You are already paying for AI. You just do not know where to plug it in yet
Most growing businesses are spending thousands a year on AI without ever making a deliberate decision to do so. AI sits inside the tools you already pay for, plus a standalone subscription or two on top. AI inside the existing tools has access to your data but no instructions. AI you go to directly takes instructions but no persistent access. A third form, agents, combines both. Most growing businesses have not deployed agents. This piece offers a four-level diagnostic for where your business actually sits today, and what to fix first.
The arithmetic is straightforward. A 10-person business on Microsoft 365 Business Standard pays roughly £1,300 a year for what comes built into the suite. Add three Copilot seats and that becomes another £900. Add two ChatGPT or Claude subscriptions for the people who use them most and that is another £400. The accounting tool runs AI in the background. So does the customer system. So does the inbox. Total annual spend on AI for that one business runs into the low thousands. For larger businesses, the figure is materially higher.
The problem is not the cost but rather that it delivers very little in its current form stacked together.
Three forms of AI you are already paying for
Three forms of AI exist in any growing business today: Embedded AI built into tools you already pay for, Standalone AI you go to directly like ChatGPT or Claude, and Agents that combine both with persistent connections to business systems. Embedded AI has access without instructions. Standalone AI has instructions without access. Agents have both.
The categories do not separate by what the products can technically do. They separate by how the business is using them.
Embedded AI. Built into tools the business already pays for. Xero. HubSpot. Microsoft 365. Shopify. The accounting AI is reading your transactions whether you instruct it to or not. The marketing tool is scoring your leads. The inbox is summarising threads. Embedded AI has standing access to your business data by default. It runs on its own schedule, with no input from you.
Standalone AI. The subscription the founder or a few staff use directly. ChatGPT. Claude. Gemini. The user opens a chat window, asks a question, gets an answer, closes the tab. Standalone AI can technically connect to business systems through what are called connectors, but in most growing businesses that is not happening. The tool is being used as a smart assistant for individual tasks. Drafts. Summaries. Research. Bounded to the user. The output rarely flows back into the business in a structured way.
Agents. AI configured with standing instructions and persistent connections to business systems. Acts on its own. Work persists across sessions. An agent might monitor invoices, follow up with overdue clients, log responses into the customer system, and escalate the difficult cases to a human. The tools to build these are now widely available. Most SMEs have not configured them where the real work happens.
The boundary between Standalone and Agents is no longer about capability. It is about how the business uses what it has. A modern Standalone subscription with three connectors set up and a clear standing instruction is functionally an agent. The same account used as a chat tool is Standalone. Most growing businesses are doing the second.
A four-level AI maturity diagnostic for your small business
The four levels below measure how an SME has actually deployed AI in its operations. Each level is tested through four questions.
1. What can AI see? Is the work of your business legible enough for AI to read it, or does it live in someone's head, in undocumented meetings, and in tools the AI cannot reach?
2. What can AI do? Can it act on the systems where the real work happens, or can it only summarise what humans already wrote down?
3. Who can extend it? Can people in your business build useful AI workflows without needing a developer, or is everything held together by one or two power users?
4. How has the business changed? Is the way work is divided, the way people are hired, the way decisions are made, materially different from two years ago? Or are you running 2024's structure with better autocomplete?
Each level also has a hard test, and a common false positive: something that looks like progress but is not. Most growing businesses sit somewhere between Level 1 and Level 2.
Level 1: Personal AI
What AI sees: only what individuals choose to feed it. A pasted document. An uploaded file. A saved note in a chat memory. Each user has their own private context.
What AI does: drafts, summarises, brainstorms. It does not act on systems of record.
Who can extend it: each user, on their own. Power users figure out clever workflows; their colleagues never see them.
How the business has changed: it has not. The org chart looks the same. The hiring plan looks the same. Some staff are faster at certain tasks. Most are not.
Hard test. If your top AI user left tomorrow, would the AI capability go with them? At Level 1, the answer is yes.
Common false positive. Hiring a Head of AI without profit and loss accountability does not move you off Level 1. Buying more AI products does not move you off Level 1. The title is not what matters. The accountability is.
