I'm very rarely tempted to respond to a Substack but I'll bite on this one! First, my credentials - I've cut turf since 2012, campaign staff since 2016, participated in probably a half-dozen of these centralized turf-cutting operations as staff (running 3 of them) and a half-dozen more as a volunteer.
I will start by saying that I 100% agree that AI turf cutting is the way of the future and someone should get to work on the project of creating a better turf cutter stat. I love cutting turf, but I also hate auditing the bad turf cut by people who aren't good at it, don't want to do it, and whose time would obviously be better spent out talking to people and winning elections. I won't shed a tear when this entire process is automated and I absolutely believe the technology exists with current AI models to produce turf that is good enough, most of the time, to justify any trade-offs.
That being said, I never allowed people to use the VAN auto-cutter as a data director (except in very rare circumstances like post-election ballot cure canvassing), and none of the data directors I know who are currently on races allow it either, and I fully stand by that decision.
I was familiar with the auto-cutter as it existed through the 2024 cycle and there were two major obstacles to using it as a time-saving tool for GOTV turf cutting specifically:
1. The lines didn't touch. I know this sounds like a petty data person nitpick, but it actually created a real issue. The only reason it makes sense in the first place to cut GOTV turf six weeks before Election Day is if you can re-use the same turfs several weeks in a row, refreshing the universe so new doors populate in the turfs you've already cut. As inefficient as manually cutting turf is, you know what's more inefficient? A canvasser walking by the home of a perfectly good target in order to get to a farther-away target because the target just isn't included in a turf at all. To do this, you need every piece of the precinct allocated to exactly one turf, even if the block/street doesn't currently have any targets. That means drawing turf boundaries so the lines touch almost perfectly, with no overlap and no empty space left un-allocated. Not only did the old-school auto-cutter NOT do this, in my experience, it actually did the opposite - the lines were as close to where the current targets were as they possibly could be, with no buffer around the targets at all. That meant manually adjusting every single turf until the lines did touch - if you didn't do that, you'd be stuck re-auto-cutting every single turf, every single week. Which brings me to...
2. It didn't actually save time. The auto-cutter didn't automate the process of running the search, loading the map, hitting "cut", saving the turf to a folder, or generating the list number. It only automated the process of actually drawing the lines, leaving a lot of manual overhead that would have to be done regardless, plus the manual auditing that would be required. Additionally, the old auto-cutter famously had a bug where it would create WAY more "corners" in the shapes than a human would if they cut the same turf. That meant manually adjusting turfs was absolutely excruciating, because to move the boundaries of a turf, you have to move every single corner. If I were to auto-cut 10 packets in a precinct, and I needed to adjust the boundaries of just one of them, I think it would take me more time to do that than it would take to cut all 10 packets by hand myself. That means "let auto-cutter do a first pass and then have a human audit and fix" just isn't a workable solution.
So that's why auto-cutting wasn't the answer for 2024. Now, VAN says they have a new, "smart" auto-cutter that can cut turf that is much closer to how a field organizer would - not crossing highways, keeping streets together, etc. I haven't personally been able to confirm how well it does that, but I haven't heard any indication that they fixed the two issues above, even though any coordinated data director would tell you that those have always been the two main obstacles to making use of the auto-cutter. If they did fix those issues, that's great. Anecdotally, I heard that the roll-out of the new auto-cutter came packaged alongside a ton of user-experience bugs in the turf cutting tool more generally that make cutting turf a worse experience now, even if you choose not to use the auto-cutter, and that doesn't inspire a lot of confidence in how well they did at implementing the new feature.
I think what's so frustrating for campaign data people is that we KNOW the technology is there for a better auto-cutter, we KNOW exactly what features it needs in order to actually be workable and save time, and we're generally tech-savvy people who are down to automate and try new things if it can actually make our lives easier, but as far as I know, NGPVAN isn't interested in hearing our feedback. Fourteen years of using VAN as a volunteer, FO, ROD, and data staffer, using nearly every single feature, in multiple states and on large and small campaigns, and I can't recall a single time anyone from NGPVAN the company reached out to me to ask why I don't use the auto-cutter they're so proud of and what they could do to make it better. None of the data staff I know who are currently on campaigns have been reached out to to test or provide feedback on the new auto-cutter. It seemed like they just assumed they knew what the problems were and how to fix them and plowed ahead without actually talking to the people who would be impacted by these changes, and in the process, made an already imperfect tool noticeably worse.
