BT Labs: applied AI research and field notes

Our thesis: the implementation gap

We track the delta between AI adoption and AI return. The foundational observation is simple and verifiable: the tools work, and the deployments often do not.

46 percent of U.S. small employer firms report that the business or its employees use AI. 7 percent of those users have fully integrated it into the business.

Source: Federal Reserve Banks, 2026 Report on Employer Firms: Findings from the 2025 Small Business Credit Survey, published 3 March 2026. 6,525 firms with 1 to 499 employees, all 50 states and DC; fielded 3 September to 14 November 2025.

56 percent of CEOs say AI has delivered no significant financial benefit to their business to date. 33 percent report gains in either cost or revenue.

Source: PwC, 29th Annual Global CEO Survey, published 19 January 2026. 4,454 CEOs across 95 countries and territories; fieldwork 30 September to 10 November 2025.

The 5 percent of companies BCG classifies as future built post 1.7 times the revenue growth and 3.6 times the three year total shareholder return of laggards.

Source: BCG press release, 30 September 2025, “AI Leaders Outpace Laggards with Double the Revenue Growth and 40% More Cost Savings”, reporting the study The Widening AI Value Gap: Build for the Future 2025. 1,250 senior executives and AI decision makers across 9 industries.

That last figure is an association between maturity and performance, not evidence that AI caused the returns. We print the caveat in the same paragraph as the number, which is the whole point of this page.

What we publish

BT Labs produces three kinds of work for operators.

• Field notes. Direct observations and workflow teardowns from our active implementations, including the ones where the number moved the wrong way.
• Frameworks. Step-by-step methodologies for auditing, deploying and measuring AI systems.
• Verified data. Curated statistics on AI productivity and adoption, filtered for small business relevance and read at the publisher’s own source before we print them.

Recent publications

Field notes: why AI fails small businesses, and how to fix it. The framework operators use to turn AI tools into systems that pay for themselves. Read the field note.

Field note: why engagements widen, and why most never get the chance. One anonymised engagement that started as paid media and ended up covering measurement, AI systems, agents and software, including the finding we withdrew and the number we cannot publish. Read the field note.

We published the 95% AI failure stat, then we checked it. The figure traces to MIT Project NANDA, July 2025, built on 52 interviews and 153 conference survey responses, and its PDF is no longer served at its own address. Here is how we checked it and what we cite instead. Read the correction.

Browse all field notes.

Our editorial standards

In an era of generated content, BT Labs holds to a short set of rules that are easy to state and expensive to follow.

01. Primary sources only. We read the number at the publishing organisation’s own page, PDF or data file before we print it. We do not cite vendor marketing studies as fact, and we do not cite an aggregator that cites a blog.

02. Retirement is part of the method. When a statistic we have published loses its primary source, or when a better-sampled successor lands, we withdraw it. We have retired figures from this site for exactly that reason rather than leave them standing because they read well.

03. Every number carries its methodology. Publisher, date, sample size and field window sit next to the figure, not in a footnote. If the sample size is not disclosed, we say that rather than omit it quietly.

04. Operator-first perspective. Our frameworks come from actual deployments in service businesses in the $500K to $5M revenue range, not from a literature review.

05. Human oversight. We build AI systems. Every piece of research published under the BT Labs name is reviewed, verified and contextualised by a person.