Field guideMiddle marketVersion 1.023 October 2026
AI in production for the middle market
A middle-market company puts AI into production by choosing one workflow that already costs it money, running it on models it can replace, with a named owner, a full record and a way to stop, and by paying for the work in phases that each end with a go or no-go. The surveys agree on the starting point: adoption is wide and shallow, and the obstacles they name are data, skills and legal clarity, not access to models.
On this page
Each figure names its source, its date and its definition.
In this page, "we" means Hikari Blue, which engineers and operates AI in production for mid-market and enterprise organizations in the United States, Switzerland and France.
The middle market has no single definition. The National Center for the Middle Market counts US companies with annual revenues between $10 million and $1 billion; RSM's 2026 survey uses $30 million to $10 billion in the US; the Census Bureau, Eurostat and the Swiss Federal Statistical Office classify by number of employees. Figures from different sources are not comparable, and none is added to another on this page.
Wide and shallow: most use AI, few have it in their core processes.
The US middle market counts about 200,000 companies, roughly one-third of private-sector GDP and about 48 million employees, according to NCMM. The table reads its AI adoption from 3 US sources, then from Europe and Switzerland for firms of comparable size.
| Measure | Figure | Source and date |
|---|---|---|
| US middle-market companies using AI | 9 in 10 | NCMM, Mid-Year 2026 Middle Market Indicator, 1,000 executives, fielded June 2026 |
| AI fully embedded across core processes | 36%, while 86% report AI partially or fully integrated | RSM US Middle Market AI Survey 2026, 1,030 respondents, fielded 5 to 16 March 2026 |
| US adopters using AI in 3 business functions or fewer | 57% | US Census Bureau, working paper CES 26-25, April 2026, reference period November 2025 to January 2026 |
| US firms using AI with no institutional adjustment | 64%; about 15% each trained staff or built new workflows | Same paper, CES 26-25 |
| EU enterprises using AI, 50 to 249 employees | 30.36%; France 30.81% | Eurostat, use of AI in enterprises, reference year 2025, news release of 11 December 2025 |
| Swiss enterprises using AI, 50 to 249 employees | 15.6% in 2023 | Swiss Federal Statistical Office, from the KOF innovation survey, data of 19 May 2025 |
Eurostat covers enterprises with 10 or more persons employed, outside the financial sector. The Swiss figure covers enterprises with 5 or more employees and uses a simplified version of the Eurostat definition.
Data, skills, legal clarity, legacy systems. None of them is a model.
Data quality
The first inhibitor in RSM's 2026 survey, at 34%. NCMM finds that 83% of middle-market executives express AI concerns, data quality among them, with privacy, implementation costs and skill atrophy.
Skills
Among EU enterprises that considered AI and did not use it, 70.89% cite a lack of relevant expertise (Eurostat, 2025). In the US, 28% name talent and skills (RSM), and 52% already work with external consultants or advisors.
Legal clarity and data protection
In the same Eurostat population, 52.52% cite a lack of clarity about legal consequences and 48.83% data protection and privacy concerns. RSM finds security and privacy at 30%.
Legacy systems
28% of RSM's respondents name legacy systems. An AI use that cannot write to the ERP or the CRM stays a demo.
Governance is not the missing piece everywhere: 67% of RSM's respondents apply governance controls before pilot or production. What the data shows missing is depth, the step from a tool many people use to a workflow that runs under control.
One workflow, then the next, each paid on its own evidence.
A middle-market company rarely has a platform team to spare. The path below is built so that no step requires one, and so that your people run the result when we leave.
- 01
First exchange · free
Describe the workflow and its principal constraint. A named senior partner answers in writing, with a clear reason if it does not fit.
You leave with: a written answer on whether the workflow is worth building, governing or stopping.
- 02
Operating Assessment · 1 week, from $18,000
1 week of work on one critical workflow: the baseline, the value, the risk, and the controls it needs before it runs in production.
