingerprints to track. And you've still got no actual safety net, because if Google nukes your accounts overnight, your Microsoft half can't suddenly pick up the slack at twice the volume to cover the gap.
You've worsened the thing you wanted to grow. And you haven't bought the thing you actually needed. That isn't diversification. That's confusion wearing a strategy hat.
What 360,000 inboxes actually show
When we started Premium Inboxes two years ago, we sold Google only. That wasn't a religious decision on my part. It was a market-fit decision.
Outlook was added to the menu only four months ago, after nearly two years of trading without it. And it wasn't added because of some big internal study saying Outlook had finally caught up on performance. It was added because a wave of legacy edu and non-profit panel collapses pushed customers in our direction looking for somewhere to land, and a wave of pure hysteria about Google "going bust overnight" was something I was tired of arguing with.
Outlook came in by sheer demand from the market. Performance was never the trigger.
Here is the number that explains everything. Across every inbox we have ever sold, roughly 96.4% have been Google. (It might be 99.4%. One of those two. I can never remember which side of the 9 it sits.)
I want you to sit with that number for a second before reading on.
Premium Inboxes sits in the middle of one of the largest cold email infrastructure operations in the world. There is a live feed of what's working and what isn't, and there are 250,000+ inboxes worth of behaviour to reference. If Outlook was significantly better, or even meaningfully comparable on a like-for-like campaign, my team would be selling it in vastly larger numbers than that.
We aren't. Because the market has already voted.
Over the last 12 months, Google has been consistently better than Outlook on positive reply rate across multiple customer campaigns we've helped run. The comparison was clean (identical copy and identical lists going through the two different sending providers in parallel), which is genuinely the only test that tells you anything useful, and that's the test we've run on real campaigns over and over again.
Can I tell you EXACTLY why Google has been better. Not really. Nobody can. Both run security policies and ML detection, both serve enterprise customers, and the second anyone figures out what's working the goalposts move. I thought my team would be the ones to crack the Google vs Microsoft question. We didn't crack it. Nobody has.
But the outcome is the outcome. When 250,000+ inboxes worth of data tell you one side is winning more meetings on identical campaigns, you stop philosophising and start allocating.
The same pattern shows up in our cancellation calls. Customers who tried 50/50 and quietly converged back onto majority-Google later because their pipeline numbers said so. One customer of ours burned through brand-new Google accounts in 2 months because he applied Microsoft-volume tactics to them and torched the lot. Another carried 95/98 warm-up score thresholds over from his old Smartlead days, then cancelled four workspaces based on a cutoff he couldn't even remember the reasoning behind.
The market keeps voting. People just don't like the verdict, so they call it "lack of diversification" and split their budget anyway.
If you split that budget on a hope that Microsoft is half as productive as Google, the maths punishes you twice. Once because you're sending fewer effective emails for the same money. And again because you've spent your insurance budget on something that isn't insurance.
How to allocate properly
Right. So what do you actually do.
You separate the two budgets completely, with no overlap between them at any layer of your business. Each one has its own objective, its own line in your books. The sending budget pays for output, and the insurance budget pays for resilience when the output side has a bad week. Those are the only two jobs at play, and they live in different places on the spreadsheet.
Step one. Name your objective.
Before you spend a single pound, you need to be honest about which of these you're actually optimising for. Optimising requires a target, and there are a few real ones.
Safety is the first. You want to know nothing catastrophic happens to your business if your primary infrastructure goes down overnight. You're risk-averse by design.
Raw performance is the second. You want every meeting you can possibly book this quarter regardless of cost. Burn rate doesn't really matter to you as long as the meetings keep landing on the calendar.
Longevity is the third. You want infrastructure that quietly runs for a long stretch without intervention.
Stability of unit economics is the fourth. You want your cost-per-meeting to stop swinging up and down with every infrastructure change you make.
These objectives pull against each other and you can't optimise for all of them at the same time, no matter how clever the spreadsheet looks. You have to pick one, and you have to actually mean it when you pick.
Step two. Size your sending budget for the objective.
Once you know what you're trying to do, the sending side becomes clearer.
If performance is the game and you're an agency, your sending budget goes entirely onto whatever is working best right now. For the overwhelming majority of cold email teams in 2026, that means Google. At this stage you stop hedging entirely and just maximise the channel that's actually producing meetings. If the accounts burn but printed 100 meetings a day each in the meantime, you bought more accounts and you carried on. The maximise game is a legitimate game to play, provided you know that's the game you're in.
If longevity is the game, the sending side is still majority Google. Longevity doesn't come from where you spend the sending budget. It comes from how disciplined you are about using it.
Step three. Set your insurance budget separately.
This is the move almost nobody makes, and it's the most important one in the framework.
You carve out a separate insurance-policy budget that has nothing to do with your primary sending. It has its own line in your books, its own review cadence, and its own success metric (which is "did I need to use it and did it work" rather than "how many meetings did it book").
The principle is non-negotiable. You carve insurance out of your spend as a fixed line item before you spend anything on primary volume, and you treat it as untouchable from then on. The exact size depends on how much your pipeline depends on cold email and how long you could survive a primary outage. That's a call you make with your finance lead. Nobody else can size it for you.