The reality is that it doesn't actually matter that much how good each turf that the auto-cutter creates is. They could be pretty bad and still represent an improvement when you take into account the time savings from people not cutting turf by hand! For the auto-cutter to be usable, what matters is how easy it is for a person to review and correct the auto-cut turfs. Until that happens, maybe county party chairs and volunteers on small local races will get some use out of it, but it won't be a viable way to cut all the GOTV turf for a state-wide coordinated campaign with multiple federal races as stakeholders.
thank you for commenting Molly super insightful! I had this half formed when I wrote this last night but I was thinking about the 20 person audit team we had on the PA Team in 2024 and what you said here is exactly it “For the auto-cutter to be usable, what matters is how easy it is for a person to review and correct the auto-cut turfs.” That’s the product feature that makes the most sense for what's missing and really interesting. Thank you for sharing!
This is pretty indicative of the discourse on field and GOTV work. I am in my early 60’s and have worked on many many campaigns and knocked on many doors - in the last two local elections, I’m sure I knocked on 1,000 doors in each.
My comment is - focusing so much on the drawing of the maps ignores the fact that true valuable field work - IDing voters, following up effectively - is often not done at all by state coordinated campaigns or local campaigns or mostly by any campaigns, and talking in great detail about “cutting” is fairly irrelevant.
True legit field work should have many workers, thorough and strategic targeting and follow up, and documented data. I’m sure this happens - say, in Iowa for the POTUS primary - but I doubt very strongly that it is anywhere near the norm.
Knocking - as anyone who has done it knows - is incredibly inefficient and unproductive, because small percentages actually answer the door, and fewer still actually take a moment to be persuaded (ditto phone banks). I know it is the most effective form of persuasion, but that notion is usually theoretical and never realized due to lack of the incredible resources needed to accomplish it properly.
Whenever a candidate or the media refer to the “1000s of doors knocked,” I know that is typically false or generally inaccurate. It is just not happening the way the theory says it should.
I would like other’s perspectives on this - maybe my experiences are not representative, but I would be very surprised.
Why cut turf at all? Your canvassers are already using an app. The app can just look at where they are and suggest "next knock" (or maybe "next few knocks" so people trust it). There's at least a decent amount of training data already. A smart algo should be able to handle dynamically allocating a walkable list among an ever-changing group of people, taking into account their transportation mode and local geography. Allow people to decline assignments and the system learns about geography / reachability, too. SMH at all the times I've been asked to cross a busy arterial while canvassing...
I'm very rarely tempted to respond to a Substack but I'll bite on this one! First, my credentials - I've cut turf since 2012, campaign staff since 2016, participated in probably a half-dozen of these centralized turf-cutting operations as staff (running 3 of them) and a half-dozen more as a volunteer.
I will start by saying that I 100% agree that AI turf cutting is the way of the future and someone should get to work on the project of creating a better turf cutter stat. I love cutting turf, but I also hate auditing the bad turf cut by people who aren't good at it, don't want to do it, and whose time would obviously be better spent out talking to people and winning elections. I won't shed a tear when this entire process is automated and I absolutely believe the technology exists with current AI models to produce turf that is good enough, most of the time, to justify any trade-offs.
That being said, I never allowed people to use the VAN auto-cutter as a data director (except in very rare circumstances like post-election ballot cure canvassing), and none of the data directors I know who are currently on races allow it either, and I fully stand by that decision.
I was familiar with the auto-cutter as it existed through the 2024 cycle and there were two major obstacles to using it as a time-saving tool for GOTV turf cutting specifically:
1. The lines didn't touch. I know this sounds like a petty data person nitpick, but it actually created a real issue. The only reason it makes sense in the first place to cut GOTV turf six weeks before Election Day is if you can re-use the same turfs several weeks in a row, refreshing the universe so new doors populate in the turfs you've already cut. As inefficient as manually cutting turf is, you know what's more inefficient? A canvasser walking by the home of a perfectly good target in order to get to a farther-away target because the target just isn't included in a turf at all. To do this, you need every piece of the precinct allocated to exactly one turf, even if the block/street doesn't currently have any targets. That means drawing turf boundaries so the lines touch almost perfectly, with no overlap and no empty space left un-allocated. Not only did the old-school auto-cutter NOT do this, in my experience, it actually did the opposite - the lines were as close to where the current targets were as they possibly could be, with no buffer around the targets at all. That meant manually adjusting every single turf until the lines did touch - if you didn't do that, you'd be stuck re-auto-cutting every single turf, every single week. Which brings me to...
