How is every Unloq answer evidenced?

What comes with each answer?

Each answer arrives with its evidence attached, however the question was phrased. The trail holds these parts:

• The plan. The steps Unloq decided to take before running anything, in order.

• The query. The SQL that ran, shown in full, so your analyst can check exactly what ran.

• The documents. Each document Unloq used, with the page and the excerpt.

• The logic. How rows became the figure: which periods were summed, which accounts were included and which were left out.

• The validation grade. Whether each figure matched the calculation that produced it when Unloq checked it.

Take a question such as “why did gross margin fall in the north last quarter?” The answer would show a plan (look up the agreed margin definition, then pull sales and cost by region, then read the supplier contract that changed), the SQL that pulled the figures, the clause in the contract with its page number, the arithmetic that split the fall between price and volume, and a pass. Every figure a manager reads can be followed back to one of those.

The logic is the part that settles most arguments about a number. Two people can pull identical rows and still total them differently, one including credit notes and one leaving them out, and only someone who reads SQL would spot which choice was made. Unloq writes that choice down next to the figure. The trail is there for whoever checks the figure before it goes in a board paper.

How is each figure produced?

Each figure is produced in code. A metric resolves to one canonical, team-owned definition before any query runs, and the model only plans while deterministic code executes. Every query is governed SQL against an approved list of tables, and the calculations run in deterministic tools, so identical inputs produce an identical result.

When Unloq says how much of a change each driver accounts for, that split is calculated in SQL and shown as a waterfall your analyst can trace back to the SQL.

What do pass, warn and block mean?

Before anyone sees an answer, Unloq cross-checks every figure against the calculation that produced it. Unloq then grades the answer:

• Pass. Every figure matches the calculation that produced it.

• Warn. The answer arrives with its caveat attached, so the reader knows what to check before relying on it.

• Block. The figure is withheld and reaches nobody. An answer that breaks a business rule you have agreed is blocked before anyone sees it.

Can Unloq say it does not know?

Yes. When your data or documents do not support an answer, Unloq says there is not enough evidence. Where the data is partial, Unloq says so.

What stops a recommendation breaching a limit?

A recommendation that would breach a limit you have set is flagged before you approve it. Before a recommendation reaches you, Unloq tests it against your budgets and limits, such as a spend ceiling or a minimum margin, and you see any breach beside the recommendation. Unloq ranks the options with the evidence attached, and a person decides. Once a recommendation is approved, it goes to a named owner in Unloq, your CRM or Slack, and Unloq can suggest who that owner should be.

What is recorded when someone approves a recommendation?

A person signs off every decision, and Unloq records the approval with the outcome you expected. Months later, Prove sets that expected outcome beside what really happened and what you did next. Unloq measures each decision against a do-nothing baseline, and it can only measure actions it can see. How that baseline is agreed, and how Unloq checks its forecasts and odds, is set out on how Unloq checks its own numbers.

For a supplier renegotiation approved in spring, the margin you expected goes on the record that day, before anyone knows the result. In our experience, the expected outcome is the line most decision papers leave out, so a year later the team cannot say whether the call worked.

What this page does not cover

Testing measures how often answers are right across a whole set of questions. Ask us for our testing figures. Why a number moved belongs to Contextualise, where Unloq ranks the drivers of a change by their share of it. On Recommend, Unloq ranks the options for what to do next, and Prove measures what each approved decision was worth once the results are in. Whether Unloq is a chatbot for your data is answered on the platform.

Ask us your question

Should your reviewer want to see the trail on a question of your own, ask us your question. Our AI engineering team works the question through with our technical founders and replies with an answer.