[UPDATE 10/04/2025]
Our WARHAMMER III Campaign AI Beta #2 is now live on Steam! Instructions for opting in to the Beta can be found in the Dev Blog below, with instructions on how to provide your feedback and bug reports available in our FAQs. Please note that you may need to restart the Steam client for the Beta to be detected. Enjoy!
Click HERE for the Campaign AI Beta #2 Feedback Survey Form
Click HERE for the Campaign AI Beta #2 Sub forum
Click HERE for the Campaign AI Beta #2 FAQs
Hey folks,
The time has come for us to talk a little more about Campaign AI, what we’ve been working on in the wake of the first Beta, and how we’ve used your feedback to help steer this big oft-unwieldy ship in the right direction.
We’re pleased to reveal that tomorrow (Thursday 10th April) we’ll be launching our next Campaign AI Beta on Steam, once again giving you the chance to test out our next stage of tweaks and adjustments and provide us with your immensely valuable feedback. If you took part in our last Beta, we’d love to get your thoughts on this latest round of changes, and if you missed out last time, this is the perfect chance to see what we’ve been working on and have your voice heard. You’ll find instructions on how to access the Beta at the bottom if this blog when it goes live tomorrow, so check back when you see us confirm it’s live.
As ever, you’ll be able to discuss these changes and provide feedback in our dedicated Campaign AI Beta forum in the CA Community, as well as via a survey that we’ll be running alongside the Beta. These links will appear here tomorrow, and you’ll also find a link to the survey in-game, on the News Feed when you’re playing the Beta itself.
Before we get into the nitty-gritty of what Rado and the team have been up to, we’d also like to share that this project will henceforth be inducted into our rekindled Proving Grounds initiative. Some of you may recall this initiative from the WH2 days, but for those who don’t, let me give you a brief introduction to what the Proving Grounds are:
The Proving Grounds first debuted back in 2020 and gave our devs the opportunity to explore radical solutions and test systemic changes to core gameplay features, while giving you the opportunity to try out these changes and provide feedback on them.
Of course, there’s no guarantee that these tested changes make it into the main game, but it allows our devs to think outside the box a little and allows you to have your voices heard on what can be some pretty sweeping changes to core parts of the game.
Basically, it’s a chance for us all to work together on improving WH3 for the greater good. The Greater Good.
With this current phase of Proving Grounds, we're quietly confident that these Campaign AI changes are moving in the right direction. These new changes we're now testing for will help to show us what further refinements we need to make before we consider bringing them into the main game, and how that might impact the difficulty of our player experience.
Looking beyond this Proving Ground, we have a team internally starting to explore changes that we think we can make to Sieges, and are hoping to bring you more on that in the summer.
With that admin out of the way, it’s once again time to hand over to the almighty Radoslav. Be sure to check back tomorrow for all the details on how to access this new Campaign AI Beta!
— Steve Coleman // Senior Community Manager // Apprentice (Proving-)Groundskeeper
Hello once again, I’m Radoslav Borisov, Principal Technical Designer at CA Sofia.
Me and the team are back to address further issues with the Campaign AI.
Before I dive into specifics I want to put a big caveat up front - we have read your feedback and are well aware there are other parts of the core game experience you all want us to work on. Several topics that come up regularly include Sieges, Autoresolve, Diplomacy, Endgame and overall difficulty. While those topics are outside the scope of the current BETA, we are making plans to work on those later this year. The scale of work on the abovementioned topics likely will not allow us to address everything, but we want to start from somewhere. The next game areas we will be working on are going to be Sieges and Autoresolve, expect more information from us later in the year.
Moving back to Campaign AI, in BETA 2 we are going to focus on the following topics:
Anti-player bias - there are a number of issues arising from the ways in which human and AI factions are unequally evaluated.
Distance scaling - we received a lot of feedback that allowed us to identify and isolate several issues with the way tasks are evaluated and prioritized for all AI factions.
Aggression - a new iteration of the changes originally introduced in BETA 1. We are focusing on keeping the AI more aggressive and dynamic, while mitigating instances of fixating on the player or suicidality.
Faction potential - a new iteration on the changes from BETA 1, with a focus on mitigating edge cases around important minor factions and introducing a bit more variance between campaigns.
If you need a refresher you can check out the BETA 1 blog here.
And now for some more details on the topics outlined above.
1. Anti-player bias
Threat vs strength
There exist two different measuring scales within the AI - threat and strength.
Threat Overview
Every faction in the game has a threat evaluation, based on a variety of factors, such as diplomatic activity, interactions with campaign objects via agent actions/occupation decisions, region count, among others. The main purpose of the threat score is to inform the AI how powerful and influential other factions even without them being allied or hostile. The abovementioned threat evaluations result in two distinct scores - global strategic threat score and individual threat score. Global strategic threat is a like a leaderboard of who's the most powerful and influential.
