Guide 04

Canvassing Operations: Territory, Routing, and Measurement

Most canvassing programs that underperform are not failing at the door. The canvassers are fine. What is wrong sits one layer up: turf cut badly, walk lists in an order that wastes an hour a shift, a manager who finds out on Tuesday that Saturday went wrong, and a metric everyone is chasing that can be raised without helping.

This guide is about that layer. How turf gets cut and rebalanced, why route order is worth more than it sounds, what a field director actually does during a shift, how to choose a stack without buying a CRM you will not use, and how to measure a canvass so that the number going up means the program is getting better. The doorstep craft it sits on top of is in the complete guide to door-to-door canvassing.

Cutting turf

A territory is one canvasser's shift. That is the definition to design against, and it immediately rules out the way most people cut turf, which is by equal record counts.

Two hundred records in a dense grid of townhouses and two hundred records spread along county roads are not comparable pieces of work. Cut by estimated walking time first, then look at the record counts and accept that they will be uneven. A program whose turfs all contain the same number of doors has almost certainly cut them wrong.

The second rule is that boundaries should follow things that exist. Streets, rivers, railway lines, precinct edges. A boundary drawn across the middle of a block produces canvassers standing on a sidewalk unsure whether the house in front of them is theirs, and it produces double-knocks where two turfs overlap.

For political programs there is an additional constraint worth accepting even when it makes the cut uglier: keep turf inside precinct lines. Election results arrive by precinct, so field geography that crosses precinct boundaries can never be compared against outcomes afterwards, and you lose the ability to learn anything from the cycle.

Rebalancing, and why it never happens

Turf is cut once, at the start, using assumptions that turn out to be wrong within two weeks. One territory is much slower than expected. Another has an unexpected apartment complex. A third turns out to be half rental. Everyone can see this by week three and almost nobody fixes it.

The reason is friction. If recutting turf means a support ticket, a data vendor, or an evening of rebuilding paper packets, it will not happen, and the program will run a whole season on a cut that was known to be wrong in week two. The operational lesson is that the ability to redraw a boundary in a few minutes is worth more than any amount of sophistication in the original cut, because the original cut is always going to be wrong.

When you do rebalance, move boundaries rather than reassigning people. Canvassers get faster on turf they know: they learn which houses have dogs, which blocks are home in the evening, and where the cut-throughs are. Shuffling people across territories throws that away.

Route order

Within a territory, the order the addresses appear in is worth real money and costs nothing to fix.

An unordered list, or one sorted by name or by ID, sends a canvasser crossing and re-crossing the same streets. A list ordered so they work up one side of a street and down the other removes that entirely. The gain is several attempts per hour, which does not sound like much until it is multiplied by every canvasser and every shift in a season, at which point it is equivalent to an extra person you did not have to recruit, train, or retain.

Ordering also has a second-order effect people miss: it makes coverage legible. When a canvasser works a street in order, a manager can look at the recorded dispositions and tell immediately whether the street was actually walked. Scattered dispositions across a turf are much harder to distinguish from cherry-picking.

Running the shift

A shift has a shape: launch, dispatch, walk, check-in, return, debrief. The two that get cut when things are busy are the launch and the debrief, and they are the two that matter most.

Launch is where the day's goal gets stated, the script gets refreshed, and anyone visibly unready gets caught before they are alone on a street. Debrief is where the data comes back, where the script gets fixed based on what actually happened, and where people decide whether they are coming back. A program that dispatches and collects, with nothing at either end, loses both its data quality and its retention.

During the walk, the field director is watching for three things, and all three are only actionable while the shift is still running. Has anyone stalled, which usually means they are stuck, lost, or having a bad time. Is any turf far off pace, which usually means it was cut wrong. Are dispositions being recorded at all, which is the one that quietly destroys a program if nobody checks.

This is the practical argument for live data over an end-of-night sync. Learning on Tuesday that Saturday went badly is a report. Learning at 4pm that one canvasser has recorded nothing since 2pm is a decision you can still act on.

How many passes, and when

A single pass through a universe reaches a minority of it. That is not a failure of the canvass, it is arithmetic: most people are not home most of the time, and the ones who are home on a Saturday afternoon are a specific and unrepresentative slice of the neighborhood.

Multi-pass planning is therefore not an optimization, it is the difference between covering a universe and covering the people who happen to be retired. The rules that matter:

• Vary the time of day between passes. Knocking the same street at the same hour three times mostly reaches the same households three times and the rest never. Alternate weekday evening and weekend daytime at minimum.
• Carry not-homes forward, drop the completed. The second pass should be mostly the addresses the first pass missed, not a fresh walk of the whole turf. This is the single biggest efficiency gain available after route order, and it is the one paper-based programs cannot do.
• Stop at three attempts for most programs. The fourth attempt at an address that has never answered is worth less than a first attempt somewhere new, right up until a deadline makes coverage of a specific universe the only thing that matters.
• Space them. Two passes in the same week feel to a household like being pestered, and they reach nearly the same people.

The planning consequence is that your shift budget has to be computed against passes, not against universe size. A 12,000-record universe at two and a half passes is 30,000 attempts, and campaigns that budget for 12,000 discover the gap far too late to fix it.

Data discipline

Three rules cover most of it.

Record every attempt, not every contact. Not-homes are data. They tell you which streets need an evening pass, and they are the denominator of your contact rate. A program that only records conversations has a contact rate of 100 percent and no idea how it is doing.

Record at the door. Anything reconstructed at the end of the night is partly invented, and the parts that are invented are the details. This is where offline capture stops being a feature and becomes a data-integrity requirement, because a canvasser in a basement corridor or on a rural road with no signal will otherwise write it down later, or not at all.

