A plumbing company with a €9,000 average job, a law firm with a €20,000 retainer, a property manager with 400 doors under contract — these businesses now have AI. They have a chatbot, a drafting assistant, a scheduling tool. Adoption happened. It happened quickly, and the survey evidence across four markets says it kept happening through 2025 and into 2026.
What did not happen at the same speed is integration. The AI sits beside the workflow rather than inside it. It drafts the email a person still has to send. It answers a question the business never routes to a person who can quote. It summarises a call that was never picked up.
Adoption is spreading faster than operational integration. That gap is where the money is.
Adoption is no longer the constraint
Four separate research programmes measured AI use in business between 2024 and 2026. They are different studies, in different countries, asking different questions of different populations. Read together they establish one thing clearly: access to AI is broadly solved.
| Market | Reported figure | What was actually measured | Source |
|---|---|---|---|
| United States | 58% | Of U.S. small businesses, the share self-identifying as generative-AI users. Up from 40% in 2024 and 23% in 2023. | U.S. Chamber of Commerce (C_TEC) with Teneo Research, 2025 |
| International | 75% | Of 3,350 SMB leaders across 26 countries and six continents, the share investing in AI in some capacity — 34% fully implemented, 41% experimenting. Fielded 3 Aug – 16 Sep 2024. | Salesforce, Small & Medium Business Trends, 6th Edition |
| Canada | 71% | Of 300 Canadian SMB decision-makers (1–250 employees), the share reporting use of AI and/or generative AI. Fielded by Edelman, 10–24 Jan 2025. | Microsoft Canada, fifth annual SMB report |
| United Kingdom | 16% | Of 3,500 UK businesses interviewed by telephone, the share using at least one AI technology. Among those adopters, only 7% reported using agentic AI. | DSIT AI Adoption Research (IFF Research & Technopolis) |
| United Kingdom | 41% | A different sample: of UK businesses that handled digitised data, the share using AI-based technologies. Of those AI users, 21% said their AI tools were integrated into existing business systems. | UK Business Data Survey 2026 |
| France | 18% | Of French enterprises with 10 or more employees, the share using at least one AI technology in 2025 — 8 points above 2024. Among non-adopters, 54% cited a lack of relevant expertise. | INSEE, Insee Première n° 2120 |
58%, 75%, 71%, 16%, 41% and 18% are different populations, not one global sample. They differ by country, by company size, by year of fieldwork, and above all by the definition of “using AI”.
Canada makes the definitional point on its own: Microsoft measures 71% by counting broad AI and generative-AI tool use, while Statistics Canada measures 12.2% by counting AI in production and service delivery. Same country. Same year. A ~59-point spread produced entirely by what the question counted. Averaging figures like these into a single headline percentage destroys the only information they contain.
The integration gap is measurable
Two of these studies asked the second question — not “do you use AI” but “is it inside your systems”. Both found the same shape of answer.
The UK figure and the property-management figure come from unrelated samples in different countries, and they are not two measurements of one quantity. What they share is the shape of the finding: a large adopting majority, a small integrating minority. The UK government research adds the third data point — among UK businesses that had adopted AI, 85% were using it for natural-language and text generation, and only 7% for anything agentic. Drafting is everywhere. Doing is rare.
France names the reason directly. Among French businesses not using AI, 54% cited a lack of relevant expertise — an obstacle reported most often by the largest companies in the sample. The barrier is not the model. It is the absence of anyone who can put the model inside an operating process and be accountable for the result.
The workflow that actually produces revenue
A high-ticket local business does not convert a lead in one step. It moves a customer through a chain of states, and each move is a place where the work can stop.
Most AI purchases improve a box. A better drafting tool makes the quote read well. A better transcription tool makes the call searchable. Neither changes whether the quote was sent, or whether the call was answered.
Where high-ticket local businesses lose money
The losses cluster at the transitions. Each one is a customer who was willing to pay, and a business that did not complete the move.
What the operational evidence measures
Three studies measured these transitions directly, in three sectors where the job value is high enough that a single failure is material. Two are vendor research and one is agency-run secret shopping; each is labelled below and in the sources.
| Sector | Transition | Measured | Evidence type |
|---|---|---|---|
| Home services | signal → identity | 52% of callers to home services businesses speak with a person. That rises to 65% for calls over 15 seconds and 73% past 30 seconds. | Invoca, 2026 Home Services benchmarks — vendor platform data, 70M+ calls |
| Home services | context → qualification | 38% of digitally generated calls answered by a person qualify as leads. | Invoca, 2026 — vendor platform data |
| Home services | next action → commitment | 45% of those leads convert during the call itself. | Invoca, 2026 — vendor platform data |
| Legal | signal → identity | 48% of law firms were effectively unreachable by phone. Of 500 firms emailed, 33% responded — down from 40% in 2019. | Clio Legal Trends Report 2024, secret shopping of 500 firms by Lux |
| Legal | context → qualification | Of firms that did reply by email, 84% replied within eight hours — but only 18% gave clear next steps or cost information. | Clio Legal Trends Report 2024 — secret shopping |
| Property management | execution → outcome | 58% AI tool adoption, and 8% reporting any fully automated workflow. | Buildium 2026 industry report — vendor research, 3,200+ respondents |
Read the home services chain end to end. Of every hundred people who call a home services business, roughly 52 reach a person. Of the answered calls from digital marketing, 38% are genuine leads, and 45% of those convert on the call. The single largest loss in that sequence happens before any AI touches anything: the call was not answered.
Legal makes the sharper point. Speed is not the failure — 84% of the firms that replied by email did so within eight hours. Quality of qualification is the failure: 18% gave the prospective client clear next steps or a cost. A firm can be fast, responsive, and still lose the instruction, because responding is not the same transition as qualifying.
Invoca and Buildium are vendor research drawn from their own platforms and customer bases. Clio commissioned an independent agency to secret-shop 500 firms. These describe the sectors studied, in the periods studied. They are not Compound Pulse results and are not presented as ours.
What a revenue system has to own
A tool improves a box. A system owns the moves between boxes, and can prove which move failed. That is the whole distinction, and it is the reason a business can hold a 58% adoption statistic and an 8% automation statistic at the same time.
Compound Pulse AI Systems is built to own and measure those transitions. Concretely, that means four commitments:
- Every signal gets an identity. A call, a form, a message and a walk-in resolve to one canonical customer record, so a second contact is never a first contact again.
- Every transition emits a receipt. Each move — qualified, routed, quoted, booked, delivered, approved — is written down with a time and an owner. A transition with no receipt is a leak, and it becomes visible the moment it stops happening.
- Qualification is a schema, not a judgement. Budget, authority, scope and urgency are captured as fields with defined values, so routing is deterministic and a high-value job reaches someone who can close it.
- Capacity is a live state. The available hour, crew or room is a fact the system can offer, not something a person has to remember to mention.
None of this requires a more capable model. Every market in the table above already has the models. What is missing is the operating layer between them and the revenue — the layer that France's 54% expertise gap describes precisely.
The measurement that matters
One number decides whether an AI investment in a high-ticket local business was worth making: the completion rate of each transition, before and after. Not tokens. Not seats. Not hours notionally saved.
If 48% of inbound cannot reach a person, the first system to build is the one that answers, identifies and qualifies — and the metric is the share of signals that reach a qualified opportunity with an owner. If quotes go out and nothing owns the second contact, the metric is the share of quotes with a recorded follow-up inside the buying window. A system that cannot report those numbers has not been integrated. It has been installed.
Adoption was the last decade's question, and the surveys have answered it. Integration is this one.