- Joined
- Jul 11, 2008
- Messages
- 1,940
- Reaction score
- 185
- Points
- 63
There's a different version that serious analysts do. They build a specific profile of how a team is likely to set up against a particular opponent, what their defensive shape looks like against teams that press high versus teams that sit deep, how their build-up changes when they're managing a lead versus chasing a game, which specific players are responsible for the behaviours that drive their attacking output. That version takes two hours per team and produces something you can actually bet against.
The LLM-assisted approach sits between these. With the right inputs and the right prompts, you can produce a profile that's genuinely specific and match-relevant in thirty to forty minutes - closer to the serious analyst version than the casual ten-minute scan, without the two-hour manual synthesis. The catch is that the quality of the output depends entirely on the quality of what you put in and how you ask for it. Generic inputs produce generic outputs. Match-specific inputs with specific prompting produce something usable.
This article covers the inputs, the prompts, and how to connect the output to your quantitative analysis rather than treating it as a substitute for it.
The Generic Team Summary Problem
Ask an LLM "tell me about [team]'s tactical approach" and you'll get three or four paragraphs of accurate, uninformative description. The team plays a high press. They like to build from the back. Their fullbacks are attack-minded. Their striker drops deep to link play. All of this is true, none of it is useful for betting a specific upcoming fixture, and the model produced it by drawing on training data rather than by analysing the specific material you needed it to analyse.
This is the failure mode the stress-testing article described in a different context. The model fills gaps with plausible generalisations when the prompt doesn't force specificity. In opposition research, the gap is enormous because "tell me about this team" is essentially an invitation to produce a press release.
The fix isn't a better general question. It's providing the specific match reports, transcript fragments, and formation data you want synthesised, and asking specific questions that can only be answered from that material rather than from general knowledge.
The distinction: "how does this team typically defend?" produces training data recall. "Based on the three match reports I've pasted, how did this team's defensive shape change between the first fixture where they faced a high press and the second where the opponent sat deep?" forces analysis of the material you've provided. The second question is the one worth asking.
What to Collect Before Running Any Prompts
The profile quality is bounded by the input quality. Three to four recent match reports from reputable tactical sources, a press conference transcript from the most recent session, and the team's formation and possession data from FBref for the last five matches. That's the minimum. Each input type carries different information.
Match reports for tactical texture. Not match reports from mainstream outlets that describe what happened - match reports from sources that describe how it happened. The Athletic's match analysis pieces where available, tactical blogs covering the competition, BBC Sport's extended match centres that include pass maps and shot locations. You're looking for reports that describe defensive shape, pressing triggers, build-up patterns, and how the team responded to specific situations. The Telegraph and Guardian occasionally produce this quality of match analysis for high-profile fixtures. For Championship and Scottish Premiership, the local newspaper beat reporters who attend every match are sometimes the best source because they describe tactical patterns from direct observation rather than from highlights.
Three to four reports gives the synthesis enough to work with. More than six and you're adding diminishing returns on input time. Prioritise the most recent three matches and include one report from a fixture against a team with a similar profile to the upcoming opponent - same pressing intensity, similar formation, comparable league position and playing style.
Press conference transcripts for upcoming match framing. The press conference transcript article covered the extraction workflow. For opposition research specifically, you want the manager's comments about the upcoming opponent, their assessment of their own team's current shape, and any tactical language around preparation. A manager who mentions working specifically on defensive transitions in training this week is telling you something about what they expect from the opponent. Most don't give information this directly, but the absence of certain topics is itself signal - a manager who doesn't mention set pieces when facing a team with a strong set piece record is either under-preparing for that threat or managing the information deliberately.
FBref formation and possession data for quantitative grounding. The five-match rolling formation data, PPDA, progressive carries allowed per match, and defensive actions by third give the qualitative analysis a quantitative context. The tactical profile you build from match reports needs to be consistent with what the numbers show - if the reports describe aggressive pressing but the PPDA suggests low pressing intensity, something is wrong with either the qualitative assessment or the quantitative reading, and that inconsistency needs resolving before you use either input in analysis.
Pull all three input types before opening the LLM conversation. Having them ready means the session is analysis rather than research, which keeps the work focused.
The Profile-Building Prompt Architecture
The opposition profile is built through a sequence of prompts rather than a single comprehensive one. A single prompt asking for everything produces a long generic response. Sequential prompts that each address a specific analytical question produce sharper outputs that accumulate into a useful profile.
The sequence has four prompts.
Prompt one: defensive shape and pressing behaviour.