This is consistent with what Microsoft observed in its 2026 Work Trend Index. Roughly 10 percent of AI users sit in what the report calls blocked agency: fluent in AI, stuck inside an organisation that cannot use them. AI-fluent staff inside an unchanged business produce blocked agency, not progress.
Mark Cuban put the same point more directly in March 2026. AI is the nervous system, not the brain. The brain is your operations. Most of the 33 million companies in the United States are running on the same operations they had two years ago and asking why AI is not delivering. The structure underneath is the answer.
Level 2: Siloed Team AI
This is where most growing businesses that have invested in AI end up.
What AI sees: each team has shared context within its own boundaries. The sales team has a shared prompt library. Operations has its own approach. Finance has built a dashboard generator. Each function has its own AI stack.
What AI does: functional workflows. AI helps sales draft proposals faster. It helps operations triage incoming work. It helps finance generate reports. The actions are bounded to the function.
Who can extend it: within a team, non-engineers can use the workflows that exist. Across teams, almost nobody can. Each function privately rebuilds the same kinds of tools.
How the business has changed: each team is faster at its own work. The org chart is largely the same. The boundaries between functions still hold.
Hard test. Does AI output in one team reach another team without a human carrying it? At Level 2, the answer is no.
Common false positive. Every team is faster. Each leader can demonstrate clear improvement in their area. The business feels like it is winning. It is not. The business has become a collection of AI-enhanced silos.
The trap of Level 2 is that it feels like progress. Compared with Level 1, where AI was an individual habit, Level 2 looks systematic. There are shared prompts. There may be metrics. But the business itself has not changed shape. The handoffs between teams still go through people. The data still does not flow.
Many growing businesses are sitting inside this gap today. They have moved past the disorganisation of Level 1 and are now feeling the limits of Level 2 without having a name for them.
Level 3: Connected AI
Level 3 is where AI starts to function as part of the business rather than alongside it. This is the level most growing businesses need to reach.
What AI sees: the whole business. Customer enquiries, quotes, invoices, payments, project status, supplier records, all available through the systems where they live. The business has become legible to a machine.
What AI does: agents act on signals from the systems where work happens. A new enquiry flows through to a customer record without anyone re-typing it. Overdue invoices generate follow-ups without someone chasing the list. Completed projects close out across the relevant systems automatically. Cross-functional, bounded, observable. The work that used to live in someone's inbox now lives in a workflow.
Who can extend it: non-engineers in the business build new workflows. The sales team can wire an agent to triage incoming leads. Operations can configure an agent to manage scheduling exceptions. The business compounds what it knows.
How the business has changed: the way work is divided is materially different. Some roles look very different from how they looked two years ago. Some have been combined. Some have been redesigned around what humans do best, with agents handling the rest.
Hard test. Can an agent take a customer enquiry through to invoiced and paid without being chaperoned at every stage? At Level 3, the answer is yes for routine cases. Approvals and exceptions come back to humans.
Common false positive. One impressive end-to-end workflow does not make a business Level 3. Level 3 is the operating posture across the business. If the demo case works but everything else still flows through people, the business is closer to Level 2 with a hero exception.
Reaching Level 3 requires three structural pre-conditions. Processes mapped, so AI knows what should happen. Systems integrated, so AI can read across them. Ownership defined, so AI knows who handles exceptions and who approves certain actions in the workflow. Without these, Level 3 is impossible. AI cannot read what is not structured. AI cannot act in a business where nobody owns the outcome.
This is the core of the GrowthPains thesis. Structure first. Automation second.
Level 4: Self-improving AI
Level 4 is where the system starts to learn from itself.
What AI sees: not just events but the relationships between them. The system maintains its own context. Last quarter's outcomes inform this quarter's actions.
What AI does: agents have decision authority within scoped domains. Within defined boundaries, agents act, learn from outcomes, and adjust. Humans review where consequence is high.
Who can extend it: non-engineers ship working tools without filing a request. A finance person builds an automated contract reviewer. A salesperson builds a prospect research workflow. The boundary between user and builder dissolves.
