Nucleus is the layer your dashboards and your AI both read from.
Not another viewer on top of a spreadsheet. A governed metric layer in your warehouse, a dense dashboard suite built on it, and an analyst that queries the exact same definitions.
How a number gets from a source system to a screen.
Read left to right. Each stage does one job, and the shape holds whatever industry you are in.
The systems that actually run the business — ERP, CRM, POS, finance, operations. Each is the authority on its own data, and each is slow or costly to query directly.
One scheduled job, early each morning. It asks each system only for what changed since its own last run, never the whole history. A missed night heals itself the next night.
One place holding every fact the business reports on. Unlimited readers at once, one writer at a time. This is the single version of the truth.
Your dashboards read that one copy. Because the data is already local and shaped, filtering by region, product or date is a lookup, not a trip across the network.
Ask in plain English. Dr. D queries the same warehouse the dashboards use, so its answer and the report on screen can never disagree.
The systems that actually run the business — ERP, CRM, POS, finance, operations. Each is the authority on its own data, and each is slow or costly to query directly.
One scheduled job, early each morning. It asks each system only for what changed since its own last run, never the whole history. A missed night heals itself the next night.
One place holding every fact the business reports on. Unlimited readers at once, one writer at a time. This is the single version of the truth.
Your dashboards read that one copy. Because the data is already local and shaped, filtering by region, product or date is a lookup, not a trip across the network.
Ask in plain English. Dr. D queries the same warehouse the dashboards use, so its answer and the report on screen can never disagree.
The rule is absolute: no metric is ever computed in application code. If hold changes, it changes in one view, and every screen and every AI answer changes with it.
Ask it the way you would ask an analyst.
Dr. D translates the question into read-only SQL against your governed views, returns the figure, interprets it, and proposes the next move. It never invents a number and it never writes to your database.
Answers come from the same views the dashboards use. If the metric layer cannot answer it, Dr. D says so.
Write, DDL and system-schema statements are rejected before execution. Row caps and statement timeouts apply.
Every question, the SQL it produced, the row count and the latency are logged for review.
Samples, not a fixed product.
Everything below is a sample screen you can tune however you want — rename it, re-cut it, swap the metric, drop it entirely. Each one is a pattern for a question executives keep asking, modelled to your data, your language and your reporting calendar. Retail, distribution, manufacturing, healthcare, field services or financial services: the metric layer underneath is the same, only the entities change.
The margin identity reconciled end to end, period movers, and the multi-year weekly view.
Three headline numbers, a segment tree with heat tints, top and bottom performers.
Bubble map by territory, gaining and slipping tables, full regional rollup.
Rank history bump chart, quartile bands, unit economics per site or per asset.
Holt-Winters bands at network, region and site level with activity timelines.
Pareto concentration, lifecycle cohorts, revenue per unit of activity.
Modelled against realised outcome, variance in points, confidence by volume.
Stacked share over time, small multiples, mix-shift in percentage points.
Cohort retention triangle, segment economics, ranked churn risk.
Utilisation distribution, idle capacity, modelled redeployment suggestions.
Contribution against payout, accrual levels, tier-by-tier returns.
Taguchi arrays, main effects, signal-to-noise factor ranking.
Four decisions that do most of the work.
None of these are about a specific technology. They hold on any warehouse.
The old way asks the source system again on every filter change. A six-week pull can be millions of rows.
Landing it locally turns minutes per click into under a second.
Rebuilding everything nightly would take all night and get slower every month. Each table re-pulls a short trailing window from its own latest date.
The job stays roughly the same length as the data grows.
Dashboards and the assistant do not each keep their own definition of revenue or active customer. They read the same columns.
Two people quoting the same metric get the same figure.
It cannot roam. It calls a fixed set of functions — run this query, fetch this report's numbers — and inherits the signed-in person's permissions.
Someone who cannot open a dashboard cannot ask the AI about it either.
You are not buying software. You are buying an analyst.
Tools do not know what your business is trying to do. The second thing you get here is someone who sits with your operators, learns the process end to end, and then decides what is actually worth measuring.
PhD and PMP, eight years inside operating businesses. The model starts from how the work runs, not from what happens to be in the warehouse.
Most teams measure what their old report measured. We define the missing ones — the leading indicators that move the lagging ones.
A read of your current ETL, models and reports: what reconciles, what silently disagrees, what nobody trusts and why.
Elasticity, mix effects, contribution and variance decomposition — so a movement in the headline can always be attributed to its parts.
One definition per metric, written in SQL, reviewed with the people who own the number. No more two versions of revenue.
Every build ends with a narrative leadership can act on: what changed, why, what it is worth, and what to do next.
What comes in the box.
Every formula — hold, RTP, rotation, net per store per day — is defined once, in SQL, in the warehouse. Dashboards read it. The AI reads it. There is exactly one version of every number.
Ask in plain English. Dr. D writes the query against the same governed views, returns the number with the interpretation, and tells you what it would do next.
Dense, fast, boardroom-grade screens: executive summary, geography, leaderboards, game economics, player retention, promotion design. No scroll hunting.
Holt-Winters forecasts with confidence bands on every location, and a severity-ranked list of what fell outside its band this week — before anyone has to notice.
Role-based access, row-level security, audit logging on every AI question. Read-only query paths. Nothing leaves your governed data estate.
One build, one fee. Modifications on request at a minimal charge. You are buying an asset you own, not a seat count that grows with your headcount.
How a build runs.
Seven steps from the first call to handover. Nothing is owed until step five.
You have seen the product and want to talk. We set up a short call, no preparation required.
Thirty to forty-five minutes on how the business actually runs, where reporting hurts, and what a good answer would have been worth.
A second call with your data people: sources, pipelines, reporting cadence, visualisation tools and the KPIs you track today.
What is broken, how we would fix it, what your portal would look like, deliverables, timeline and price. No commitment to this point — you can walk away with the report.
We settle payment terms, you release the first instalment, and the build begins.
Milestones land as the contract sets out, each one reviewed with you so course corrections happen early, not at the end.
Final delivery, a walkthrough for your team, documentation and a working session on the questions they will actually ask.