Keep the record on the address, not the person walking it. Canvassers and reps turn over constantly. If the callback, the note about the dog, and the fact that the homeowner works nights live in one person's head or on their paper packet, all of it leaves when they do. This is the single largest silent loss in high-turnover programs.

Quality control

Every canvassing program eventually has to answer a question nobody enjoys asking: is the data real? Not because canvassers are dishonest, but because the incentives point one way and the supervision is thin. A person alone on a street, measured on doors, with a long walk to the next house, is in a situation that produces optimistic record-keeping even among people acting in good faith.

The two failure modes look different. Skipping is a canvasser quietly not knocking doors they expect to fail, then recording them as not-home. It feels efficient and it destroys your not-home list, which is the input to the second pass. Fabrication is recording conversations that did not happen. It is rarer, and much worse, because a fabricated ID goes into the universe you build your final week on.

Three checks catch nearly all of it, and none of them require treating anyone as a suspect.

• Look at the geography of the dispositions. A canvasser who walked a street in order leaves an orderly trail. Scattered records across a turf, or a cluster of dispositions recorded from one spot, are worth a conversation.
• Look at the timing. Records entered in a burst at the end of a shift mean they were reconstructed, which at best means the details are approximate.
• Call back a sample. Ring five reported contacts a week and ask whether someone came by. This is standard practice in paid canvassing operations and it is the only check that actually verifies a conversation happened.

The management posture that works is to make the checks routine and visible rather than investigative. Canvassers who know the geography is visible record honestly, and the handful who were skipping stop. Announcing an audit after you already suspect someone is a much worse position to be in.

One caution worth stating: implausibly good numbers deserve the same attention as implausibly bad ones. Programs reliably investigate underperformers and celebrate overperformers, which is exactly the wrong way round if what you care about is whether the file is real.

Choosing a stack

Canvassing tooling splits into three categories that get confused with each other in evaluations.

• Data platforms own the file. Strong on targeting and matching, usually weaker as a phone app, often priced and gated around organizational eligibility.
• Field sales trackers own the rep. Strong on pipeline, quotas, and management reporting, with canvassing as a module rather than the center of the product.
• Field operations tools own the walk. Strong on turf, routing, offline capture, and coverage, and expected to hand data off to whatever holds the customer or voter record.

Most bad purchases come from buying the wrong category rather than the wrong product. A team whose actual problem is that turf takes two days to recut will not be helped by better pipeline automation, however good the demo was.

The practical evaluation questions are narrow: what happens to a disposition entered with no signal, how long does it take a manager to redraw a boundary, when does the manager see the data, and what does a seasonal tripling of headcount cost. Those four separate the field faster than any feature grid.

The metrics that resist gaming

Doors knocked is the default metric and the worst one, because every cheap way to raise it harms the program. Skip the not-home record, re-walk the easy street, cut the conversation short: all three make the number go up.

Use these instead.

• Coverage. Share of the universe attempted at least once. Cannot be gamed by re-walking, because a second attempt at the same address does not increase it.
• Contact rate. Conversations over attempts. Diagnoses list quality and time of day. Collapses if not-homes go unrecorded, which makes it a useful integrity check as well as a performance one.
• Outcome per conversation. IDs, appointments, or signatures divided by conversations. Isolates script and training from throughput.
• Attempts per canvasser hour, by turf type. Never pooled across turf types, because comparing a rural canvasser to a suburban one on this number drives out your rural people.

Working out whether it paid

Cost per door is the wrong denominator. Compute cost per conversation and cost per outcome, and put recruitment and training into the cost, because in a high-turnover program those dominate everything else and are invisible if you only count hours walked.

For commercial teams, attribute closed revenue back to the territory as well as the rep. Reporting only by rep silently blames people for territory quality, rewards whoever was handed the good turf, and gives you no way to tell a good rep on bad turf from a bad rep on good turf. Those two look identical on a leaderboard and require opposite responses.

For political programs the honest answer is that attributing votes to a canvass is hard, and the useful proxy is coverage of the identified universe rather than any attempt at causal credit. Keeping turf inside precinct lines at least leaves the door open to comparing results afterwards.

Common questions

How big should a canvassing territory be?

Size by walking time, not door count. A territory is what one canvasser can attempt in one shift, commonly 60 to 120 doors on suburban streets for a three-hour shift and far fewer rurally. Equal record counts produce wildly unequal shifts.

Does route order actually matter?

Yes, and it is the cheapest available improvement. Ordering a walk list by street removes backtracking, which is worth several attempts an hour across every canvasser and every shift.

What should a field director look at during a shift?

Whether anyone has stalled, whether any turf is far off pace, and whether dispositions are being recorded at all. All three are only actionable while the shift is still running.

How do you measure canvassing ROI?

Cost per conversation and cost per outcome, with recruitment and training included in the cost. For commercial teams, attribute revenue to territory as well as rep so the two can be told apart.

What is the most common canvassing operations mistake?

Managing to doors knocked. It is the easiest number to raise and the easiest to raise in damaging ways. Coverage and outcome per conversation resist gaming.

Where to go next

If this is a campaign, the field-program version of these decisions, with deadlines and volunteer dynamics attached, is in the guide to political canvassing and GOTV. If it is a sales team, the economics and follow-up side are in the complete guide to door-to-door sales.

Beacon is the tool this guide describes: territory management, route optimization, offline capture, and live coverage reporting, built so a boundary can be redrawn in the middle of a shift.