2. It didn't actually save time. The auto-cutter didn't automate the process of running the search, loading the map, hitting "cut", saving the turf to a folder, or generating the list number. It only automated the process of actually drawing the lines, leaving a lot of manual overhead that would have to be done regardless, plus the manual auditing that would be required. Additionally, the old auto-cutter famously had a bug where it would create WAY more "corners" in the shapes than a human would if they cut the same turf. That meant manually adjusting turfs was absolutely excruciating, because to move the boundaries of a turf, you have to move every single corner. If I were to auto-cut 10 packets in a precinct, and I needed to adjust the boundaries of just one of them, I think it would take me more time to do that than it would take to cut all 10 packets by hand myself. That means "let auto-cutter do a first pass and then have a human audit and fix" just isn't a workable solution.
So that's why auto-cutting wasn't the answer for 2024. Now, VAN says they have a new, "smart" auto-cutter that can cut turf that is much closer to how a field organizer would - not crossing highways, keeping streets together, etc. I haven't personally been able to confirm how well it does that, but I haven't heard any indication that they fixed the two issues above, even though any coordinated data director would tell you that those have always been the two main obstacles to making use of the auto-cutter. If they did fix those issues, that's great. Anecdotally, I heard that the roll-out of the new auto-cutter came packaged alongside a ton of user-experience bugs in the turf cutting tool more generally that make cutting turf a worse experience now, even if you choose not to use the auto-cutter, and that doesn't inspire a lot of confidence in how well they did at implementing the new feature.
I think what's so frustrating for campaign data people is that we KNOW the technology is there for a better auto-cutter, we KNOW exactly what features it needs in order to actually be workable and save time, and we're generally tech-savvy people who are down to automate and try new things if it can actually make our lives easier, but as far as I know, NGPVAN isn't interested in hearing our feedback. Fourteen years of using VAN as a volunteer, FO, ROD, and data staffer, using nearly every single feature, in multiple states and on large and small campaigns, and I can't recall a single time anyone from NGPVAN the company reached out to me to ask why I don't use the auto-cutter they're so proud of and what they could do to make it better. None of the data staff I know who are currently on campaigns have been reached out to to test or provide feedback on the new auto-cutter. It seemed like they just assumed they knew what the problems were and how to fix them and plowed ahead without actually talking to the people who would be impacted by these changes, and in the process, made an already imperfect tool noticeably worse.
The reality is that it doesn't actually matter that much how good each turf that the auto-cutter creates is. They could be pretty bad and still represent an improvement when you take into account the time savings from people not cutting turf by hand! For the auto-cutter to be usable, what matters is how easy it is for a person to review and correct the auto-cut turfs. Until that happens, maybe county party chairs and volunteers on small local races will get some use out of it, but it won't be a viable way to cut all the GOTV turf for a state-wide coordinated campaign with multiple federal races as stakeholders.
thank you for commenting Molly super insightful! I had this half formed when I wrote this last night but I was thinking about the 20 person audit team we had on the PA Team in 2024 and what you said here is exactly it “For the auto-cutter to be usable, what matters is how easy it is for a person to review and correct the auto-cut turfs.” That’s the product feature that makes the most sense for what's missing and really interesting. Thank you for sharing!
This is pretty indicative of the discourse on field and GOTV work. I am in my early 60’s and have worked on many many campaigns and knocked on many doors - in the last two local elections, I’m sure I knocked on 1,000 doors in each.
My comment is - focusing so much on the drawing of the maps ignores the fact that true valuable field work - IDing voters, following up effectively - is often not done at all by state coordinated campaigns or local campaigns or mostly by any campaigns, and talking in great detail about “cutting” is fairly irrelevant.
True legit field work should have many workers, thorough and strategic targeting and follow up, and documented data. I’m sure this happens - say, in Iowa for the POTUS primary - but I doubt very strongly that it is anywhere near the norm.
Knocking - as anyone who has done it knows - is incredibly inefficient and unproductive, because small percentages actually answer the door, and fewer still actually take a moment to be persuaded (ditto phone banks). I know it is the most effective form of persuasion, but that notion is usually theoretical and never realized due to lack of the incredible resources needed to accomplish it properly.
Whenever a candidate or the media refer to the “1000s of doors knocked,” I know that is typically false or generally inaccurate. It is just not happening the way the theory says it should.
I would like other’s perspectives on this - maybe my experiences are not representative, but I would be very surprised.
Why cut turf at all? Your canvassers are already using an app. The app can just look at where they are and suggest "next knock" (or maybe "next few knocks" so people trust it). There's at least a decent amount of training data already. A smart algo should be able to handle dynamically allocating a walkable list among an ever-changing group of people, taking into account their transportation mode and local geography. Allow people to decline assignments and the system learns about geography / reachability, too. SMH at all the times I've been asked to cross a busy arterial while canvassing...