Step four. Choose the insurance vehicle.
Insurance isn't "the same sending operation on a different provider with the same warmup and the same volumes". Insurance is something genuinely uncorrelated with your primary sending. The whole point is that when your primary infrastructure has a bad day, your insurance isn't failing for the same reasons at the same time.
The options here are Microsoft, IONOS, self-hosted SMTP. It genuinely doesn't matter who, as long as the supplier and the account-source path are uncorrelated with your primary.
The one tactic we actually use ourselves on the insurance side is ~5-10 accounts direct from Google, paid in full at retail USD pricing. You keep them warm and live but you don't lean on them for primary volume. They sit there earning their keep simply by existing as a fallback when something else breaks.
Notice what's NOT on this list. "50/50 Microsoft and Google through the same reseller as my primary." That arrangement isn't insurance. That's a different shape of the same sending operation, which means it's exposed to most of the same risks (industry-wide ML detection changes, list-quality issues, sender-pattern policy shifts) that your primary is exposed to.
Real insurance is uncorrelated risk. A different supplier, a different setup, ideally a different category of account altogether.
Optimise or maximise? different framework for each
Here is the bit that nobody in the audience likes hearing.
The overwhelming majority of cold email teams reading this think they're optimising. In reality they're maximising. They just don't know it yet, and they're using optimiser language to describe a maximiser operation. The two games look identical from the outside if you don't know what you're looking at.
A team that is genuinely optimising has a fixed objective and a real constraint behind it. "I need to maintain this exact level of pipeline with the longest-lived infrastructure I can build." That team works the five things Google and Microsoft are actually watching. Sender reputation and authentication, sending patterns and warm-up, engagement signals, technical fingerprints, list quality. And they work all five in parallel rather than picking one to obsess over.
The team that is maximising has a totally different mental model. "I need as many booked meetings as possible from the infrastructure I can afford this month." That team is going to buy a lot more accounts than the optimiser would, send harder on each one, accept higher churn as a normal cost of doing the business they're in, and replace burned infrastructure without flinching. They aren't careless about any of this. They've just made a different trade on what to prioritise.
Most agencies are maximising. And they should be honest about it. Because when accounts print 100 meetings a day each, you just buy more accounts. It doesn't matter whether the accounts burn quickly. The client's going to stay because the pipeline keeps filling, and the agency's going to keep its retainer. That's the maximise game. It's a real game and plenty of agencies are winning it.
The problem comes when a maximiser tells themselves they're an optimiser. That's when you get the 50/50 split conversation. That's when you get the customer who builds a custom warm-up score monitoring system carried over from his Smartlead days, then flags four workspaces as underperforming based on a 95/98 cutoff he can't even remember the reasoning behind. The vocabulary of optimisation got applied to a maximise game. The framework didn't fit. Accounts got cancelled that didn't need to be cancelled, and spend got shuffled around for no real reason.
So before you build your insurance allocation, get honest about which game you're actually playing.
Closing
One last thing on this.
Anybody who tells you disconnections don't happen on properly-set-up infrastructure is living in la la land. Disconnections happen on every provider, every reseller, every account-source path. Even if you walked into Google's headquarters in California and bought your accounts in cash from the Google HQ desk, you would still have disconnections eventually. It's not if it goes wrong, it's when it goes wrong.
That's why the insurance budget exists in the first place. It has nothing to do with Google being bad or Microsoft being better. It has everything to do with the fact that uncorrelated backup is the only honest answer to the question "what happens when this provider has a bad day."
You don't buy fire insurance because you think your house is about to burn down. You buy it because the cost of being wrong is too high to absorb out of cash flow. The same logic applies to a sending operation that your business depends on for pipeline.
Your primary infrastructure should be optimised (or maximised) for output. Your insurance budget should be sized for the cost of being wrong, then parked somewhere that won't fail for the same reasons your primary fails. Those are two completely separate decisions that need two completely separate line items on the spreadsheet. They almost never benefit from being smashed together into one 50/50 split.
Stop calling that diversification when it isn't.
Actual diversification happens when you've sized your sending operation honestly for the game you're playing, sized your insurance for the cost of getting it wrong, and parked the insurance somewhere that won't fail for the same reasons your primary fails. If your "diversified" setup is two halves of the same supplier relationship, the same warm-up regime applied across both halves, with the same underlying sending pattern across all of it, then there is no diversification anywhere in the picture. All you have is a smaller sending budget and a hopelessly inadequate insurance policy, sitting next to each other on the same invoice.
Sort the budget split out properly before your next provider outage. The cost of getting this wrong scales with how much your pipeline depends on cold email, and for most teams reading this, the answer to that question is "too much."
Keep it simple.
Richard
Want help executing this?
At Premium Inboxes we build the sending budget side of this for you.
Official Google + Microsoft accounts, fully configured DNS, ready to deploy in under 12 hours.
Hit the link in my bio and use code “LINKEDIN20” for 20% off our done-for-you infrastructure set-up
