Territory: demand by region, and why Unknown beats a guessed country — Open Pulse
A demand chart is only as honest as the location data underneath it, and the location data under public signals is far shakier than a heat map suggests. What the research measured about geotags, profile locations and inferred geography, and the design that forced on us: a ranked table rather than a map, deterministic resolution rather than a model, and an Unknown row that is counted rather than hidden.
What this page answers
What does this post cover?
A demand chart is only as honest as the location data underneath it, and the location data under public signals is far shakier than a heat map suggests. What the research measured about geotags, profile locations and inferred geography, and the design that forced on us: a ranked table rather than a map, deterministic resolution rather than a model, and an Unknown row that is counted rather than hidden.
Who is it for?
Sales and RevOps leaders who staff, size and quota regions off a demand chart
Why does Territory show a table instead of a map?
Because roughly two thirds of signals name no location at all, and a map has no honest way to draw that. A choropleth renders missing data as empty countries, which reads as "no demand here" rather than "we could not tell". A ranked table can carry an Unknown row and a coverage line next to the regions it did resolve, so the gap is visible instead of implied.
What does an Unknown row actually mean?
That the signals in it are real and we could not place them. It is not an error state and it is not a filtered-out bucket: the signals are counted, grouped and one click from their source posts. Unknown is pinned to the bottom of the table regardless of size, because it describes our coverage rather than your market.
Why not use a model to infer the location?
Because a model resolves everything to something confident, and confident is the failure mode that costs money here. Research on inferring location from post text found roughly 52.8% accuracy at city level, which is close to a coin flip for the granularity a territory decision actually needs. A deterministic parser that names what it can name and abstains otherwise produces a chart you can act on.
At a glance
- Published
- 2026-09-28
- Written for
- Sales and RevOps leaders who staff, size and quota regions off a demand chart
- Category
- informative
- Reading time
- 9 minutes
- Keywords
- territory demand analysis, sales territory planning, intent data by region, geotagged social data accuracy, location inference
Overview
A territory view is a beautiful thing. One map, demand arranged by region, and suddenly staffing, quota and expansion decisions look obvious.
Which is exactly the problem. A demand chart is only as honest as the location data underneath it, and the location data under public signals is shakier than the heat map suggests. So here is the uncomfortable reading from the research, followed by the two design decisions it forced on us: Open Pulse shows demand as a ranked table rather than a map, resolves every location deterministically, and reports what it cannot resolve as Unknown. A wrong country on a demand chart is worse than an admitted gap, because people make hiring decisions on charts.
The location under the heat map
Start with the raw material. Territory planning in the intent-data era increasingly means overlaying signal data onto territory maps: heat maps of intent density by region, comparisons of intent volume to rep assignments, identification of under-served patches. Fullcast, a territory planning vendor, puts the prerequisite plainly in its own planning guide: clean and accurate data is essential for creating benchmarks and measuring success.
Now look at what "location" means for a public post. Researchers have measured this repeatedly, and the numbers are consistent.
Almost nothing is geotagged. Between 0.35% and 3% of tweets carry geotags, with most studies landing around 1 to 2%. The other 97% and more has no coordinates attached at all.
Geotags are not what they look like. A geotag can indicate where the text was written or a place mentioned in the text, and Middleton et al. found that geotags can sit many kilometres away from where the subject matter actually is. A post about a vendor's outage in Frankfurt, written from a train in Brussels, can pin itself to either.
Profile locations are worse. Hecht, Hong, Suh and Chi, in the first in-depth study of the Twitter profile location field, found that 34% of users did not provide real location information at all: jokes, celebrity names, sarcasm, or nothing. Traditional geographic tools are fooled by every one of them.
Inferring location from text is roughly a coin flip at city level. A convolutional neural network predicting a user's location from a single tweet reached 52.8% accuracy at city level and 92.1% at country level. That paper is from 2017 and methods have improved since, but the gap it exposed is structural: coarse geography is inferable, fine geography mostly is not.
Geotagged posts do not represent the population. Karami et al., working with more than 88,000 users and 170 million tweets, found significant differences between geotagged and non-geotagged users across nearly three quarters of measured features. The people who geotag are not a sample of the people who post.
None of this is anyone's fault. People post about problems, not about their coordinates. But it means every demand-by-region chart built on public signals is standing on an inference layer, and the honest question is what the chart does when the inference fails.
Why a guessed country is worse than an admitted gap
Most tools answer that question by never asking it. They fill every row. The geocoder returns its best guess, the chart fills in, and the map looks complete.
Completeness is reassuring and it is also the most expensive kind of wrong. A territory chart that quietly assigns 40 phantom signals to the DACH region does not just mislead, it hires. Headcount follows the chart. Quota follows headcount. A wrong country compounds into salaries and missed quarters.
An admitted gap does not compound. Unknown sits in the table, visible, counted, inspectable. It tells the reader exactly one true thing: these signals exist, and we could not place them. That is a smaller claim than a guess and a more useful one, because the response to a gap is investigation while the response to a guess is confidence.
