Adoption Without Capability Reporting

How 22 Major Firms Publicly Describe Their AI Programmes · Richard Foster-Fletcher · MKAI · Revised working paper, June 2026

Abstract

This study examines how 22 publicly listed organisations described their artificial intelligence programmes in prepared management remarks and, where necessary, supplementary public reporting from the 2025 and 2026 reporting period.

The coded inventory applies a three-part taxonomy. Adoption and Containment records statements about usage, deployment scale, cost, efficiency, investment, and governance coverage. Capability Reporting records measured business outcomes directly attributed in the reporting to AI reasoning or synthesis. Product and Marketing Claim records feature descriptions, comparative assertions, positioning language, and capability assertions that cannot be verified from outside the firm.

Across the 22 organisations recorded in the reconciled table, the inventory contains 404 Adoption and Containment statements, 2 Capability Reporting statements, and 308 Product and Marketing Claims. The ratio is 202 to 1 to 154. Both Capability Reporting statements come from Alphabet. No other organisation in the sample produced a statement meeting the stated threshold.

The study is a disclosure study. It reports what appeared in prepared public communication. The performance of the AI programmes covered, the quality of internal measurement, and the business value generated by those programmes fall outside its scope. The finding is that, in this sample, public reporting gives considerably more visibility into AI adoption, usage, scale, efficiency, investment, governance, and product positioning than into measured AI contribution to business work.

1. Introduction

Public reporting about artificial intelligence has become a regular feature of enterprise communication. Large firms describe AI in earnings calls, annual reports, investor presentations, and supplementary public documents. These sources often contain material about deployment, customer adoption, usage, cost, efficiency, investment, and governance coverage.

This paper examines a narrower question. It asks whether prepared public reporting contains measured statements about what AI contributed to business work, separated from surrounding adoption, automation, efficiency, product, or commercial claims.

The study reports a coded analysis of explicit AI-related statements from 22 publicly listed organisations during the 2025 and 2026 reporting period. Prepared management remarks were preferred because the study records what organisations chose to put into public reporting. Analyst Q&A was excluded because it reflects what management was asked, rather than what the organisation selected for prepared disclosure.

The central finding is direct. Across the sample, the inventory records 404 Adoption and Containment statements, 2 Capability Reporting statements, and 308 Product and Marketing Claims. Both Capability Reporting statements come from one Alphabet earnings call. No other organisation produced a statement meeting the stated threshold.

The paper makes no inference about enterprise AI value, internal evidence, concealment, or implementation quality. The evidence reaches a narrower finding: the public reporting examined here rarely discloses measured AI contribution separately from adoption, efficiency, governance, commercial scale, or product positioning.

The contribution of the paper is therefore empirical and documentary. It provides a coded inventory of AI-related statements from a cross-sector sample, applies one consistent taxonomy, and shows the distribution of public claims across the three coded groups. The interpretation is intentionally limited to what the reporting contains.

2. Method

The study applies a three-part coding taxonomy to explicit AI-related statements in prepared management remarks delivered during the 2025 and 2026 reporting period. Where the chosen earnings call contained little or no AI-related material, supplementary public reporting from the same period was used. Supplementary sources included investor day transcripts, annual reports, Form 10-K filings, and sustainability reporting where AI was discussed.

Source selection followed a consistent procedure. First, the most recent full earnings call transcript available at the time of coding was identified. Prepared management remarks were extracted. Where those remarks contained fewer than three AI-related statements, one supplementary source from the same fiscal period was added where available. Analyst Q&A was excluded for every organisation in the sample.

Each coded statement was assigned to one of three groups. The groups are defined below.

2.1 Coding taxonomy

Adoption and Containment records statements about usage, seats deployed, customer counts, revenue from AI products, cost reduction, efficiency gains, cycle time improvements, governance audits, infrastructure investment, deployment scale, and similar measures. Statements reporting labour displacement, throughput growth, manual step reduction, or service-speed improvement were coded here because they do not isolate AI reasoning or synthesis from the surrounding workflow.

Capability Reporting records measured business outcomes directly attributed in the reporting to AI reasoning or synthesis. A qualifying statement had to include both a claimed AI contribution and a measured outcome that separated that contribution from ordinary deployment, automation, usage, or commercial scale. Examples that would qualify include AI identifying a risk that existing methods missed, AI generating a forecast that outperformed a defined baseline, or AI producing an option that changed a measured decision outcome.