Individual threat scores are individual for every faction and reflect their view. Global strategic threat score is used as a baseline for these calculations, but factions take additional parameters in considerations, such as proximity, diplomatic attitude, direct aggressive actions, among others. The impact of threat scores are wide reaching and affect deal evaluation, diplomatic attitude, task generation. There are additional distinctions between a faction's main threat and common threats for several factions, but those are a subject for another time.
Strength Overview
Whenever the AI needs to understand how strong any force is in battle it does a strength evaluation.
Said strength evaluation takes into consideration everything the AI knows about it - units, experience, bonuses, stances, among others. Every time a task is being evaluated a strength evaluation is performed on the target. Depending on the result of the evaluation all owned forces are evaluated to find a force we can justify resourcing. Finding a force with a good strength ratio against the target is critical, and failing to do so requires additional resources be expended to recruit.
What we changed
In this beta we are changing the threat player actions generate to be between 90% and 120%.
Additionally a new multiplier that changes the player's strength from 100% always to between 100% and 120%, based on AI personality and campaign difficulty level. Also the vast majority of tasks of the “attack human forces/settlements” variety have been removed, and the ones that were not removed have had their priority greatly reduced.
Why we changed It
The current biggest source of anti-player bias in the game is the threat system. There exists a multiplier that changes the threat player actions generate from 50% to 200%, based on AI personality and campaign difficulty level. This causes player threat to scale very differently compared to the AI. The goal of these changes is to have the AI evaluate the player much closer to how it evaluates other AI's. The addition of a strength multiplier aims to align the way different systems perceive the player, leading to more consistent behavior overall. Since the threat system affects a variety of other systems, such as task generation, diplomatic attitude calculations, and deal evaluation, improvements in the AI's behavior in those areas is expected as well.
A prime example of an issue these changes aim to resolve; If an AI faction borders the player it will go into a defensive posture, mostly due to tasks of the “defend regions neighboring threat” variety, but factions that are close but not actually bordering the player will aggressively attack due to tasks of the “attack threat” variety.
How will these changes affect you in-game?
Changing the level of threat that player actions generate means that the Campaign AI will evaluate player action with less bias, while retaining the ability to respond appropriately when attacked. Meanwhile the multiplier should help make the Campaign AI’s aggressiveness scale more closely with the difficulty settings you choose. Finally, removing and de-prioritising the Campaign AI’s preference for attacking human players over AI factions will take the target off your back, unless warranted by your playstyle.
2. Distance Scaling
Inside the AI when tasks are being evaluated, before actually being assigned to a specific resource for execution, their priority gets modified based on how far away the target is from the asset the AI wants to use. The AI’s perception of "how far away" is measured in turns. For existing assets the calculation takes into account movement extends, and where appropriate the usage of stances like tunneling/forced march. When the asset being used is a fresh army that needs to be hired the calculation additionally takes into account the distance to recruitment region, and time to recruit.
Scaling overview
Currently the time and distance scaling is constrained so tasks have a minimum priority if they take more than 4 or 5 turns, for recruiting and non-recruiting respectively. The intention behind the current setup is to restrict the AI's horizon so it doesn't venture out from it's holdings without a very good reason. For a majority of cases this works fine, but we noticed several cases where a short horizon does not allow the AI to do certain map traversals without access to tunneling stance or movement bonuses. One example is the Southlands World's Edge Mountains province - the AI struggles to invade it normally.
What we changed
To combat this in this BETA we are going to introduce a scaling component, based on campaign progression, that changes both time and distance scaling dynamically.
Current values are:
Minimum of 20% priority at above 4 turns away when recruiting;
Minimum of 10% priority at above 5 turns away when not recruiting;
New values when AI faction has more than 15 regions:
Minimum of 15% priority at above 6 turns away when recruiting;
Minimum of 15% priority at above 7 turns away when not recruiting;
New values when AI faction has more than 30 regions:
Minimum of 20% priority at above 7 turns away when recruiting;
Minimum of 20% priority at above 8 turns away when not recruiting;
Why we changed it
The current setup works well at restraining the AI from breaking up it’s territory and avoiding cases of province-hopping and border gore. This holds true for the early game very consistently, but as turns pass and empires become bigger it also results in undesirable behavior.
Two types of issues we are aiming to resolve with these changes are:
Certain regions/provinces being inaccessible to some factions when AI.
Larger AI empires stagnating and armies recruited in central regions getting “stuck” there due to borderlands being far away enough so that tasks targeting enemies in/around them hit minimum priority.
How will these changes affect you in-game?
You should see AI empires expanding more consistently and see less “stuck” armies idling in their core provinces that are too far away from danger to provoke action. This should make the mid and late game more dynamic as well as making powerful AI empires spread their assets out more appropriately, simultaneously making the Campaign AI more ambitious while opening up strategic opportunities to attack them as their borders expand and soft spots organically appear in their defences.
3. Aggression
We’ve further tweaked how the AI assesses the strength of a rival factions, with a strong focus on Very Hard and Legendary difficulty.
What we changed
Average enemy strength and threat modifiers for strategic calculations reduced from 20% to 10%.
Very Hard/Legendary modifiers for defensive/cowardly factions now allowed to exceed 100%.
