Subtraction, Not Substitution
September research on content marketing, customer service and internal audit raises one practical question: when an organisation removes human work, how does it check that the technology has taken over the whole job?
Research published in September raises a practical question about AI adoption: when an organisation removes human work, how does it check that the technology has taken over the whole job?
AI business cases often assume that a new system will absorb work previously done by people. The expected savings depend on that assumption.
Reducing a budget can happen immediately. Checking that the work is still being done properly takes longer.
September findings on content marketing, customer service and internal audit suggest that this gap deserves closer attention. The studies measure different things and cannot establish a common cause. They do, however, raise questions about the capabilities organisations maintain as they adopt AI.
Content marketing is losing practices associated with better results
Orbit Media has surveyed content marketers annually since 2014. Its September edition records the weakest results in the survey’s history: fewer than 14% of respondents report strong results, roughly half the proportion recorded in 2022. Nearly one in five cannot say whether their content delivers results at all. [1]
AI adoption among respondents exceeds 92%. The survey finds no measurable relationship between AI use and stronger results.
It does find a decline in practices historically associated with success: original research, collaboration with outside experts, keyword research, human editing, paid promotion and consistent measurement.
Collaboration with external experts has the strongest association with results, at 2.6 times the benchmark. Yet the proportion of marketers using it has fallen from 25% in 2017 to 7% today.
The survey does not establish why organisations have reduced those activities. It does leave a question for anyone investing in content generation: how will the organisation maintain the research, expertise and distribution that help content produce results?
Customer service savings need a longer accounting period
Two Gartner releases in September describe different parts of the same operational problem.
A survey of more than 3,500 customers found that only 27% would try a chatbot again after a single bad experience. It also found that 49% would have used one had it been offered, while just 7% actually did. [2]
There is willingness to use these systems. Companies still have to make them accessible and reliable enough for customers to return.
A separate Gartner forecast estimates that 30% of employees laid off in AI-driven reductions will need to be rehired by 2029, often at higher cost. Among the customer service AI use cases Gartner evaluated, roughly a quarter returned positive value, a quarter returned negative value, and 42% could not be assessed. [3]
The rehiring figure is a forecast. It identifies a possible future cost that deserves a place in today’s investment decisions.
Staff reductions and lower costs per contact appear quickly in the accounts. Unresolved problems, lost customers and depleted expertise can take longer to measure. Their effects may also appear in another department’s budget.
A credible assessment needs to account for those later costs and check that customers can still get their problems resolved.
Oversight requires operational authority
In a Gartner webinar poll of 161 chief audit executives, 93% of audit functions reported using AI. 60% lacked a formal strategy, only 15% had defined use cases, and more than half had not started measuring the return. [4]
The sample is small and self-selected. It cannot represent every audit function. It suggests that even teams responsible for evaluating controls can adopt AI before establishing how to manage its use.
ABBYY’s research, conducted by Opinium across six markets, adds another concern. Among leaders whose organisations already use AI, 82% reported a formal governance framework. Only 35% reported having a way to switch the system off. [5]
Effective oversight needs practical authority. Someone must be able to identify a problem, take responsibility and intervene.
That requires time and resources. In the ABBYY research, 70% said data governance requirements slowed deployment, while 57% said governance made it easier to move beyond the pilot stage. Work done during deployment can help a system operate reliably at scale.
Qualitative research by Kris Johnson Ferreira and Jordan Tong, covering five firms, describes a division of responsibilities: agents conduct analyses and surface trade-offs; people provide context, set guardrails and make final decisions. [6]
That human contribution remains part of the operating cost.
What organisations still need to pay for
The geographical limits of this evidence matter. The content research provides no geographic breakdown, and the Gartner customer service figures cited here do not specify one. ABBYY includes the United Kingdom, France and Germany among its six markets. These findings provide limited grounds for conclusions specific to Switzerland or Italy.
The practical question remains relevant to the organisations I work with in both countries: which capabilities must stay in place for an AI investment to deliver its expected value?
AI has sharply reduced the cost of producing many kinds of output. Original research, external expertise, editorial judgement, reliable escalation to a person and accountable system ownership still require investment.
Before removing an activity or a role, an organisation needs to identify everything it contributes and decide how each function will be covered. After implementation, it needs evidence that the arrangement works.
That is the distinction behind “subtraction” and “substitution”. A reduction in spending is easy to record. Establishing that the full job is still being done takes more work.
— Stefano
Sources
- Andy Crestodina, “Blogging Statistics 2026: What 1,042 Content Marketers Told Us About What Works”, Orbit Media Studios, 8 September 2026.
- Gartner, “Gartner Survey Finds Only 27% of Customers Would Try a Chatbot Again After a Negative Experience”, 2 September 2026.
- Kristen Doerer, “Are AI-driven customer service cuts here to stay? Gartner predicts rehiring”, CX Dive, 11 September 2026, reporting Gartner’s release of 9 September.
- Gartner, “Gartner Survey Finds 93% of Audit Functions Use AI, but 60% Lack a Formal Strategy”, 10 September 2026.
- ABBYY and Opinium, “Who Answers for AI: The Governance Gap”, 9 September 2026.
- Kris Johnson Ferreira and Jordan Tong, “How AI Agents Orchestrate Work Across Silos”, Harvard Business Review, September–October 2026.