Product and Marketing Claim records feature descriptions, comparative assertions, positioning statements, forward-looking claims, and capability assertions that cannot be verified from outside the firm. Examples include statements about unmatched accuracy, industry-leading reasoning, better decisions, or new product capabilities where no measured outcome is disclosed.

2.2 Coding process

The coding taxonomy was implemented through an AI-assisted pipeline. Source material was supplied to a language model with the coding brief, the definitions above, and worked examples. The model returned a coded inventory containing verbatim quotes, speaker identification, source, date, proposed classification, and classification notes.

The author reviewed the inventory, with particular attention to statements coded as Capability Reporting, flagged ambiguities, and statements near the boundary between efficiency and capability. Where review identified classification errors or insufficient evidence, the author adjusted the coding.

A ten per cent spot-check was conducted by running the same source material against the coding brief in a second session and comparing the output. The spot-check produced minor adjustments in five cases. The adjustments moved statements between Adoption and Containment and Product and Marketing Claim. No adjustment moved a statement into or out of Capability Reporting.

The coded inventory records the three groups as A, B, and C for operational consistency. This paper uses the named labels throughout.

3. Sample

The reconciled sample used in this version includes 22 publicly listed organisations across four broad groupings. Six are AI vendors or cloud platforms: Microsoft, Salesforce, Adobe, Oracle, Alphabet, and Meta. One is enterprise software and consulting: Accenture. Four are financial services firms: JPMorgan Chase, Goldman Sachs, Morgan Stanley, and HSBC. Eight are retail, consumer goods, and industrial firms: Shopify, Walmart, Target, Unilever, Procter & Gamble, Mondelez, General Motors, and FedEx. Three are media and consumer services firms: Disney, Netflix, and Spotify.

The selection criteria were pragmatic. Each organisation was publicly listed at the time of the relevant reporting cycle. Each had discussed AI in public reporting during the 2025 or 2026 period. Each operated in a sector where AI had been identified by the firm itself as commercially relevant. The sample was constructed to provide cross-sector coverage rather than statistical representativeness.

The sample is weighted towards North American firms because public reporting in that jurisdiction is dense and chronologically aligned with the reporting window. European and Asia-Pacific organisations are represented, although they are under-represented relative to global enterprise AI activity.

Three further organisations were coded during earlier research sessions and excluded from the reported sample. Their exclusion reflects coding and comparability concerns rather than a substantive finding about any of them.

The time window covered by the sources is the 2025 and 2026 reporting period. For most organisations, the source is a single earnings call from the fourth calendar quarter of 2025 or the first calendar quarter of 2026. Where that call contained little AI-related material, one supplementary public source from the same period was used.

4. Findings

The aggregate counts are 404 Adoption and Containment statements, 2 Capability Reporting statements, and 308 Product and Marketing Claims. The ratio across the three groups is 202 to 1 to 154.

The distribution shows a strong concentration of public reporting in Adoption and Containment and Product and Marketing Claim. Capability Reporting is almost absent under the stated threshold.

4.1 Per-organisation counts

Table 1 records the coded statement counts per organisation, where A means Adoption and Containment, B means Capability Reporting, and C means Product and Marketing Claim.

4.2 Capability Reporting statements

Two statements meeting the Capability Reporting threshold were identified. Both come from Alphabet’s Q4 FY2025 earnings call, delivered on 4 February 2026, and both were spoken by Philipp Schindler, Chief Business Officer.

The first statement reads: "For instance, Aritzia, Canada’s premier fashion house, used AI Max to find new high-value customers that traditional strategies miss, delivering an 80% incremental uplift in conversion value for Q4." This is the clearest Capability Reporting statement in the sample. It reports an AI-enabled identification task against a stated comparison with traditional strategies, and it reports a measured commercial outcome.

The second statement reads: "AI Max enabled the L’Oréal Group to maximize its presence across the full consumer journey, fuel its consumer growth, and increase revenue for DTC brands like NYX by 23%." This statement was coded as Capability Reporting with an ambiguity flag. The revenue movement is measured, and the statement attributes that movement to AI Max, although it does not describe the underlying reasoning mechanism.

Several statements approached the threshold and were reviewed closely. A Salesforce customer described closing $2.7 million with Agentforce. That statement reports revenue connected to an agent-enabled workflow. The reporting does not separate the agent contribution from the surrounding sales workflow, user actions, or process design. It was therefore coded as Adoption and Containment.