How the business has changed: hierarchy has shifted. Some roles look more like managing a workflow than executing one. Compensation may be tied to AI proficiency.
Hard test. Do non-engineers in your business ship working AI tools, and does the system improve from its own use? At Level 4, the answer is yes.
Common false positive. Agent sprawl. A hundred small automations nobody owns is not a self-improving system. It is a maintenance problem in waiting.
Most growing businesses are not at Level 4 yet, and that is fine. Most do not need to be. Reaching Level 3 first is the work that matters.
Above Level 4 sits Level 5. A self-driving organisation that notices, decides, acts, and learns without human prompting. It is theoretical for now. Almost no business operates there yet. The question for growing SMEs is not how to reach Level 5. It is whether the business is describable enough for any of this to work at all.
Why most small businesses are stuck at Level 1 or Level 2
Over the past six months we have spoken with business owners and founders across a range of growing companies. Every one of them told us their team uses ChatGPT or Claude. Not one had it wired into the systems where the actual work happens. People drafting emails. People summarising calls. People using AI privately, on the side. Useful in the moment, invisible to the business. By the diagnostic in this post, every business we spoke with was sitting at Level 1.
The cause is structural, rather than technical.
In May 2026, Daniel Miessler made the point bluntly. AI is execution. Execution is powerless when it does not know what to execute. The companies AI is helping are the ones that can already articulate what they are solving for customers, what their goals are, which metrics they are moving, where the work happens, and at what cost. Most growing businesses cannot answer those questions consistently across quarters. AI cannot fix that. It can only accelerate whatever the business already is.
Microsoft made the same point with vendor data. The 2026 Work Trend Index found that organisational factors, not individual capability, do most of the work in determining whether AI delivers value. The constraint is not the AI. It is the operating model: how the business runs day-to-day across people, process, systems and data, and governance. AI cannot redesign work the business has not redesigned itself.
In GrowthPains terms, this means processes are not mapped, systems are not connected, ownership is not defined. AI cannot read what is not there. AI cannot act in a business where nobody owns the outcome. Adding more AI tools to a disorganised business produces a faster disorganised business.
Hiring more AI-fluent staff does not fix this. They become the blocked agency Microsoft described. Buying more tools does not fix it. The new tools sit on top of the same disconnected operations. Running another tool review does not fix it. The review identifies the same gaps that were identified last year.
The work that fixes Level 1 and Level 2 is the work most growing businesses keep avoiding. It is unglamorous. It is structural. It is the operating model.
The asymmetry tells you what to fix first
Most businesses do not score evenly across the four diagnostic questions.
A business may have AI that can see a lot (clean data, integrated tools, accessible records) but no agents wired to systems of record. AI knows everything and does almost nothing. Investment goes into agent integration, not more data work.
Another business may have AI that can do useful things (workflows running, tasks moving) but only one or two technical people can extend it. Investment goes into low-code tooling and process documentation so more people can build.
Another business may have built useful tools and have non-engineers extending them, but the business itself is still hiring against the 2024 plan. Capacity has expanded, the structure has not caught up. Investment goes into role redesign.
The asymmetry is the diagnostic output. The lowest-scoring question of the four is the next intervention. Score each question Level 1 to Level 4. The gap is what you fix first.
This is one of the reasons a generic "adopt AI" programme rarely works. It treats every business as if the constraint were the same. It is not.
Legibility is the moat, not AI
The competitive battlefield has temporarily shifted away from AI itself.
Before AI becomes a moat, organisational legibility is the moat. The ability to articulate goals, workflows, decisions, and economics in a form a machine can read.
Microsoft's 2026 study found that organisational factors (culture, manager behaviour, team practices) accounted for roughly twice the AI impact of individual effort. The operating model does the heavy lifting. Companies that treat AI readiness as a tooling or training problem are misdiagnosing it. The real prerequisite is structural clarity.
The first wave of disruption is unlikely to come from AI-native giants out-AI-ing incumbents. It is more likely to come from small coherent operators outcompeting larger chaotic ones because they can actually deploy AI against well-defined work.
