Get in touch
Get in touch
Our evidence-based approach to avoiding wasted dashboards, models and AI.
Most dashboards, models and AI systems don't fail because the technology was wrong. They fail because the organisation around them was never going to act on what they produced. Decision mapping is Elliptica's method for catching that risk before it happens.
The checks run as a conversation, and each one rests on named, independent research rather than an opinion about best practice.
Before anything technical: who's actually going to use or present this day to day, and how do data conversations usually go for them?
Nearly two-thirds of employees feel anxious about data, and 30% avoid it entirely, a confidence deficit rather than a knowledge deficit. It's rarely named directly. It shows up as a room going quiet around a chart.
If this goes brilliantly, what decision changes as a result, and who makes it?
Most BI and analytics tools reach a median of just 14.3% employee adoption, commonly because they were built around whatever data was already available, not a decision anyone actually needed to make. The right answer is rarely the newest or most complex one either: a simple, transparent model has matched a proprietary black-box system's accuracy on real high-stakes data.
Who else needs to be comfortable with this before it can actually be acted on: a board, a minister, a funder?
Two-thirds of boards rate their own board packs weak or poor, and a 2025 review of 119 studies across 32 countries found the same pattern in government: institutional fragmentation and political resistance, not the quality of the evidence, are what usually stop it reaching a decision. A decision that works on paper still has to survive the room it's presented in.
Do the roles, incentives and workflows around this decision need to change alongside any new tool?
Only 30% of digital transformations succeed on technology alone, rising to 80% when the human system around it changes too. The same pattern holds for AI specifically: of 1,803 executives surveyed, 75% rank it a top-three priority but only 25% see significant value, a gap attributed to a 10-20-70 split, just 30% of value from algorithms and technology combined, 70% from people and process.
Who ends up using this day to day, and do they have both the subject knowledge and the technical comfort to run with it?
Domain knowledge and technical skill are usually staffed and trained as if they're separate jobs, and the mismatch runs both ways: one organisation's own promotion criteria turned out to reward domain knowledge but never once mention data ability.
How many tools are already floating around the team, and who is keeping track of them? Can the results they produce be repeated and verified?
95% of enterprise GenAI pilots never reach production. Traceability is a named standard, not a slogan: NIST's AI Risk Management Framework lists explainability and interpretability among the specific markers of trustworthy systems. The same test applies to a spreadsheet, a model or a dashboard: can someone else reproduce the result, and check it?
Start with the environment, not the technology. What the decision is, who has to act on it and what the organisation can carry are what determine which technical solution fits. Asked in conversation, these questions surface where a decision is at risk before the wrong deliverable gets built around it.
The Elliptica Visualisation Scorecard judges a dashboard or data tool on what it tells the people who use it, not only on whether its numbers are right. It joins two traditions: the rigour of Tufte and of Cleveland and McGill, and the usability of Nielsen and the WCAG 2.2 accessibility standard.
A first-time visitor can state the main message within five seconds, and each title states the finding rather than naming the axes.
Comparisons use bars, lines and dots, never angles or areas, and nothing on the screen is there for decoration.
Bar axes start at zero, an effect looks as large as it is in the data, and every view shows its source and date.
Anyone can operate it unaided, on a phone as well as a desk, and it meets WCAG 2.2 AA for contrast and screen readers.
Sources: DataCamp, The State of Data and AI Literacy (2026). BARC/Eckerson Group, Strategies for Driving Adoption and Usage with BI and Analytics (2022). Rudin, "Stop Explaining Black Box Machine Learning Models for High Stakes Decisions and Use Interpretable Models Instead," Nature Machine Intelligence (2019). Board Intelligence, Effective Board Reporting (2020). Suazo-Galdames, Saracostti and Chaple-Gil, "Scientific evidence and public policy," Frontiers in Communication (2025). BCG, Flipping the Odds of Digital Transformation Success (2020). BCG, From Potential to Profit: Closing the AI Impact Gap (2025). McKinsey, "Analytics Translator," Harvard Business Review (2018). Koloski et al., "Data Literacy in Industry," Harvard Data Science Review (2025). MIT, The GenAI Divide: State of AI in Business (2025). NIST, Artificial Intelligence Risk Management Framework, AI RMF 1.0 (2023). Tufte, The Visual Display of Quantitative Information (2001). Cleveland and McGill, "Graphical Perception," Journal of the American Statistical Association (1984). Nielsen, "Enhancing the Explanatory Power of Usability Heuristics," Proceedings of ACM CHI ’94 (1994). W3C, Web Content Accessibility Guidelines 2.2 (2023). Full publication details ›