Deliverables: Time to Stop established, the workflow's value quantified, a named owner, a plan to put it into production.
- 03
Engineering and operation · in phases
The workflow is engineered into your systems and deployed in stages against the baseline. Each phase ends with a go/no-go, so you never pay for the next one on a promise.
Evidence: before-and-after report, cost per operation with the full cost of the AI counted, the record of every action.
- 04
Transfer · on a dated plan
Operation moves from us to your team on a dated plan, written in the contract from day one, or through an on-site AI pod that works inside your teams until the use runs, then hands over.
Evidence: runbooks, dated literacy records of the people who operate and verify, the owner's sign-off.
4 choices that keep the cost and the risk where a mid-sized company can carry them.
Replaceable models
Each task runs on the model it needs, frontier models by API or local open-weight models in your cloud or on premises. Switching provider is a routing change, not a rebuild.
Full costs, counted
Gains are reported net of integration, maintenance, inference and supervision. A use that does not pay for itself shows it in the before-and-after report.
A named owner and a named partner
One person on your side owns each workflow; one partner on ours answers for the engagement. Nobody hides behind a platform.
Compliance as a record, not a department
The record of every action, the stop path and the literacy records answer the regulator without a compliance team built for the occasion.
Size exempts no one. The rules follow the use, in each market.
The texts below reach companies of every size when they use AI in the situations they cover. Each status was read in its primary source.
| Market | Text | What it asks | Status |
|---|---|---|---|
| United States | NIST AI RMF 1.0 | A voluntary framework of outcomes: govern, map, measure, manage. | Voluntary since January 2023; under revision |
| Texas | HB 149, Texas Responsible Artificial Intelligence Governance Act | Prohibited uses, enforcement by the attorney general, a defense tied to recognized frameworks. | In force since 1 January 2026 |
| European Union | Regulation (EU) 2024/1689, Article 4 | Measures to support the AI literacy of staff and other persons acting on your behalf. | Applies since 2 February 2025 |
| European Union | Regulation (EU) 2024/1689, Articles 12, 14 and 26 | Logs, human oversight and deployer duties for high-risk uses such as recruitment or creditworthiness. | From 2 December 2027 (Annex III) |
| Switzerland | Federal Act on Data Protection, Article 21 | Inform the person of an automated individual decision and allow a review by a natural person on request. | In force since 1 September 2023 |
This is not legal advice. Whether a text applies to a given system is a question for counsel.
- NIST AI RMF implementation, control by control The 15 controls mapped to NIST AI RMF 1.0 subcategories, with their evidence.
- AI literacy and Article 4 of the EU AI Act What Article 4 requires, since when, and the records that show it is met.
- AI audit trail: definition and EU AI Act mapping The 9 fields of the record, mapped to Articles 12, 19 and 26.
- The EU AI Act for Swiss companies When the EU regulation reaches a company established in Switzerland, and what it requires.
What mid-sized companies ask before the first workflow.
What counts as the middle market?
In the US, the National Center for the Middle Market counts companies with annual revenues between $10 million and $1 billion, about 200,000 companies. Other sources use other bounds, such as $30 million to $10 billion for RSM, or employee size classes for statistical offices.
How many middle-market companies use AI in production?
Use is wide: 9 in 10 US middle-market companies report using AI (NCMM, June 2026). Production depth is not: 36% have AI fully embedded across core processes (RSM, March 2026), and 57% of US adopters use it in 3 business functions or fewer (Census Bureau, April 2026).
Do we need an AI team before we start?
No. We engineer and operate the workflow with your team, then hand it over on a dated plan, or place an on-site AI pod that works inside your teams until the use runs. What you need from day one is a named owner for the workflow.
What does it cost to start?
The first exchange is free. The Operating Assessment takes 1 week on one workflow, from $18,000. Engineering and operation are scoped from it and paid in phases, each ending with a go/no-go.
Are we locked into one AI vendor?
