Axon Enterprise — Scaling the AI Era
Download case PDF (6 pages)Request teacher’s pack
Decision date and information cutoff: 3 October 2026
Decision scope: How to scale AI-assisted reporting and related AI Era products over the next 12 months
Case convention: The management assignment and alternatives below are constructed for classroom discussion. Axon has not disclosed this specific pending decision or deliberation. Company disclosures and published research are historical. No post-cutoff outcome is included.
The decision
Axon’s second-quarter 2026 results showed rapid growth across its connected devices and software businesses. The company reported that AI Era revenue had grown nearly 700% year over year. Yet the release did not provide an absolute AI Era revenue base, product-specific margins, customer usage, or renewal data. [S1]
At the same time, independent research on Axon’s Draft One report-writing tool had produced a more complicated productivity picture. A preregistered field experiment found no statistically significant reduction in total report-completion time. A companion survey found generally positive officer and supervisor perceptions, but also reported editing and workflow issues. The studies were conducted at one agency and did not evaluate downstream case outcomes. [S5]–[S6]
For this case, assume a cross-functional Axon strategy group is preparing a recommendation for the company’s next 12 months of product deployment, customer guidance and evidence collection. This group and assignment are fictional. The task is to recommend how Axon should scale Draft One and related AI Era products while balancing adoption, customer value, economics and trust.
Your assignment: Select a scale posture and explain what evidence would justify expanding, narrowing or slowing it. Address which workflows should be prioritized, what customers should be told about expected benefits, what safeguards should be monitored, and how Axon should measure value. Use only information available by 3 October 2026.
A product inside a broader platform
Axon began as a connected-device company and now describes a public-safety ecosystem that includes body-worn and in-car cameras, drones, digital evidence, records management, real-time operations and software tools. Management argues that linking devices, data and workflows makes the overall platform more useful. The public filings report company and segment results; they do not quantify the value of each connection or isolate AI product economics. [S1]–[S2]
Draft One was introduced in April 2024. Axon described it as software that turns body-worn-camera audio into a first-draft police-report narrative. Its launch materials said an officer must review and approve the report before submission. Axon also cited early agency trials and its own report-quality study when presenting the product’s potential. Those were company statements, not an independent multi-agency evaluation. [S3]
Axon later placed Draft One alongside other AI capabilities in its AI Era Plan. In an April 2025 product announcement, the company described the plan as including Auto-Transcribe, Draft One, Smart Detection and Policy Chat, and announced Axon Assistant for Axon Body 4. [S4] These tools address different tasks: searching or transcribing evidence, drafting a narrative, finding people in video, answering policy questions and providing voice-based assistance. Their bundle creates a platform strategy question: how much value can a growing set of AI features add to customer workflows when the company has limited public product-level usage and cost data?
Axon said Draft One generated more than 100,000 incident reports and saved officers 2.2 million minutes in 2024. [S7] Those figures are company-reported. The cited disclosure does not provide a research design, counterfactual, distribution of savings across users, or a conversion from minutes to staffing, response times or service outcomes.
