Get in touch
Get in touch
Elliptica chooses the deliverable to fit the decision it serves, not the reverse: a brief, a model, a dashboard, a presentation.
Most engagements begin with the same first step, and go on to one or more of the four that follow. Each is scoped around the decision it has to support: which lever to fund, how a board prices a risk, what evidence a report needs to survive review.
A written brief that matches technology and strategy to the decision environment you actually work in. It is the cheapest way to find out whether a build is worth commissioning, and what it would have to do to be worth it.
It runs as a conversation, and it starts with the conditions the evidence has to work in rather than the technology: the people who will use it, the decision itself, the environment it lands in, the organisation around it, the team and how the tool would be governed. Those are what determine which technical solution fits, and each is set out in full under decision mapping.
Forecasts, economic modelling, cost–benefit analysis and return on investment, plus composite indices and statistical models where a number has to carry weight in public. For when the evidence has to hold up under review, not just produce a figure.
The right method is the simplest one that fits the decision, not the newest. Every model is documented so a reviewer can follow how a result was reached and reproduce it.
Dashboards and data tools built around the people who will use them, not handed to them, plus the plumbing underneath: harmonising sources that disagree, storage and integration with the systems you already run.
Maintenance is 60% of lifetime cost, so every build minimises it: it runs on what you already own as a live view rather than a one-off analysis, and ownership stays with you. Most analytics tools reach a median of 14.3% employee adoption because they are built around available data instead of a decision (BARC/Eckerson, 2022). That is the failure this work is designed to avoid.
One integrated, governed system scoped to a workflow, not a scattering of tools. Retrieval-augmented generation and validation-first summarisation over your own documents, with human review grounding every output rather than replacing judgment.
Every decision it makes is auditable, and it goes to production rather than staying a demo. That last part is where most of these end: 95% of organisations investing in enterprise AI see no measurable return, and pilots built with an external partner reach full deployment about twice as often as those built entirely in-house (MIT, 2025).
Presentations of findings to the audiences that have to act on them: boards, funders, staff, ministers, conferences and the public. Analysis that is right but unheard changes nothing, and the room is usually where that is decided.
This draws on a background teaching, and on years presenting flagship indices to press, policymakers and boards: knowing where an audience gets stuck, and how to make a difficult result land without flattening it.