No. The architecture runs frontier models by API (Anthropic, OpenAI, Mistral, Google) and local open-weight models in your cloud or on premises, by configuration. Switching models is a routing change, not a rebuild.
Do AI rules apply to a company our size?
Yes, when the use falls within them. Article 4 of the EU AI Act reaches providers and deployers of any size; high-risk duties follow the use, from 2 December 2027 for Annex III systems. The EU Act lowers fines for SMEs, it does not exempt them.
Reference documents and primary sources.
Every figure and status on this page comes from the sources below, read on 23 October 2026.
Reference documents
- Enterprise AI Agents The offer: agents engineered into your workflows, measured on your operations, full costs included.
- Enterprise AI agents implementation, phase by phase 4 phases, 3 controls on every agent and the evidence each phase leaves.
- Modernize your platform and put AI into production When legacy systems hold AI back: modernization and the AI operating layer as one engagement.
- On-site AI pod A team of senior operators inside your teams until the use runs, then the handover.
- AI Operating Layer Reference Architecture v1.0 4 planes, 4 testable controls and the 9-field record.
- Nvidia booked $96 billion. Most US AI adopters stop at 3 business functions. Briefing of 1 September 2026 on the Census Bureau's measure of AI depth.
Primary sources, consulted on 23 October 2026
- National Center for the Middle Market · Middle Market 101 (opens in a new tab) Definition: US companies with annual revenues between $10 million and $1 billion.
- NCMM · Middle Market Indicator, Mid-Year 2026 (opens in a new tab) Fielded June 2026, 1,000 executives: AI use, concerns, investment. PDF.
- RSM · Middle Market AI Survey 2026 (opens in a new tab) Released 21 July 2026, fielded 5 to 16 March 2026, 1,030 respondents.
- RSM · Middle Market AI Survey 2026, building blocks of AI adoption (opens in a new tab) Use of external consultants or advisors, current and planned.
- US Census Bureau · CES 26-25, The Microstructure of AI Diffusion (opens in a new tab) Bonney et al., April 2026: depth of use and organizational change.
- Eurostat · 20% of EU enterprises use AI technologies (opens in a new tab) News release of 11 December 2025, reference year 2025; size classes from dataset isoc_eb_ai.
- Swiss Federal Statistical Office · Other ICT use in enterprises (opens in a new tab) AI use by size class in 2023, from the KOF innovation survey, data of 19 May 2025.
- Texas Legislature Online · HB 149, bill history (opens in a new tab) Signed 22 June 2025, effective 1 January 2026.
- Fedlex · Federal Act on Data Protection (FADP), SR 235.1 (opens in a new tab) In force since 1 September 2023, Article 21 on automated individual decisions.
How to cite this document.
This URL is permanent and always serves the current version. Each change of substance increments the version and is listed in the changelog below, so a citation stays traceable to the text it quoted.
Preferred citation
Hikari Blue, "AI in production for the middle market v1.0", 23 October 2026, https://hikariblue.com/ai-in-production-middle-market
Section anchors are stable: #status, #data, #obstacles, #path, #design, #rules, #faq, #references, #cite.
Changelog
v1.0 · 23 October 2026. First public version.
Editorial correction · 25 October 2026: license extended to this page (CC BY 4.0).
License. Published under CC BY 4.0: free to cite, quote and reuse with attribution to Hikari Blue and a link to this page. The license covers Hikari Blue’s text, tables and diagrams on this page; it does not cover third-party data, quotations or legal texts, which keep their own terms, nor the Hikari Blue name and logo, trademarks of UNITED4 LLC.
Bring the workflow that costs you the most.
30 minutes with a named partner. We look at one workflow you run today and tell you, in writing, whether it is worth building, governing or stopping, what the first phase would prove and what it would cost. You leave with that answer, whether or not we work together.
- One workflow, judged on value and risk
- What the first phase would prove
- What it would cost, full costs included