Most management tools answer a simple question: where does the company stand today? That is useful, but incomplete. An executive, an advisory firm, an investor or a business development team also wants to know who to compare with, what has changed, which relationships explain the gap and which questions deserve to be examined next. Northstar was born from this requirement: turning company accounts and signals into a comparative, temporal and explainable reading.
01 — An executive’s question, not a table of ratios
Northstar’s golden use case fits in a few sentences: “Is my company doing well? Compared with whom? Why? What do comparable companies that are doing better do differently?” The wording is deliberately accessible. The product must not ask its user to become a financial analyst to understand a gap.
This question changes how the system is built. An isolated ratio becomes an incomplete object. It needs a definition, a value, a period, a benchmark, an evolution, a decomposition, a source and a limit. Northstar therefore does not try to multiply indicators, but to connect the ones that actually change how a situation reads.
02 — Building Company 360 before comparing
A comparison is only reliable if you know who you are comparing. Northstar relies on an entity model that connects a company, its identifiers, its financial years, its activity, its establishments and the public events that can put its trajectory in context. The SIREN company number serves as the identity pivot; period and provenance stay attached to every observation.
The separation between public and private data is structural. Public accounts can form a shared base. An advisory firm, a fund or a company can then bring its own histories, exports or documents. This data stays visible within its own scope, but it can be interpreted with the same analytical reference frame. The system gains depth without turning confidentiality into a secondary variable.
which entity?
which year?
which source?
03 — A benchmark is first a decision about scope
Northstar does not compare a company with an abstract average. It builds an explicit cohort: comparable financial year, similar activity and dimensions that are actually available. Normalised accounts feed a set of financial metrics; V1 similarity relies on a deterministic, weighted distance.
This choice matters for two reasons. First, it makes the score inspectable: users can see which dimensions were matched and which are missing. Second, it prevents the system from giving artificial precision to a cohort that is too thin. An honest benchmark must be able to say it does not have enough material, rather than disguising an absence as certainty.
04 — The benchmark says where. The trajectory asks when.
A snapshot can position a company. It does not yet tell its path. The time dimension makes it possible to distinguish a temporarily high margin from a lasting improvement, growth that generates cash from growth that consumes it, or a break from mere measurement noise.
Northstar organises this reading around annual sequences: starting position, evolution, breakpoints, groups of trajectories and the characteristics that differentiate the observed paths. Caution is essential: a historical pattern can flag an area to investigate; it does not automatically become a causal recommendation.
The case presents the analytical logic of the trajectory and the controls around it: every reading stays tied to its scope, its period and its limits.
05 — Explaining a financial signal
A Financial Explanation engine does not just write a sentence around a figure. It unfolds a chain: indicator, definition, value, benchmark, evolution, decomposition, interpretation, possible causes, company-specific evidence, peer evidence, lines to examine, confidence and limits.
A working capital example shows the difference. Saying that working capital to revenue is high describes a gap. Connecting receivables, inventory, suppliers, cash and trajectory makes it possible to ask a more useful question: is growth consuming cash, and which component deserves investigation? The system does not decide in the executive’s place. It makes the next conversation more precise.
06 — The role of the LLM: tell, never fabricate
The Northstar chain separates financial facts, deterministic calculations, the benchmark, trajectories, explanation rules, the evidence package and the narrative. An LLM can help turn this material into understandable text. It must not invent a figure, rewrite a missing period or turn a correlation into advice.
This architecture makes the output reviewable. An important sentence points back to the metric and its source. A hypothesis carries its label. Missing data stays missing. Explainability is not a varnish applied after the model: it is a property of the pipeline.
07 — Five entry points, one engine
The core of Northstar stays the same: Company 360, Financial Understanding, Comparable Universe, Benchmark, Trajectory, Explanation, Evidence. Uses then change according to the decision.
- 01SME management — understand the situation and choose the next investigations.
- 02Portfolio advisory — repeat the reading across hundreds of companies and spot atypical trajectories.
- 03Investment screening — rank cohorts and look for profiles that deserve a closer look.
- 04Due diligence / M&A — put a target in context with real comparables and its history.
- 05B2B sourcing — find companies that resemble the best customers or partners.
08 — A verified core, continuous controls
The current core includes a normalised financial contract, cohort benchmarks, deterministic similarity, API endpoints and a structured public enrichment layer. These elements form a credible foundation for the instrument.
Every capability is exposed with the same standard: the reasons behind a score must be readable, sector references consolidated and historical series checked before any strong claim. At Northstar, the rules are explicit and the limits are part of the product.
Northstar does not promise a magic answer.
It builds a better question, with its evidence.