The pitch sounds identical

Every major platform now offers some version of the same story: AI agents that automate your workflows, answer your team's questions, and connect your business data. The language is interchangeable. The results are not.

The gap between the marketing and the reality matters most for small and mid-sized businesses, because SMBs cannot absorb a bad technology bet the way a 10,000 person enterprise can. A wrong call here means six months of sunk costs and a painful migration.

What each platform actually costs

Published pricing for AI agent capabilities across the four major platforms, converted to GBP at current rates. These are entry-level figures for a team of 20.

Platform AI agent cost Base platform required 20-person monthly total Lock-in
Microsoft Copilot ~£15/user/month M365 Business (~£10/user/month) ~£500/month Annual commitment
Salesforce Agentforce ~£20/user/month + £1.60/conversation Starter Suite (~£20/user/month) ~£800+/month Annual contract
Google Gemini Bundled with Business Standard Business Standard (~£11/user/month) ~£220/month Annual for lower rate
HubSpot Breeze ~£0.40/resolved conversation Pro Hub (~£72/month base) ~£150+/month Annual typical
Custom AIOS One-time build from £4,500 None Under £5 infrastructure You own the code

The numbers tell the first half of the story. A 20 person team on Microsoft Copilot pays £6,000 a year for AI agent access alone, on top of their existing M365 licences. Salesforce pushes past £9,600. These are recurring costs with no exit ramp.

The compounding problem: most SMBs do not use just one platform. They run a mix of Microsoft, Google, Xero, Slack, Simpro, and half a dozen vertical tools. Each platform's agents only see their own data. So you end up paying multiple per-seat fees for AI that cannot talk across your actual business.

Five problems every platform shares

1. Per-seat pricing punishes growth

Every platform charges per user, per month. Hire five more people and your AI bill jumps by £75 to £100 a month, permanently. A custom system charges nothing per seat because the infrastructure cost is fixed at pennies.

2. Agents only see one silo

Microsoft's agents read Microsoft data. Salesforce's agents read Salesforce data. Google's agents read Google data. None of them can natively reach into Xero, Simpro, your job management system, or the spreadsheet your operations manager actually runs the business from. Connecting external data requires custom integrations that typically add 20% to 40% to the published licence fee.

3. You own nothing

The agent logic, the prompts, the workflow configuration, the trained context: all of it belongs to the platform. Cancel your subscription and it disappears. With a custom AIOS, you own every line of code. You can modify it, move it, or hand it to any developer. It is yours.

4. Vendor lock-in is structural, not just contractual

After 12 months of building workflows, training agents on your data, and integrating your team's processes into the platform, switching costs are enormous. This is by design. The longer you stay, the harder it becomes to leave. Platform incentives and your business interests diverge over time.

5. Hidden costs surface late

Consumption billing (Salesforce's per-conversation charges, HubSpot's outcome pricing) sounds affordable at low volumes but becomes unpredictable as usage grows. Practitioners report 70x to 120x cost spikes when agents run multi-step workflows compared to simple queries. You will not see this in the sales deck.

What a custom AIOS looks like in practice

A custom AI Operating System is a single, purpose-built platform that replaces the patchwork. It connects the data sources your business actually uses, automates the workflows your team actually runs, and gives you an AI layer that sees across every part of the operation.

When enterprise platforms make sense

To be fair: if your business already runs entirely inside one ecosystem (all Microsoft, or all Salesforce) and you have 200+ employees, the platform agent may be the right call. The integration is native, the per-seat cost amortises across a large workforce, and you probably have an IT team to manage it.

But most SMBs are not in that position. Most SMBs use three to five different tools, have under 50 people, and cannot justify £500+ a month for AI that only sees a fraction of their data.

The real question

The decision is not "which platform's AI agent should we subscribe to." The decision is whether you want to rent intelligence from a vendor who controls the terms, or build an operating system you own outright.

One path costs more every year. The other costs once.