The latest operating signal
Axon reported second-quarter revenue of $904.4 million, up 35.3% year over year. Software & Services revenue was $397.8 million, up 36.2%; Connected Devices revenue was $506.6 million, up 34.6%. The release attributed Software & Services growth primarily to new users and increased adoption of premium offerings, including the AI Era Plan. [S1] The filing shows the broader segment includes more than AI products, so its growth cannot be assigned to Draft One. [S2]
Exhibit 1. Selected Q2 2026 indicators
| Measure | Q2 2026 | Year-over-year comparison | What it does and does not show |
|---|---|---|---|
| Total revenue | $904.4 million | +35.3% | Company-wide growth |
| Software & Services revenue | $397.8 million | +36.2% | Includes multiple software and service offerings |
| AI Era revenue | Not disclosed in dollars | Nearly +700% | Growth from an undisclosed base; not a product margin or usage metric |
| Annual recurring revenue | $1.639 billion | +39% | Company-defined SaaS metric, not AI-specific |
| Net revenue retention | 126% | 124% a year earlier | Company-wide customer expansion/retention metric; not attributable to AI Era |
| Software & Services gross margin | 71.3% | 75.6% | Segment measure, not AI product economics |
| Adjusted Software & Services gross margin | 75.1% | 78.9% | Non-GAAP segment measure |
Source: Axon Q2 2026 shareholder letter and Form 10-Q. [S1]–[S2] Revenue and margins are reported; AI Era growth is company-reported. Margin declines were attributed primarily to higher professional-services mix and scaling new offerings. Axon stated that software-only gross margin continued to exceed 80%, but did not publish an AI Era or Draft One margin. [S1]
The same quarter’s release noted increased investment in people and AI among factors in higher operating expenses. Company-level operating cash flow, revenue guidance and contracted bookings describe Axon’s wider business and do not establish whether an individual AI feature creates attractive incremental returns. [S1]–[S2]
What the independent evidence says
Adams and coauthors studied Draft One in the patrol division of Manchester Police Department in New Hampshire. In a preregistered randomized trial, 85 officers were assigned to treatment or control groups, and researchers analyzed 755 reports. They measured report duration using records-management-system timestamps from opening a report to submission. In the preregistered mixed-effects model, the treatment estimate was a nonsignificant reduction of 29.66 minutes (standard error 39.62). Alternative models and a one-year analysis of 6,084 reports were also statistically nonsignificant. [S5]
The measured outcome was total report-completion duration, not simply the minutes spent drafting a narrative. The study did not evaluate objective report quality, supervisor review time, later investigative work, court outcomes, community trust or staff redeployment. The authors caution that their study was conducted at one agency and early in the product’s use. Its estimate is not an effect forecast for every department or report type. [S5]
A follow-up survey at the same agency linked responses to the field experiment. Its final analytical sample contained 76 people: 25 treated officers, 25 control officers and 26 supervisors. Treated and control officers did not differ significantly in their perceptions of usefulness, speed, quality, training or job ease. Among treated officers, 52% reported problems such as missing details, timeline mismatches or extensive editing. Supervisors perceived improvements in quality, completeness and submission speed, but those ratings were perceptions rather than blinded quality measurements. The authors noted single-agency, survey and downstream-outcome limitations. [S6]
Exhibit 2. Evidence and uncertainty
| Evidence | Observation | Boundary |
|---|---|---|
| Randomized field trial | No statistically significant reduction in end-to-end report duration; treatment estimate −29.66 minutes (SE 39.62) | One agency; 85 officers and 755 reports; no objective quality or downstream outcome test |
| One-year robustness analysis | No statistically significant duration effect in 6,084 reports | Same department and product context |
| Companion survey | Positive overall views; no significant treated/control differences; 52% of treated respondents reported tool issues | Same agency; self-reports and supervisor perceptions, not independent quality testing |
| Axon’s launch and usage claims | Company cited early time savings, report counts and a company-run quality study | Company-reported; methods and counterfactuals are not equivalent to the independent field trial |
Sources: [S3], [S5]–[S6], [S7]. The field trial and survey are related studies, not independent replications.
A competing product approach
Truleo introduced its Field Notes report-writing feature in July 2024. In its launch release, Truleo described a workflow that could use transcribed body-camera audio, work with different camera and records-management systems, require officer review and redact personally identifiable information. Truleo also claimed “up to 50%” report-time reduction and improved quality and accuracy. These are vendor statements; no independent validation or head-to-head test is provided here. [S8]
A Department of Justice COPS Office overview published in January 2025 also described Axon and Truleo as providers of AI-assisted police-report products at that time. It is a dated government overview, not a comprehensive market census. [S9] The public evidence in this packet does not establish comparative performance, pricing, market share, customer win rates or switching costs.