"I'm going to paste three match reports covering [team]'s last three fixtures. Read them carefully and answer these specific questions from the reports only - do not draw on general knowledge about this team. How did the team defend in each fixture - what was their shape out of possession, where did they set their defensive block, and did this change based on the match state? Where did they press and what were the triggers - which specific situations or player positions caused them to engage higher up the pitch? Were there specific moments in the reports where the defensive shape broke down, and what caused those breakdowns? Do not summarise the matches - answer only these questions with evidence from the reports."
Prompt two: build-up patterns and attacking structure.
"Using the same three match reports, answer these specific questions. How did the team initiate attacks - through goalkeeper distribution, central build-up, direct play, or transitions? Which players were most involved in carrying the ball out from the back and into midfield? Were there specific patterns in how they created chances - recurring combinations, specific areas of the pitch where attacks originated, particular players who appeared repeatedly in chance creation? How did the attacking structure change when they were ahead versus when they were level or behind? Again, answer from the reports only."
Prompt three: match-specific adaptation.
This is the prompt that separates useful opposition research from generic team profiling. You're asking the model to identify how this team specifically adapts to the type of opponent they'll face in the upcoming fixture.
"Based on the match reports I've provided, identify how [team] performed in the fixture where they faced [description of opponent profile most similar to upcoming opponent - e.g. 'a high-pressing team that played a 4-3-3 and looked to press from the front']. Specifically: how did their defensive shape change compared to the other fixtures, how did their build-up adapt to the pressing intensity they faced, and were there any visible vulnerabilities or strengths that appeared specifically in that matchup type? Then contrast this with their performance against [description of opponent at the other end of the tactical spectrum]. I want to understand how adaptable this team is tactically and where their system shows strain under specific types of pressure."
This prompt requires that one of your three match reports is against a team with a similar profile to the upcoming opponent. That's why the input collection step specified including a report from a stylistically similar fixture. If you don't have that match in your inputs, this prompt can't be answered from the material and the profile has a significant gap.
Prompt four: the betting-relevant synthesis.
"Based on everything you've extracted from the match reports, identify the three most specific and actionable observations about [team] that would affect how you'd price the upcoming fixture against [opponent]. Not general strengths and weaknesses - specific patterns in how they play that the market's standard model is likely to be pricing from averages rather than from current form and tactical specificity. For each observation, explain what market it most directly affects - total goals, Asian Handicap, specific player props, or in-play market dynamics - and why the match-report evidence suggests the market might be mispriced in that area."
That final prompt is the bridge from tactical analysis to betting decision. The answer to it isn't the bet - it's the list of hypotheses to check against your quantitative analysis before deciding whether there's a position worth taking.
Connecting the Tactical Profile to Quantitative Analysis
The tactical profile produces qualitative hypotheses. The quantitative analysis tests whether those hypotheses are supported by the data. Both are necessary. Neither is sufficient alone.
The integration process works like this. The tactical profile flags three specific observations. For each one, you identify the quantitative test that would confirm or challenge it.
If the profile identifies that the team's high defensive line creates vulnerability to balls in behind when the opposition can bypass their press - that's a testable hypothesis. The quantitative test is the team's xG conceded from counter-attacking situations in their FBref data, their defensive line height as implied by how far up the pitch they concede possession, and the upcoming opponent's progressive carry rate and direct play tendency. If the numbers support the tactical observation, you have a confirmed hypothesis and a specific market implication. If the numbers contradict it - if the team's xG conceded from transitions is low despite the match reports suggesting vulnerability - something needs resolving before you act on the hypothesis.
The resolution might be that the match reports are reflecting a sample of unusual fixtures where the vulnerability appeared, which the longer-run data shows is not typical. Or it might be that the data is reflecting a phase of the season where the vulnerability didn't exist but the recent match reports are showing it emerging. The press conference transcript sometimes helps here - a manager who has recently been discussing defensive transition work in training sessions is consistent with an emerging vulnerability that hasn't yet shown up strongly in the aggregate data.
Either way, the integration step is the analytical work that the profile can't replace. The profile surfaces hypotheses efficiently. Whether those hypotheses are worth betting on requires the quantitative grounding.
The Comparison Profile: Building Both Sides
Opposition research covers one team. A match profile requires understanding both.
Run the same four-prompt sequence for both teams in the fixture, using their respective recent match reports. Then run a fifth prompt that synthesises the two profiles:
"I'm going to paste two tactical profiles - one for [home team] and one for [away team] - built from recent match reports. Identify the specific tactical interactions that are likely to be most significant in this fixture. Where does one team's offensive approach interact with the other's defensive vulnerabilities? Where does each team's system create problems that the other's approach is well-equipped to handle? What specific match situations are most likely to be decisive given the two profiles - for instance, if one team presses high and the other's build-up is reliable under pressure, the press is unlikely to be the key variable; if the build-up is vulnerable to pressing, it may be central. Produce a match-specific prediction of how the game is likely to unfold tactically, and identify the market implications of that prediction."