4.3 Sectoral observations

AI vendors and cloud platforms account for a large share of the total statement volume. Their reporting frequently covers seats deployed, revenue run rates, token volumes, infrastructure investment, customer counts, partner integrations, and product capabilities.

Financial services firms report fewer AI-related statements overall. JPMorgan Chase is more detailed than the other financial services firms in the sample because the sources include numeric efficiency material from its 2025 Annual Report and 2025 Investor Day alongside its Q1 2026 remarks. Goldman Sachs, Morgan Stanley, and HSBC report AI primarily through investment, efficiency, adoption, or market-context language.

Retail, consumer goods, and industrial firms generally report automation coverage, supply chain integration, content generation, product discovery, productivity, and deployment across stores, distribution centres, or production systems. Media and consumer services firms report creator tools, recommendation systems, content workflows, and engagement-related claims.

The observed pattern is broadly similar across the sample. Where organisations speak about AI in prepared public reporting, they most often speak about deployment, usage, efficiency, commercial scale, governance, or product positioning. Measured Capability Reporting remains rare under the stated threshold.

4.4 Two boundary examples

One Microsoft statement illustrates the difference between governance volume and capability measurement. On Microsoft’s FY26 Q2 earnings call, Satya Nadella said: "24 billion Copilot interactions were audited by Purview this quarter, up 9X year-over-year." The statement reports audit coverage at very large scale. It does not report what the AI produced in those interactions, whether the outputs improved work quality, or how output quality compared with a baseline.

One Accenture statement illustrates the boundary between capability language and measured capability. On Accenture’s Q2 FY2026 earnings call, Julie Sweet said that a set of AI-enabled platform capabilities could "analyze large volumes of data, initiate routine actions, and support better decisions in real time." The statement asserts better decisions. It provides no disclosed measure for better decision quality. It was therefore coded as Product and Marketing Claim.

5. Discussion

The prepared public reporting examined in this study contains many statements about AI adoption, usage, deployment, efficiency, cost, investment, governance coverage, and product positioning. It contains very few statements about measured AI contribution to business work.

This finding concerns public disclosure only. Under the threshold used in this study, the reporting examined here rarely discloses measured capability.

Several explanations could fit the pattern. Firms may have capability evidence internally and choose not to disclose it. Capability may be difficult to separate from surrounding workflow changes. Firms may prefer to disclose commercial, operational, and governance quantities because those are easier to report publicly. The study does not adjudicate among these possibilities.

The coding separates two kinds of claim. An efficiency claim, such as reduced triage time or lower labour cost, can show that a workflow changed. It does not by itself show that AI reasoning or synthesis produced a measured business outcome. A product claim, such as better decisions or superior accuracy, may describe a claimed capability. It remains outside Capability Reporting unless the reporting provides a measure that can be read from outside the firm.

The two Alphabet statements show the threshold in practice. They make a claim about an AI-enabled product, connect that product to a measured commercial outcome, and provide a comparison or attribution that the reporting itself presents. The second remains borderline, which is why it is flagged. The rest of the sample contains many capability-adjacent claims without that level of measurement.

The study therefore supports a narrow claim about public evidence. Prepared reporting in this sample makes AI visible as adoption, scale, cost, efficiency, governance, revenue, infrastructure, and product narrative. It rarely makes AI visible as measured contribution to business work.

6. Limitations

First, the study uses public reporting. It measures what firms said in earnings calls, investor day transcripts, annual reports, and filings. Internal evidence, private regulatory disclosure, and operational experience are outside the evidence base.

Second, the prepared-remarks rule excludes material. Some earnings calls included more AI discussion in analyst Q&A than in prepared remarks. That material was excluded consistently because the study concerns selected public reporting rather than reactive questioning.

Third, the time window is short. Most organisations are represented by one earnings call, sometimes supplemented by one further source from the same reporting period. A longer longitudinal study could produce a different distribution.

Fourth, the Capability Reporting threshold is deliberately strict. A looser threshold would record more Capability Reporting statements. The strict threshold was used to avoid counting adoption, automation, efficiency, or product description as evidence of measured AI contribution.

Fifth, the sample is purposive rather than statistically representative. It is a cross-sector sample of publicly listed organisations with visible AI commentary. The findings should be read as evidence from this sample rather than as a population estimate for all public firms.

Sixth, the coding process used AI assistance. The author reviewed the coded inventory, adjudicated ambiguous cases, and conducted a spot-check. A full independent human recoding exercise would strengthen reliability.

