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.