Why we changed it
These changes do not aim to introduce new behaviors, but rather to increase the precision of calculations.
Introducing margins of error or modifiers to the raw calculations the AI performs has been used as a method of introducing difficulty and behavior variety, but both feedback and data show that wider margins disproportionately worsen the gameplay experience. The abovementioned margin of error is the leading cause of AI factions misjudging their chances and suiciding armies or not leaving enough defenders in threatened territories.
How will these changes affect you in-game?
These changes should prevent the AI from being overly aggressive when it makes no sense to be, resulting in strategically poor decision making that in turn can result in players fighting one sided battles that don’t make sense in the context of their campaigns, and are aren’t fun to play.
4. Faction potential
After reviewing data and reading feedback from BETA 1 the team and I reached the conclusion we need a new category of faction potential for certain minor factions. One of the primary goals of the faction potential system is to regulate the strength of individual factions, when played by the AI, in a dynamic way. This rather direct way of manipulation is very effective, but it also has some downsides. The main downside is that the more variance we introduce, be it by shifting more of faction potential’s regulation to a random roll, or by introducing more categories, the less we can rely on averages. I believe that the best approach is to continue tweaking faction potentials with subsequent updates, but the goal for the current Beta is to establish a baseline we can build on top of.
With all of that said we are aiming for the following:
All major factions should win their respective theaters at least some of the time.
Minor factions, that are not starting enemies for major factions, should have an opportunity to survive at least until the mid game.
The player should be able to influence the development of factions they’ve met.
Developments under the shroud should have sufficient variance, to the point where it’s hard to accurately predict the state of a theater before scouting it.
What we changed
Lower the faction potential of a few outlier factions that have been overperforming:
Oracles of Tzeentch, Blooded Wanderers, Cult of Pleasure, Khemri.
Increase the faction potential of several underperforming factions:
Clan Moulder, Warhost of the Apocalypse, Hag Graef, Itza, The Bloody Handz, Caravan of Blue Roses, Puppets of Misrule, Bonerattlaz;
New category of faction potential will be applied to the following minor factions:
All Elector factions; Caledor, Ellyrion, Saphery; Overlords of Zharrduk; Varg, Goromadny Tribe, Kul, Skaeling; Zhufbar, Karak Hirn, Greybeard's Prospectors; Imperial Wardens, The Jade Custodians; Crooked Moon Mutinous Gits, Teef Snatchaz; Ropsmenn Clan; Xlanhuapec, Defenders of the Great Plan; Clan Verms, Clan Spittel, Clan Morbidus; Rakaph Dynasty; Templehof, and Mousillon.
Why we changed it
We heard you loud and clear that having a variety of minor factions survive and potentially thrive is something you want to see. We have chosen these specific minor factions for their gameplay importance in their respective theatres, and we’ve avoided factions that are starting enemies of playable factions where possible. Additionally major factions that were over or underperforming have been adjusted accordingly.
How will these changes affect you in-game?
In a campaign these changes to Faction Potential should make sure that major factions assert the expected level of control over their theatres e.g. Reikland over The Empire etc., but without wiping out every minor faction within reach. This should establish a more reliable baseline that keeps more major and minor factions in play for longer, allowing players to influence the development of the factions they meet and making a bigger strategic impact. The new degree of unpredictability under the shroud should also create more variety between different campaign playthroughs while also preserving the strength and influence of major factions.
Miscellaneous fixes
Several additional bugs around how certain AI factions navigate and make use of campaign features:
- Sisters of Twilight will now try and actively pursue a specific item set in the Forge of Daith;
- Disciples of Hashut, The Legion of Azgorh and The Warhost of Zharr will better navigate the Hell-Forge and focus on obtaining high tier manufactory options;
- Vampire Counts factions will prioritize going for the Bloodline they are thematically aligned with;
- Improved AI handling of the Bloodletting mechanic;
- [Realms of Chaos specific] All Cathay factions will now properly try and capture back bastions;
- [Realms of Chaos specific] Kurgan Warband will now properly choose targets after destroying bastion settlements;
Thanks for reading, and enjoy the Beta!
— Radoslav Borisov // Principle Technical Designer
How to download the Beta branch
If you would like to opt-in to this Beta branch and try these changes today, here's how you can do it:
Open your Steam client and go to your Library.
Right-click on Total War: WARHAMMER III and select 'Properties'.
Navigate to the 'Betas' tab.
In the drop-down menu, select the option ‘proving_grounds - Campaign AI Beta #2'
Steam will then download the necessary files and begin the patching process to install the Beta.
The download is very small, Steam will only need to patch around 248MB of game data.
How to submit your feedback
If you would like to complete the feedback survey once you've played the Beta, here's how you can do it:
1. Follow this link
2. Complete the survey and submit your feedback
If you would like to discuss your experience of playing the Beta with us and other players, here's how you can do it:
1. Follow this link to the Campaign AI Beta #2 Sub forum
2. Post in it
Thank you in advance for your feedback and for discussing your experiences with the Campaign AI Beta #2!