The distinction is strategically relevant, but unresolved: Axon can connect AI tools to its own devices, evidence and software; an alternative vendor may emphasize compatibility across existing equipment and records systems. Either posture could appeal to different agencies. The case provides no evidence that one architecture consistently wins.
Three possible scale postures
The fictional strategy group is considering three approaches:
- Accelerate platform-wide adoption. Make AI Era features a central part of customer expansion, continue adding functions, and promote productivity and workflow value broadly. This approach emphasizes speed, product breadth and the opportunity to deepen existing customer relationships.
- Scale with evidence gates. Keep products available while requiring stronger evidence before expanding use cases or making broad productivity claims. Define customer-level measures for usage, end-to-end time, objective quality, editing burden, supervision, cost and renewal.
- Limit deployment to selected workflows. Prioritize report types or agency settings where customers can show good fit and monitor results; defer broader recommendations until multi-agency evidence is stronger.
Each path has costs. Broad expansion may build customer familiarity and increase use of the platform, but it could amplify uncertain costs or trust problems before results are clear. Evidence gates may improve learning and credibility but require time, customer cooperation and operational measurement. Restricting deployment can reduce exposure to weak-fit use cases but may also limit adoption and delay product learning.
The public record does not disclose an internal decision deadline, product-level investment hurdle, unit economics or a formal rollout plan. Choose a posture as an analyst in a constructed exercise; do not assume one is already under consideration inside Axon.
Your recommendation
Prepare a two-page strategy memo for the fictional group. Recommend a scale posture for the next 12 months and explain:
- Which customer workflows or settings should be prioritized, and why?
- What should Axon say about efficiency, report quality and human review?
- Which outcomes should customers and Axon measure before broadening deployment?
- How should product usage, cloud/model costs, implementation effort and customer renewal be tracked?
- What evidence would change your recommendation?
Discussion questions
- What is Axon’s strategic advantage in AI-assisted reporting, if any? Which parts are supported by the public evidence, and which remain hypotheses?
- How should the company interpret rapid AI Era growth alongside the null time result in one agency?
- Do favorable perceptions or supervisor impressions create a commercial benefit even if total report time does not fall? What evidence would be needed?
- Should the company scale broadly, impose evidence gates or limit deployment to selected workflows?
- Which safeguards and measurements belong in the product, customer contract, agency policy or independent evaluation?
Sources and conventions
S1. Axon, Q2 2026 earnings release and shareholder letter, 5 August 2026. See Q2 summary results, segment commentary and forward-looking operating metrics.
S2. Axon, Form 10-Q for quarter ended 30 June 2026, filed 6 August 2026, accession 0001628280-26-053646. See Note 2, “Nature of Products and Services,” and Part I, Item 2, MD&A.
S3. Axon, Draft One launch announcement, 23 April 2024.
S4. Axon, AI Era Plan product announcement, April 2025.
S5. Ian T. Adams et al., “No man’s hand: artificial intelligence does not improve police report writing speed”, Journal of Experimental Criminology 22 (2026), 137–154; published online 2 October 2024. See methods pp. 140–145, Table 2 p. 147 and Appendix Table 3.
S6. Hunter M. Boehme et al., “Writing at the speed of hype: officers’ post-experimental perceptions of AI report writing”, Journal of Experimental Criminology 22 (2026), 649–661; published online 21 June 2025. See data/sample, Tables 1–3 and discussion/limitations.
S7. Axon, 2024 results release, 25 February 2025, “5 Big Innovative Leaps in 2024.” Its reported usage and time-saved figures are attributed company claims, not independent outcome estimates.
S8. Truleo, “TRULEO Enables AI-Powered Police Reports”, 16 July 2024, paragraphs describing product workflow, review/redaction and claimed benefits. News provided by Truleo; performance and safeguard statements are not independently validated here.
S9. U.S. DOJ COPS Office, “Using AI to Write Police Reports”, January 2025.
All numbers are company-reported or study-reported as labeled. Percentages are not used to infer undisclosed AI revenue or product returns. The case contains no fictional Axon operating targets, forecasts or internal deliberations.