Artificial intelligence behind the platform

What the engine is

The engine is a statistical analysis system that turns very large volumes of market data into short written readings. "AI" here means machine-learning models trained to detect pattern regularities, combined with rule-based checks that encode plain market logic, such as how volatility is measured and how events are matched to reactions. It is software we operate and inspect, not a third-party black box bolted onto the site.

Its job title is analyst's assistant, not oracle. It exists because the volume of raw information relevant to even a short watchlist exceeds what any person can read daily; the engine reads it all and writes down what mattered. The quality that makes it useful is discipline: it applies the same tests every session, without fatigue, mood or a position of its own.

What it reads

Inputs fall into four families. Market data: prices, volumes and order-flow statistics from the venues whose names appear on our homepage chips. Trend and volatility measures: computed features describing each asset's structure and regime. Event data: results announcements, economic releases and news items, time-stamped and matched to price context. History: each asset's own past behaviour, which the engine uses as its baseline for what is normal.

What it does not read is you. The engine does not use your personal data, your balances or your trading history to produce its readings; summaries are computed per asset and per market, identically for every user. Your watchlist decides which readings you see, nothing else.

How it summarises

The engine's output is built in three stages. First, feature extraction computes the measurable facts: where price sits in its range, how volatility compares with the asset's own recent history, which levels were crossed. Second, pattern comparison asks how the current configuration resembles stretches of the past and what tended to follow them. Third, composition writes the summary: a short reading that names its evidence and its confidence, in English rather than code.

Summaries are deliberately phrased as readings rather than instructions. "Trend intact, volatility rising, reaction to results muted" is a state of the world you can act on or ignore; "buy" would be a decision the engine is not built, and we are not permitted, to make for you. The phrasing is the product of policy, not modesty.

The reasoning trail

Every alert and summary links to its reasoning trail: the specific signals that fired, their values, and the rule or model that flagged them. This is the most important feature on the platform and the one we would keep if forced to choose. It lets you audit the machine, disagree with specific evidence, and build a personal sense of when the engine's readings deserve weight, which is a form of knowledge no score can substitute for.

The trail also protects the honesty of the whole exercise. An analysis system whose conclusions cannot be inspected drifts toward whatever sounds wise; one whose every sentence cites its inputs has nowhere to hide. Your adviser uses the same trails on review calls, so the conversation is anchored to evidence rather than recollection.

Where it runs and how often

The engine runs continuously against live data, with readings refreshed on a schedule set by your plan: standard cadence on Basic and Advanced, highest cadence on Premium. Event-driven items, such as results announcements and their reactions, are processed as they occur. Computation is server-side; your browser only receives the finished readings, which keeps the dashboard light on any device.

What the engine is not

It is not an autonomous trader. It holds no keys, places no orders and cannot move funds; the read-only exchange connection described on the security page makes that impossible at the permission level. It is not a profit guarantee or a predictor: no statistical system eliminates market risk, and claims to that effect from any service are false. And it is not personal investment advice; it is information, offered for your own judgement.

Limits and errors

The engine's readings are probabilistic and sometimes wrong. Patterns break at news, rare events by definition resemble nothing in history, and model confidence can be high exactly when the situation is novel. Two design choices manage this: summaries carry confidence phrasing rather than certainty, and unusual configurations are flagged as unusual rather than forced into familiar templates. Your adviser's review adds a second judgement layer where it counts.

We also publish no performance claims for the engine, for a simple reason: any number we could quote about past signals would invite the wrong conclusion about future ones. The reasoning trail lets you evaluate the engine on your own watchlist, which is the only sample that matters to you.

Human review

Readings are automated; judgement is staffed. Your adviser reviews flagged configurations on your monthly call, answers why a summary reads as it does by walking the trail with you, and escalates genuine data anomalies to the desk that maintains the engine. The division is deliberate: machines for coverage and consistency, people for context and accountability, and you for decisions.

Your control

You choose the assets followed, the alert levels and the plan; you can widen, tighten or mute anything; and the decision log records what you concluded and when. If you want the engine's contributions reduced to a bare minimum, alerts only, no summaries, the settings allow it. The service is arranged so that the more you understand it, the less of it you need.

Getting started

Register on this page's form and the first call walks you through a live example, an asset you name, its current reading and its reasoning trail, so the engine explains itself before you fund anything. Plans start at the £250 Basic minimum with a £19 monthly subscription, and the products page shows the other pillars, equities monitoring and personal support, that share the same engine and the same account.

Worked example: one reading, opened up

Suppose the summary line for an asset reads: trend intact, volatility rising from its own recent low, no event attached. Opening the trail shows the three inputs behind it: the range calculation that defines "intact", the volatility ratio against the asset's trailing baseline, and the empty event table that earned the third clause. Each number is date-stamped and each rule is named, so "intact" is a claim you can check rather than a mood.

That same reading, a month later, might add that the configuration resembles two prior stretches, and name what followed each. This is the engine's pattern comparison at work, and it is phrased as resemblance deliberately: two precedent stretches are context, not a prophecy, and the confidence phrasing says so. Reading the trail turns the summary from an assertion into an argument, which is the only form of analysis worth having.