7. Closing observations

The coded inventory reports how 22 publicly listed organisations described their AI programmes in prepared public communication during the 2025 and 2026 reporting period. Across those organisations, the reconciled table records 404 Adoption and Containment statements, 2 Capability Reporting statements, and 308 Product and Marketing Claims.

The study establishes what this reporting contains and what it leaves unmeasured for the external reader. Whether the AI programmes covered by the reporting generate capability in the work of the firm remains outside the evidence base.

The public record contains detailed language about use, scale, deployment, cost, infrastructure, governance, revenue, and product features. It contains little measured reporting about what AI changed in business judgement, synthesis, forecasting, risk identification, option generation, or decision quality.

For readers of public reporting, that leaves an unresolved measurement gap. AI appears frequently in the public record; its measured contribution to business work is recorded far less often.

8. Research process and AI use

The collection and preliminary coding of the statements analysed in this paper used AI assistance. Source transcripts and reports were supplied to a large language model with the coding brief described in Section 2, including the taxonomy, definitions, Capability Reporting threshold, and worked examples.

The author reviewed the full inventory, with particular attention to statements coded as Capability Reporting and to all flagged ambiguities. A second session, using the same source material and coding brief, was run against a randomly selected ten per cent of statements for spot-check purposes. Differences were reviewed by the author.

The coded inventory is available as supplementary material. It includes verbatim quotes, speaker identification, source, date, classification notes, ambiguity flags, the coding brief supplied to the language model, adjustments made during author review, and the spot-check record.

9. Disclosures

Conflict of interest: The author is Chair of MKAI and founder of Reality & Reason. This research was conducted independently, with no third-party funding.

Use of AI: This research used large language models for source extraction, preliminary coding, and review support. The author reviewed the coded inventory, adjudicated ambiguous cases, and takes full responsibility for the research design, coding decisions, findings, and final text.

References

Accenture plc. (2026, 19 March). Q2 FY2026 earnings call transcript.

Adobe Inc. (2026, 12 March). Q1 FY2026 earnings call transcript.

Alphabet Inc. (2026, 4 February). Q4 FY2025 earnings call transcript.

Federal Express Corporation. (2026, 19 March). Q3 FY2026 earnings call transcript.

General Motors Company. (2026, 27 January). Q4 2025 earnings call transcript.

Goldman Sachs Group, Inc. (2026, 13 April). Q1 2026 earnings call transcript.

HSBC Holdings plc. (2026, 25 February). Q4 FY2025 fixed income earnings call transcript.

JPMorganChase & Co. (2025, 19 May). 2025 Investor Day transcript.

JPMorganChase & Co. (2026, 23 February). 2026 Company Update transcript.

JPMorganChase & Co. (2026, 6 April). 2025 Annual Report.

JPMorganChase & Co. (2026, 14 April). Q1 2026 earnings call transcript.

Meta Platforms, Inc. (2026, 28 January). Q4 FY2025 earnings call transcript.

Microsoft Corporation. (2026, 28 January). FY26 Q2 earnings call transcript.

Mondelez International, Inc. (2026, 4 February). 2025 Form 10-K.

Mondelez International, Inc. (2026, 17 February). CAGNY 2026 presentation.

Mondelez International, Inc. (2026, 15 April). 2025 Snacking Made Right Report.

Morgan Stanley. (2026, 15 April). Q1 2026 earnings call transcript.

Netflix, Inc. (2026, 16 April). Q1 FY2026 earnings call transcript.

Oracle Corporation. (2026, 10 March). Q3 FY2026 earnings call transcript.

Procter & Gamble Company. (2026, 24 April). Q3 2026 earnings call transcript.

Salesforce, Inc. (2026, 25 February). Q4 FY2026 earnings call transcript.

Shopify Inc. (2026, 11 February). Q4 2025 earnings call transcript.

Spotify Technology S.A. (2026, 10 February). Q4 FY2025 earnings call transcript.

Target Corporation. (2026, 3 March). 2026 Financial Community Meeting, including Q4 and Full-Year 2025 earnings.

Unilever plc. (2026, 12 February). H2 FY2025 earnings call transcript.

Walmart Inc. (2026, 19 February). Q4 2026 earnings call transcript.

Walt Disney Company. (2025, 13 November). Fiscal Year 2025 Annual Financial Report / Form 10-K.

Walt Disney Company. (2026, 2 February). Q1 FY2026 earnings call transcript.