The interaction analysis is where opposition research earns its most specific value. A team with excellent build-up quality facing a team that presses aggressively is a different fixture from the same team facing a team that sits deep - and both are different from the same build-up quality facing a team that presses inconsistently and creates transition opportunities when the press is beaten. The interaction between the two profiles is what determines the match script, and the match script is what determines which market types have meaningful edge available.
Calibrating the Depth to the Bet Size
The full workflow - four prompts per team plus the interaction prompt, built from three match reports each - takes between forty-five minutes and an hour for a fixture you haven't researched before. Faster for teams you track regularly and have profiles built up from previous weeks.
That's the right investment for a significant prop bet, a large Asian Handicap position, or a fixture where the tactical interaction is central to why you think the market is mispriced. It's too much for a small-stakes bet in a competition you know well where the signal is primarily quantitative and the tactical check is confirmatory rather than central.
Calibrate the depth to the decision size. A quick version of the workflow - one match report per team, a single synthesis prompt focusing only on the most relevant tactical interaction for the specific market you're considering - takes twenty minutes and is appropriate for smaller positions. The full workflow is for larger ones.
The fixture screening tool identified which fixtures warrant deeper analysis. This workflow is what the deeper analysis looks like for tactically complex matchups where the quantitative signals alone don't close the case. Both tools in sequence - screen first, profile second for flagged fixtures - is the right operational architecture.
The Limits of the Approach
The tactical profile is only as current as the most recent match reports you've fed it. A team that has shifted formation or tactical approach in the last week - because of a new signing, an injury to a key positional player, or a managerial adjustment following poor results - may not have that change captured in any available match report yet. The press conference transcript sometimes catches this in advance. More often it's only visible in the match itself.
This is the genuinely novel situation gap the AI pricing problem article identified. No amount of match report synthesis catches a tactical change that hasn't appeared in a match yet. The profile tells you what the team has done. The combination of press conference signals and the screening tool's squad anomaly flags tells you where the most likely changes might be. Neither gives you certainty about a team that's mid-transition.
The other limit is quality of available match reports by competition. Premier League fixtures have extensive tactical analysis available within twenty-four hours. Championship fixtures have variable coverage. Scottish Premiership, lower-division European competitions, and most domestic cups have sparse tactical reporting that limits what the synthesis can extract. The workflow works best in competitions with rich written coverage. Below that threshold, the quantitative data carries more of the analytical weight by necessity.
Anyway. The profile isn't the bet and it isn't meant to be. It's the structured qualitative layer that sits between the screening tool's initial flag and the decision to take a position. Used correctly - as one input that confirms or challenges a quantitative hypothesis rather than as a self-contained analysis - it produces the kind of match-specific understanding that generic team summaries never get close to.
The gap between teams you have a genuine match-specific read on and teams the market is pricing from seasonal averages is where the edge in this type of analysis lives. The workflow narrows that gap systematically rather than leaving it to chance.
FAQ
Q: Can I run this workflow on press conference transcripts alone if I can't find quality match reports for a specific competition?
Partially. Press conference transcripts carry the upcoming-match framing and some tactical signal - the manager's emphasis on defensive preparation, their comments on the opponent's strengths they're most concerned about, any tactical language around specific phases of play. What they don't carry is the observed evidence of how the team actually executed. A manager describing their defensive approach and a match report showing how it looked in practice are different quality inputs. With transcripts only, weight the qualitative output accordingly - it's hypothesis generation rather than evidence-based analysis, and the quantitative checks carry proportionately more importance in the integration step.
Q: How do I handle managers who are deliberately uninformative in press conferences?
The press conference workflow article covered the baseline approach to evasive managers - the baseline communication profile makes deviation visible even when the baseline is evasive. For the opposition research workflow specifically, an evasive manager is a partial problem rather than a complete one. The tactical profile depends more heavily on match reports than on press conference content anyway. The transcript contributes the upcoming-match framing; the match reports carry the tactical evidence. If the transcript is uninformative, weight the match report synthesis more heavily and treat the upcoming-match framing as uncertain rather than absent.
Q: How quickly does a tactical profile go stale?
A profile built from three matches covering the last three to four weeks is usable for roughly one to two weeks before the oldest match is far enough in the past that tactical drift becomes a meaningful concern. The exception is mid-season tactical shifts - a manager who changes formation or pressing approach in response to a run of poor results can make a profile built last week unreliable for this week's fixture. The press conference transcript is the leading indicator of a shift in progress. If the manager's language around preparation has changed significantly from the previous week, rebuild the profile from the most recent two matches rather than relying on the older one that may predate the change.
