Ask
Ask anything about the company, and see how it was worked out
Ask reads the question, finds what it is about, and works it through everything the twin holds: the ledger, the graph, every line of every file, the industry profile and the runs on record. The answer comes back in a moment, with every figure marked, every step shown, how sure it is and the file that would make it surer.
Build a twinHow it shows its working
No language model is needed for any of it. With one configured, the model phrases the worked answer and is given nothing else to work from.
An answer you can audit is worth more than one you have to trust.
Ask how concentrated revenue is and the answer names the largest customer and its share, counts the top five, measures the whole book as if it were a number of customers of equal size, sets that against the band for your industry, checks when the largest contracts end, and quotes the clause in the contract file that matters. Each of those carries a number you can click, and the number opens the row, the line and the file.
The mechanism
What happens to a question
- 01
It is read
Against thirty three concepts a company is organised by (revenue, margin, runway, concentration, churn, renewals, pricing, suppliers, people and the rest) and against every name on your graph, so "Meridian" means your customer and not a word. A question that is a decision in disguise is recognised as one.
- 02
The numbers are looked up
The ledger rows the concept stands on, each with its value, its band, where it came from and the line in the file it was read from.
- 03
The arithmetic is done and written down
A concentration index, a runway, a payback period, a customer lifetime, a market size implied by your share. Every calculation is its own row, with the formula beside it.
- 04
It is set against your industry
The band this model holds for companies like yours, with what sitting above or below it usually means, labelled as the model's assumption rather than a survey.
- 05
Every file is searched, line by line
Prose is split into sentences and a spreadsheet row is read with its column names. A paragraph that opens with a customer's name keeps that name for every sentence in it, so feedback is found under the customer who gave it.
- 06
The record is checked for trouble
Contracts of the largest customers ending soon, single sourced suppliers, runway on the simulation, a stretched team, figures well outside the industry band, open alerts and the last stress test.
- 07
It concludes, and says how sure it is
Confidence is computed from where the numbers came from, each weighted by how far it can move a decision. The answer names the upload that would firm it up, where that upload goes, and the questions worth asking next.
What it answers
Questions it works through, and what it works out
| Ask about | For example | What it works out |
|---|---|---|
| Money | Why is our margin where it is | Gross profit, the bought in share and the largest supplier behind it, price against competitors, pass through, the industry band and the rule of thumb that applies |
| Cash | How long will our cash last | The monthly flow on the ledger, runway if it is negative, and the year ahead on the simulation with the lowest point and the share of runs below zero |
| Customers | How concentrated is our revenue | The largest share, the top five, the book as customers of equal size, renewals of the largest within a year and the clauses that bind them |
| Risk | Which customers are at risk | Every named customer scored on contract end, rating and status, blended with the trained network where it earned its place, ranked by likelihood and exposure |
| Price and market | How do our prices compare with competitors | Your level against each competitor, what a move of the size you name does to volume on the first round, how much revenue is held by contract |
| Supply and people | Who are our most important suppliers | Spend by supplier, single sources, lead time weighted by spend, cost per head, revenue per head against capacity |
| The year ahead | What will the next twelve months look like | Revenue, profit, cash and customers at month twelve, the band around each, and how companies like yours usually get into trouble |
| The record itself | How confident is the model | How much of the model is your record, weighted by leverage, and which defaults to replace first |
| Anything by name | What does the contract with Meridian say | The customer, supplier, competitor or department, its figures, its connections on the graph and the lines in your files that name it |
Ask reads your twin and nothing else: no other accounts, no web, and no general knowledge it is willing to offer as fact beyond the industry profile, which it labels every time.
Decisions
A decision is answered from the record, then handed to Analysis
Ask "what happens if we raise price 10 percent" and Ask answers what the record says about price: your level against competitors, how sensitive demand is, what a move of that size does on the first round, and how much revenue is held by contract until renewal. Then it offers to run it.
Analysis runs the decision against doing nothing on the same seeds and says what the answer rests on. It works the other way too: a compound question to Analysis has its parts about the record answered from the record, so "how concentrated is our revenue and what happens if we lose our largest customer" gets a lookup for the first half and a simulation for the second.
- One click from an answer to a run of the same question
- The size you asked about is the size it reads for
- Parts of a question about the record are never simulated as if they were decisions
With a key
The same working, written as a paragraph
When a language model is configured it is handed the finished answer: the headline, the findings, the confidence and the numbered evidence. Its brief is narrow. Write this up, mark every sentence with the row it came from, compute nothing new beyond simple arithmetic, and say so when a row is an industry default.
Any sentence that comes back with a number and no marker is counted and shown, so a paragraph that drifted from its evidence is flagged rather than trusted. The worked answer stays on the screen under it either way.
Honestly
What Ask will not do
It is an analyst of one company's record, working to written rules. The edges of that are worth knowing before you rely on it.
- It reads questions against thirty three concepts and your graph. An oddly phrased question can miss, and when it does it says so and suggests what it can answer
- It will not tell you what your market does next year beyond what the simulation of your own company says
- It will not invent a figure; every number is a row, a line in a file or arithmetic written down beside it
- Its confidence measures where the numbers came from, not whether the world will agree with them. The decision register is where that gets checked
Ask your own twin how concentrated your revenue is
The first question most people ask is not about the future. It is whether the model has understood their company, and that is the question Ask answers best, with every figure one click from the file it came from.