Edition 02 · September 2026 · current
Your company already adopted AI. The result is not showing up, and technology is not the reason.
Preparation was automated and nobody redesigned the review. When that happens, three things occur at once, and none of them shows up on a dashboard.
In one year, finance functions with AI embedded in their processes went from 22% to 42%. In a separate measurement, across 1,719 companies of every kind, the share that can show a significant impact on operating results is still 6% and did not move. They are two different samples and they do not subtract from each other, but they point at the same hole: adoption is fast and the result does not show up.
That distance is not explained by the models, which got better. It is explained by where the automation went. In almost every organization, preparation was automated: whoever builds the report, reconciles the account, drafts the first version of the contract. Nobody redesigned the review.
01
The cost moved to a different pocket
You measured the saving where it is always measured: hours, people, cycle time. That saving is real and it sits in your budget. The new cost does not.
Your business case did not fail: it was paid from another pocket, which is why the saving shows in your area and not in the consolidated result. The first projects to go are the ones that cannot show the whole number.
02
The review was left without an owner
Preparation is easy to automate because it repeats. Review is not, because it is judgment. In practice the system prepares, and the same person who used to review now approves faster and with less context.
A separation that stood for decades quietly broke: whoever prepares does not approve. It works until the day somebody asks who authorized a decision from last month, on what data, and who reviewed it.
03
The rung that trained people broke
This is the part that shows up on no dashboard. Preparation was not just work: it was the training. Nobody learns to review a contract without having drafted fifty, or to close a month without having reconciled a hundred accounts.
This year the employment gap for ages 22 to 25 came in 19% below what was expected, up from 15% the year before, and it is not layoffs: it is less hiring. Added up, those decisions mean fewer people able to review in 2031, exactly when there is far more to review.
04
The map by function
Where AI is entering inside the company, function by function, and what each figure says about the same problem: preparation was automated and the review stayed as it was.
The function that moved money to AI fastest, and the first to run short: 56% say the budget does not cover what they already started.
How much the use of sales agents grew in three months. It is also the first function where vendors had to add hard spending caps.
Predictive maintenance doubled. It is the adoption with the easiest return to measure, because the failure you avoided counts itself.
Report a productivity gain, and only 20% better decision quality. The gain is real in the task and not yet in the judgment.
A class action over automated candidate screening has already been served, with age and race claims moving forward. It is the only function where the risk is already judicial.
That is the difference between handling complexity well or badly, and 97% of organizations ran a complex project last year. The gap is not in the tooling.
Mature governance and a named executive owner. An agent security top ten and an agent control standard were both published this year: the standard arrived before the owner did.
Seven functions, one pattern: adoption enters wherever the work repeats, and the review stays where it was.
05
How this looks in your industry
Banking and insurance
66%of European insurers already run generative AIThe three US banking regulators revised their model risk guidance and left generative and agentic AI out of scope. The model deciding on a customer does not fall under the framework your committee uses to review models, and that gap is filled by the contract and the policy.
Health
81%of physicians use AI, against 38% in 2023It did not arrive through an institutional project: it arrived through the clinical route, one person at a time. The question is no longer whether it is used, it is who answers for what the tool suggested and with what record.
Retail and consumer
138%growth in a year in traffic referred by AI modelsAnd it converts 54% better than the rest. It is the best traffic you have and it cannot be bought with ad spend: it is won with product data a machine can read and cite. That is catalog and datasheet work, not campaign work.
Travel and hospitality
1%of nights sold are booked through language modelsLess than 1%, across 325 million nights. Agents already discover, but they do not yet book: investing today in closing the agentic booking runs ahead of the fact; investing in being found and cited by one does not.
None of these four industries shares a regulator, a size or a sales cycle. All four share the same problem.
06
What you can ignore this month
Which model to use, because that decision stopped moving the result and changes every quarter. Buying an AI specific security platform, which is now a purchasing checkbox.
Full autonomy, pursued by only 22% of projects. And the figure that “95% of AI pilots fail”, which measures the pilot stage and not production results.
Last year's question was what to adopt. This year's question is what to redesign.
07
What to review this month
- Ask for the full cost of a process you automated this year, including the consumption paid outside your budget. If it takes more than a week to arrive, that is already the finding.
- Take one automated decision from last month and ask for the trace: who authorized it, on what data, who reviewed it. Time how long it takes to appear.
- Count how many people under 27 joined your area this year and compare it with three years ago.
08
How this report is put together
Every figure published has a primary source, a date and a sample size. What does not have one does not go in, however widely it circulates. If your industry is not in this issue, it is because this month we did not find a figure for it with a primary source, and we would rather leave the space empty.
The sources for this edition, one by one
Embedded AI in finance 22% to 42% (Consero, 18 May 2026, n=102) · significant impact on operating results 6% (McKinsey, 25 Aug 2026, n=1,719) · spend in business units, consumption charges and project cuts (Zylo, 2026) · mature governance 21% and named executive owner 14% (Deloitte, 2026, n over 3,000) · confidence in passing an independent AI governance audit within 90 days: 78% not confident, 73% running autonomous AI and 20% that have tested failure plans (Grant Thornton, Apr 2026) · employment gap ages 22 to 25 (Stanford, 12 Aug 2026, on ADP payroll data) · US banking regulators model risk guidance (17 Apr 2026) · insurers running generative AI 66% (EIOPA, 2 Feb 2026, n=347, 25 countries) · physicians using AI 81% (AMA, 2026) · AI referred traffic 138% and conversion 54% (Adobe Analytics, 17 Jun 2026) · nights booked through language models (Booking Holdings, 5 Aug 2026, across 325 million nights) · full autonomy 22% (S&P Global, 2026) · marketing budget 15.3% and 56% with insufficient budget (Gartner, 11 May 2026, n=401) · quarterly growth of sales agents 97% (Salesforce, 26 Aug 2026) · predictive maintenance 9% to 18% (Fluke and Censuswide, 7 May 2026, n over 600) · productivity 45% and decision quality 20% in finance (Gartner, 20 Jul 2026, n=204) · complex projects 88% against 14% and 97% that ran one (PMI, 12 May 2026) · agent security top ten and agent control standard (OWASP, 2026) · class action over automated screening served in March 2026, with age and race claims advancing in June 2026 (US court record).
Eddie F. Monge Morales · Founder
The previous ones
This is the current edition. The next ones get published here, one a month.
The current edition is read here, in full and without asking for anything. Past editions arrive when you subscribe.
Next edition: October 2026. One topic a month, with its mechanism and its window. No commercial follow up and no calls.
Subscribe to the Report
One edition a month. The previous ones arrive when you subscribe. No commercial follow up.
Get it every monthHow long does your organization take to answer the three questions above?