This is the SME asymmetry. A 15-person business can answer the goals-workflows-economics questions in an afternoon. A 10,000-person business needs a project. A coherent small business has, for the first time in a generation, a structural advantage over a less coherent large one. The window favours the small and the clear.
The strategic question flips. The question is not "what can AI do for us?" The question is "are we describable enough for AI to do anything for us at all?"
Most growing businesses cannot answer that second question, and most do not realise it is the right question.
The shift that matters
The shift is not from less AI to more AI. It is from one question to another.
From "where can we use AI?" to "which metric are we moving?"
Level 1 and Level 2 keep asking the first question. Level 3 and Level 4 only exist because the business has answered the second.
In March 2026, HSBC made this shift visible at scale. The bank appointed its first Chief AI Officer. The choice was deliberate. David Rice is an 18-year HSBC veteran. Most recently he was the Chief Operating Officer for HSBC's Corporate and Institutional Banking business. Not a data scientist. Not a chief technology officer. A business operator with profit and loss accountability, tied to a 17 percent return on tangible equity target for 2026 to 2028.
The signal: AI is not a tooling initiative. It is an operating model decision, owned by someone whose name is on the financial outcome.
The same logic applies at smaller scale. The SME equivalent is not hiring a fractional Chief AI Officer or paying a consultancy to write a strategy. It is the founder owning AI as a profit and loss outcome rather than delegating it. Owning the metric AI is supposed to move. Owning the workflow it operates within. Owning the financial impact it produces.
If the founder does not own those three things, nobody in the business will.
How to assess AI maturity in your small business
Use the four diagnostic questions as a self-assessment.
1. What can AI see in your business today?
Level 1: each individual sees only what they paste in.
Level 2: each team has shared context within its own boundaries.
Level 3: AI can read across the whole business through systems integration.
Level 4: AI sees relationships between events, not just events.
2. What can AI actually do in your business today?
Level 1: drafts, summaries, brainstorms for individuals.
Level 2: functional workflows within team boundaries.
Level 3: agents act across systems with bounded authority.
Level 4: agents have scoped decision authority and learn from outcomes.
3. Who in your business can extend AI workflows?
Level 1: each user, on their own.
Level 2: a few power users, within their team.
Level 3: non-engineers, across functions.
Level 4: anyone, including non-engineers shipping production tools.
4. How has the way your business runs changed in the past two years?
Level 1: it has not.
Level 2: each team is faster, structure is the same.
Level 3: roles, hiring, and decision-making are materially different.
Level 4: hierarchy has shifted toward managing AI workflows.
Score each question. The lowest score is the next intervention.
Operating-model redesign compounds in a way that AI procurement does not. Tools can be bought in a day. Operating models take months. The advantage compounds for businesses that start now, and it widens against businesses that keep buying tools and hoping.
Stop asking where AI fits. Start asking which number it moves, and whether the business is describable enough for it to move anything at all.
Not sure where to start? Take our free Operational Health Check. It takes 2 to 3 minutes and shows you where to focus first.
If you already know what needs fixing, book a call.
Stats Referenced in This Post
Approximately 10 percent of AI users sit in "blocked agency." Microsoft, 2026 Work Trend Index Annual Report.
Organisational factors account for roughly twice the AI impact of individual effort. Microsoft, 2026 Work Trend Index Annual Report.
HSBC appoints first Chief AI Officer; 17 percent return on tangible equity target for 2026–2028; David Rice is an 18-year HSBC veteran and former COO of Corporate and Institutional Banking. HSBC press release, 23 March 2026.
The four-level AI maturity diagnostic. Ann Miura-Ko (Partner, Floodgate VC) , X post, 1 May 2026.
"AI is the nervous system, not the brain"; 33 million companies in the United States. Mark Cuban, TBPN interview, 19 March 2026.
"AI is execution. It is powerless when it does not know what to execute." Daniel Miessler, Most Companies Aren't Anywhere Near Ready for AI, 2 May 2026.
Subscription pricing arithmetic (Microsoft 365 Business Standard, Microsoft Copilot, ChatGPT Plus, Claude Pro): figures are illustrative rounded estimates based on standard 2025/2026 UK SaaS pricing.