This is what a Mirmara report looks like, from the executive brief to the 18-month roadmap. Halvard Group is fictional and all data is illustrative, but each chapter follows the structure of a real report. Hover, click and test each exhibit.
Every organisation runs AI on human infrastructure. Few can see it. This report makes it visible and open to discussion.
A Mirmara diagnostic answers one question: where do our people really stand on AI, and what should be done first? The chapters below follow the structure of the report.
Sample report for a fictional company. Figures are illustrative; the structure, instruments and outputs are real.
In January, Halvard provided AI assistants to 400 of its 540 employees. Licence usage has since plateaued, and a company-wide training programme is under consideration. Before committing that budget, Halvard commissioned this diagnostic to establish what is limiting adoption.
Score bands: 0–39 Nascent · 40–59 Emerging · 60–79 Established · 80+ Embedded
Licence logs show 34% of seats active, yet 61% of respondents report using AI weekly, and 23% use tools Halvard has not approved. Actual use is higher than the organisation can currently observe.
– of AI-advice ties point to three colleagues, and one person is the only link between Engineering and the rest of the organisation. The leadership team receives –.
Fear of judgement (31%), unclear purpose (24%) and fairness concerns (19%) are each cited more often. The planned training targets the fourth-ranked barrier.
Training addresses a knowledge gap. Where the barrier is fear of judgement, unclear purpose or perceived unfairness, training alone is unlikely to change usage. Mirmara therefore diagnoses before budget is allocated.
Human infrastructure is the relationships, emotions, beliefs, behaviours and values through which a workforce adopts AI and integrates it sustainably into everyday work. Mirmara scores it on five layers, using a 30-item questionnaire, combines them into the Human Infrastructure Index, and locates each layer in the organisation’s system. Select a layer on the chart or the tabs to see what sits behind its score.
Mirmara reads culture as a system, following Daimler’s reculturing model (2022). An organisation stays coherent when Purpose, Strategy and Culture are connected, and Culture itself is made up of Values, Process and Practice. AI is an external load on the system that can pull two of these elements apart, and each layer reads one such seam. The score shows how large the gap is; the location shows where to act. Select a seam, or hover a corner for its definition.
AI adoption spreads through relationships. Policy alone rarely changes behaviour; people tend to start when someone they trust shows them how. Social Network Analysis maps these informal paths: people are the dots, and “whom do you ask about AI?” are the lines.
Five executives, six managers and 56 team members. Following this chart, an AI rollout would start at the top: announce through leadership, cascade through managers.
Each line connects a person to a colleague they turn to for help with AI. Larger, brighter dots are asked more often. Note how the leadership team recedes.
Start AI adoption from two different points and observe how it spreads through this network over 12 weeks.
Illustrative diffusion model on this sample network: each week, a person adopts with a probability that rises with the number of colleagues they ask who already use AI. It shows a mechanism, not a forecast. Averages are over 300 runs each.
Involve informal influencers in workshops and focus groups. Their endorsement is likely to carry more weight than an announcement.
One person linking two groups is a single point of failure. The map shows where a structural hole needs a second connection.
The colleagues everyone asks rarely have this in their job description. Identifying them early can reduce the risk of burnout and attrition.
The map is a stylised view of 67 nodes. In a real report it is built from consenting participants, presented in aggregate and anonymised; individuals are never identified to the client.
A gap is a point where what the organisation states and what people experience have diverged. Each gap is plotted by its severity in the data and by its leverage, meaning how much of the AI investment depends on closing it. Bubble size shows the share of the workforce affected.
Employees understand what AI can and cannot do, so general training is likely to add the least value here.
It shows what stronger human infrastructure looks like inside Halvard. Its norms are worth studying and adapting, not copying, since Operations works under different conditions.
Formal and informal influence coincide. This is the clearest internal example of leader-led adoption.
Every action names the gap it closes and an owner. Sequence matters: later horizons are unlikely to succeed before earlier ones are in place. Filter by gap to see how each is addressed.
The methods complement each other: the questionnaire provides breadth, interviews depth, and Social Network Analysis the relational layer. Together they ask not only what people do with AI, but why.
30 items across the five layers. Anonymous, segmented by function and level.
Five to seven conversations across roles. Interviews explain the reasons behind the survey patterns.
A short sociometric survey with explicit opt-in consent. Reported only in aggregate.
Kick-off, communication plan, works-council alignment.
Questionnaire and network survey live. Interviews held.
Scoring, segmentation, network analysis, triangulation.
Written report and leadership readout with Q&A.
Optional re-measurement of the weakest layers.
Research foundations. Edmondson (1999) on psychological safety · Davis (1989) on technology acceptance · Cross, Borgatti & Parker (2002) on network analysis for collaboration · Deci & Ryan (2000) on self-determination · Schepman & Rodway (2020) on attitudes towards AI · Bandura (1977) on social learning · Daimler (2022) on reculturing.
Whether AI delivers value depends heavily on people and process, which are also the areas where leaders have the least visibility. Independent research points in the same direction.
BCG’s 10-20-70 framework attributes about 70% of AI value to people and process, 20% to technology and data, and 10% to algorithms.
Leaders estimated that about 4% of employees use generative AI for at least 30% of their work. Employees reported about 13%.
of organisations report no measurable P&L return on generative AI despite $30–40bn of enterprise spending, based on interviews, a survey and 300 public deployments.
of 1,150 surveyed US workers received AI-generated “workslop” in the past month; each instance took about 1h56m to resolve.
A 30-minute conversation. Together we talk through a possible diagnostic approach and what you are seeing in your team or organisation.
Talk to MirmaraHalvard Group is a fictional company. All figures in this report are illustrative.