Tactical Briefing: AI-Generated Pre-Match Storylines, Decoded
A football fan reading a probability split has two questions in their head. The first is "what does the model say?". The second is "what is the story behind that number?".
The probability answers the first. Tactical Briefing answers the second.
This article explains what Tactical Briefing does, how it is generated, and what to expect from it on different types of fixtures.
What the briefing actually contains
A typical Tactical Briefing is 200 to 300 words and covers four areas:
Recent form trends. What pattern have the two sides shown in their last five fixtures, and how does that pattern interact with the upcoming opposition? A side outperforming its xG in attack against weak defenses is a different read than a side suppressing its opponents' xG against strong attacks.
Key matchups. Where on the pitch does the fixture get decided? Often this is a winger versus a full-back, a striker versus a center-back pairing, or a midfield duel that controls possession. The briefing identifies one or two matchups specifically.
Notable absences. Who is out, and how much does it matter? The briefing pulls from the lineup feed and identifies absences whose impact would be felt strongly given the rest of the side's structure.
Tactical context. Title race or relegation battle? Cup-tied players? Recent rotation patterns? European fixture in midweek? These contextual factors do not always make the analysis output but they shape how a fan should read the fixture.
The four areas are not always present in every briefing. Some fixtures have no notable absences. Some fixtures are mid-table mid-season with no broader context. The briefing covers what is actually relevant to that specific fixture, not a fixed checklist.
How the briefing is generated
The briefing runs through three phases.
First, the base analysis completes. This produces the structured data that the briefing will be grounded in: probability split, recent form summaries, head-to-head record, expected lineups, injury and suspension list, league context.
Second, this structured data is composed into a prompt. The prompt instructs the LLM to write a short narrative covering the four areas above, with specific constraints: do not invent statistics, do not contradict the structured data, do not produce betting suggestions, keep the tone analytical rather than promotional.
Third, the LLM (Amazon Bedrock Nova 2 Lite) generates the briefing. Generation typically takes 2 to 4 seconds. The result is rendered in the app under the analysis result.
The grounding is the critical piece. The LLM is not asked "tell me about Arsenal vs Chelsea." It is given the actual match data and asked to produce a narrative that is consistent with it. This stops the model from hallucinating fictional injuries, made-up recent results, or invented tactical schemes.
When the briefing adds the most
Three categories of fixture get the most value from a Tactical Briefing.
High-stakes fixtures with subtle context. A relegation six-pointer where one side has won three in a row and the other has not won in eight, but the recent form scaler is misleading because the eight-match drought includes fixtures against the league's top half. The briefing surfaces the underlying read: this is a closer fixture than the form table suggests.
Fixtures with notable absences. A side missing two of its top three goalscorers, with a backup striker who has scored once in 14 appearances. The briefing flags the absence as material and explains what it changes about the team's likely shape.
Big-match context. A title race fixture between two sides separated by two points with three matchdays left. The briefing covers stakes, head-to-head history, recent form against quality opposition, and any tactical patterns the two sides have shown when meeting each other.
In each of these cases the probability number alone does not communicate why the fixture is interesting. The briefing fills that gap.
When the briefing adds little
Two categories of fixture get less value from the briefing.
Routine fixtures with no context. A mid-season mid-table fixture between two stable sides with no notable absences and no recent form anomalies. The briefing in this case produces correct but dull commentary: "both sides are mid-table, both have steady form, expect a tight contest." It is not wrong but it is not informative.
Fixtures where the data is sparse. Lower-division fixtures, regional cup fixtures, or fixtures very early in a season. The briefing has fewer data points to work with and produces less specific commentary as a result. Premium subscribers in these cases often find the structured analysis output more useful than the narrative.
How to read it
The briefing is one input among several. The probability output is the model's mathematical read. The briefing is the same model's verbal read on the same data. They should agree, and when they do not, that disagreement is itself information.
The briefing is also a check on your own read. If you opened a fixture expecting one side to dominate, and the briefing argues for a tighter contest with specific reasoning, that is a moment to either defend your prior or update toward the briefing's read. The briefing is not the truth; it is a structured second opinion.
Tactical Briefing is included in Premium tier on iOS, Mac, and Android. The same briefing is generated regardless of which platform you open the fixture on, since the underlying inference runs server-side and the result is the same.