E7, Agent 365, and Copilot Credits: The Microsoft Billing Shift That Changes Your Margins

A Nottingham accountancy firm we work with signed off a Microsoft 365 renewal last spring without a second thought. Same seat count, same familiar SKU, a modest annual increase they'd budgeted for. Three months later their bill had grown by nearly a fifth — not because anyone bought more licences, but because a handful of staff had started using an AI assistant that quietly billed by the action. Nobody had lied to them. Nobody had upsold them. The cost simply appeared, line by line, in a place they weren't looking.

That story is about to become far more common. The Microsoft licensing changes landing in May and June 2026 — the arrival of E7, the rollout of Agent 365, and the shift to consumption-based Copilot Credits — represent the biggest structural change to how Microsoft bills SMEs since the move to per-user cloud subscriptions. For businesses, the risk is a bill that grows without a clear decision behind it. For IT partners reselling those licences, the risk is worse: margin that erodes silently while the invoice you pay Microsoft goes up and the invoice you send your client stays flat.

Let's take the pieces apart.

What's actually changing

The familiar Microsoft 365 world was mostly predictable. You picked a plan — Business Premium, E3, E5 — you multiplied by headcount, and you knew roughly what the year cost. Add-ons like Copilot were a fixed per-user amount you could plan around: buy the seat, get unlimited use, done.

The 2026 restructure breaks that predictability in three ways.

E7 arrives as a new top tier. Sitting above E5, E7 bundles the advanced security, compliance, and AI capabilities that Microsoft has been assembling piecemeal. On paper it looks like a tidy upgrade path. In practice it's a magnet for scope creep, because a lot of what E7 includes was previously sold — or trialled — as separate add-ons. If a client is already paying for a scatter of point solutions, E7 might genuinely save money. If they're not, it's an expensive default that bundles capability they'll never switch on.

Agent 365 formalises AI agents as a licensed product. This is the one most people haven't priced yet. Agents — automated assistants that carry out multi-step tasks rather than just answering questions — move from experimental features to a governed, licensed layer. Agent 365 covers the management, identity, and security wrapper around those agents. The wrinkle is that the licence to govern agents and the cost to run them are two different things, and the running cost is consumption-based.

Copilot Credits replace flat-rate AI billing in key scenarios. This is the big one. Instead of every AI interaction being covered by a fixed per-user Copilot subscription, a growing share of AI activity — particularly agent actions and heavier automation — draws down a pool of consumption credits. Use more, pay more. And crucially, the meter runs whether or not anyone is watching it.

How Copilot Credits actually bill in practice

Here's where the theory becomes a real invoice.

Under the credit model, actions have a cost. A user asking Copilot to summarise a meeting is cheap. An agent that runs autonomously — pulling data from three systems, drafting a document, cross-checking a record, and sending an email — consumes far more, because it's chaining multiple operations together. Each step draws credits.

The practical consequences matter:

• Cost scales with usage, not headcount. Two firms with identical licence counts can end up with wildly different bills depending on how enthusiastically their people use agents.
• A single power user can move the needle. One operations manager who builds a clever agent to reconcile invoices overnight can quietly consume more credits than the rest of the company combined.
• Overages default to being charged. When a credit pool runs dry, usage doesn't necessarily stop — it often spills into pay-as-you-go pricing, which is where surprise bills are born.

The uncomfortable part for SMEs is that credit consumption is genuinely hard to forecast before you've lived with it for a quarter. You don't know how your team will use agents until they start, and by then you're already spending.

Where the margin traps sit for IT partners

If you resell Microsoft licences under a managed-service agreement, the credit model is a direct threat to the flat-fee structure most of the industry runs on.

The classic managed-service proposal says: pay us a fixed monthly amount per user and we'll handle your Microsoft 365 estate. That works beautifully when the underlying cost is a fixed per-user licence. It falls apart when a meaningful chunk of the underlying cost is metered consumption you don't control and can't cap without switching off features your client relies on.

Picture the sequence. You quote a fixed managed fee that includes Copilot. Your client's team gets productive with agents. Their credit consumption climbs. Microsoft bills you for the overage. You're contractually committed to a flat fee. The gap comes straight out of your margin — and because it accumulates quietly across many small transactions, you might not spot it until the quarter closes.

How to structure procurement and proposals so this doesn't bite

The fix isn't complicated, but it does require you to stop treating AI consumption like a fixed licence. A few concrete moves:

Separate fixed and variable costs in every proposal. Split your managed-service pricing into a fixed platform-and-support fee and a clearly labelled variable line for consumption-based AI usage. Bill the variable part on actuals plus a handling margin. This is how mobile data and cloud infrastructure have been sold for years, and it's the honest way to price something metered.

Audit before you recommend E7. Don't let the shiny top tier become the default. List what the client actually uses today, price the equivalent capability, and only recommend E7 where the bundle genuinely beats the sum of the parts. For a lot of SMEs, Business Premium plus one or two targeted add-ons will remain the right answer.

Set credit budgets and alerts from day one. Treat the Copilot Credit pool like a spend cap. Agree a monthly credit allowance with each client, put monitoring in place, and alert both sides when consumption crosses 70 and 90 per cent. Nobody likes a surprise, and clients respect a partner who flags a rising cost before it lands.

Govern agents deliberately. Agent 365's management layer exists for a reason. Decide who can create agents, review what they do, and retire the ones that aren't earning their keep. An unused agent that still runs on a schedule is pure cost leakage.

Reprice on a review cycle, not a renewal cycle. Twelve months is too long to leave consumption pricing untouched. Build a quarterly usage review into your service agreement so both fixed fees and variable allowances can be adjusted against real data.

The bottom line

The 2026 changes aren't a trap that Microsoft is springing on anyone deliberately — they're a shift from predictable per-seat pricing to a mixed model where some costs are fixed and some are metered. The danger is entirely in treating the new world like the old one.

For SMEs, that means understanding that your AI bill now depends on how you use it, not just how many people you employ. For IT partners, it means redrawing your proposals so metered consumption is priced as metered consumption, with visibility and alerts built in.

Get ahead of it now, while it's a planning exercise, and it stays a manageable line on a budget. Wait until June, and it becomes a very awkward conversation about an invoice nobody expected. If you'd like us to model what E7 and Copilot Credits would mean for your specific setup before the changes land, that's exactly the kind of thing we're here for — get in touch and we'll run the numbers with you.