The most credible economic case for generative AI in a small professional firm is not “replace the professionals”.
It is increase the amount of high-quality work each professional can supervise.
Research, document transformation, first drafts, structured comparisons and administrative preparation can consume enormous amounts of time without themselves being the final source of professional value. AI can compress part of that work.
The judgment that makes the work trustworthy remains a different function.
Key takeaways
- AI leverage comes from reducing the cost of iteration. Research, structuring, drafting and transformation can often begin faster.
- Automation and accountability must remain separate concepts. A system producing an answer does not transfer professional responsibility to the system.
- The likely operating model is leaner, not human-free. Current evidence supports meaningful productivity potential but not a simple story of wholesale job removal.
What was visible by late 2023
By the end of 2023, generative AI had already moved beyond a technology demonstration. ChatGPT and GPT-4 had shown that general-purpose language models could support writing, coding, summarisation and analytical workflows.
What was not yet established was the size of the productivity effect across ordinary firms or whether that productivity would translate into fewer employees, greater output or simply more work being attempted.
Interpretation: leverage is a better model than substitution
Professional businesses contain different types of work.
Some tasks require judgment, accountability, negotiation and contextual understanding. Others require finding, reorganising, comparing, formatting or transforming information.
Historically, both categories consumed expensive human time.
Generative AI matters because it can lower the cost of the second category and therefore leave more professional time available for the first.
The mechanism
A lean professional workflow can increasingly look like this:
source material → machine-assisted extraction and structuring → human analysis → machine-assisted drafting or challenge → professional verification → final accountable output.
If the first machine-generated interpretation is accepted as truth, the technology can scale mistakes. If it is treated as an intermediate layer, it can scale preparation.
This is why AI can produce leverage without justifying “fully automated advice”.
Evidence since 2023
OECD research published in 2025 found GenAI use among 31% of more than 5,000 SMEs surveyed in seven countries. Among users, many reported stronger performance, while 83% said their overall staffing needs had not changed.
ILO’s 2026 review similarly found real but uneven productivity effects and limited evidence, so far, of large-scale employment displacement attributable to generative AI.
That combination supports a more restrained conclusion: AI can change how work is organised before it changes how many people organisations employ.
The strongest countercase
AI can also create hidden work.
Outputs must be checked. Confidential information must be protected. Poor prompts and weak source material produce poor results. Employees need training. Tool proliferation can fragment rather than simplify workflows.
In high-stakes professional work, reviewing a sophisticated but incorrect answer can sometimes be harder than producing the answer properly in the first place.
The relevant measure is therefore reliable output per unit of total professional effort.
Scenarios, not forecasts
A lean boutique uses AI extensively for preparation while senior people retain analysis and client responsibility.
An agent-assisted firm delegates increasingly complex internal workflows to software while maintaining explicit approval gates.
An automation plateau emerges if verification, confidentiality and liability costs rise almost as quickly as model capability.
Practical consequences
For a professional firm, useful implementation starts with process mapping.
Which activities are repetitive? Which require sources? Which contain confidential data? Where can an error cause material harm? Who approves the result?
Those questions should determine the technology architecture.
The international dimension matters too. If AI allows a small firm to serve more countries with fewer people, the firm’s physical footprint may shrink while the importance of the remaining footprint increases. Founder residence, management location, employee presence, banking, data handling and regulated activities remain jurisdiction-specific.
AI can make a firm lighter.
It cannot make responsibility disappear.
Sources
- OpenAI, GPT-4 Research: https://openai.com/index/gpt-4-research/
- OECD, Generative AI and the SME Workforce, 2025: https://www.oecd.org/en/publications/generative-ai-and-the-sme-workforce_2d08b99d-en.html
- ILO, The impact of GenAI on jobs, productivity and work organisation, 2026: https://www.ilo.org/publications/impact-genai-jobs-productivity-and-work-organization-review-empirical
Disclaimer
This Insight provides general business and technology analysis. It is not legal, tax, employment, data-protection or professional-regulatory advice. Appropriate controls depend on the activity, information processed and jurisdictions involved.