This is a general principle, not a quirk of one product. Any system that aggregates uncertain labels into a decision surface has the same choice: propagate the uncertainty, or launder it into certainty. Dashboards launder by default, because empty cells look like bugs and filled cells look like answers. The fix is to treat the gap as a first-class value, counted and shown, rather than as a rendering failure.
How Open Pulse resolves location
In Open Pulse, Territory is the view that answers "where": the same demand data grouped by place, with each row carrying how confident we are in the location. Accounts answers "who": one company that three listeners have touched is one row, joined across all of them rather than scattered between them. Territory is the same data grouped by region, for the people who staff and size regions.
Three decisions sit underneath it.
It is a ranked table, not a map. Roughly two thirds of signals name no location at all, because a Reddit thread is a person rather than a place. A choropleth renders that as an empty world, which reads as "the product found nothing" rather than "we do not know". A table can carry an Unknown row and a coverage line, and on this screen those are the two most honest things on it.
It is parsed deterministically, never by a model. The underlying location is free text: "Toronto, ON", "Remote: EMEA", "London/Hybrid". A model would resolve every one of those to something confident. So this parses what it can name, the same input always produces the same output, and anything unrecognised is filed under Unknown rather than guessed.
Unknown sorts last, whatever its size. Every other row is ranked by signal volume, then by account count. Unknown is pinned to the bottom even when it is the biggest row, because it is a coverage statement rather than a territory, and letting it top the chart would make the screen about our parsing rather than about your market.
A few consequences fall out of this that are worth stating plainly.
The score does not pretend geography is intent. A signal's score is built from seven components, including fit and evidence quality. Location feeds the Territory grouping, not the signal's worth. A high-intent signal with an Unknown location is still a high-intent signal.
Unknown is honest about the source mix. Reddit and Hacker News return full text; LinkedIn, X and the job boards return snippets; Facebook is best effort. Some sources carry reliable location cues and some carry almost none. A deterministic resolver with an explicit Unknown state reflects that mix instead of smoothing it over.
Territory is a Pro feature for a reason. Territory and Accounts sit on Pro and above, alongside the pipeline board, competitor intelligence and score tuning. They are the views for teams making staffing and investment decisions, which is exactly where a guessed country does its damage.
What to do with the Unknowns
An Unknown row is not a dead end; it is an instruction. Three moves, in order of usefulness.
- Read the signals themselves. Every Unknown groups real posts, each one click from its source. A cluster of Unknown signals complaining about the same rival is not a geography problem, it is a competitor-intelligence opportunity. The "where" is uncertain; the "what" is not.
- Ask whether the region is missing or the cues are. If your market is Germany but most of your demand resolves as Unknown, the honest reading is that your sources do not carry location cues for that audience, not that Germany has no demand. That is a search-plan problem, and the search plan is editable: keywords, exclusions and per-source queries can all be changed on a running listener without losing its signals.
- Treat a shrinking Unknown as a quality metric. As you tune the listener and rate signals, better-targeted queries produce posts with better cues. The Unknown share moving down is evidence the plan is improving. The Unknown share staying flat while signal quality rises is evidence the audience simply does not geolocate, which is also information.
What you should not do is what most charts invite: mentally reassign the Unknowns to whichever region would confirm the plan. The whole point of the explicit gap is to make that move visible to yourself.
The chart is a decision surface, so it should read like one
Territory exists for one kind of morning: the one where you decide where the next rep goes, which region gets the budget, where the expansion bet lands. Those decisions deserve a chart that distinguishes what it knows from what it guessed, and refuses to do the second.
Deterministic resolution, confidence on every row, and an Unknown state that is counted rather than hidden. That is the whole design, and it is the direct consequence of what the research says about location data.
If you want to see what your demand looks like when the gaps are shown instead of filled, book a demo and bring your hardest region. The Unknowns are part of the tour.
References
- Serere, Resch and Havas, "Enhanced geocoding precision for location inference of tweet text using spaCy, Nominatim and Google Maps", PLOS ONE, March 2023. Source for the 0.35% to 3% geotag range.
- Hecht, Hong, Suh and Chi, "Tweets from Justin Bieber's Heart: The Dynamics of the Location Field in User Profiles", CHI 2011, Palo Alto Research Center. Source for the 34% figure on profile location fields.
- Huang and Carley, "On Predicting Geolocation of Tweets using Convolutional Neural Networks", arXiv, April 2017. Source for 52.8% city-level and 92.1% country-level accuracy. A 2017 method; newer work exists.
- Karami et al., "Analysis of Geotagging Behavior: Do Geotagged Users Represent the Twitter Population?", ISPRS International Journal of Geo-Information, June 2021. Source for the representativeness finding.
- "Comparing Methods to Collect and Geolocate Tweets in Great Britain", 2018, which collects the 1 to 2% geotag rate and quotes Middleton et al. on geotag inaccuracy.
- Fullcast, "Optimize Sales Territory Planning for Maximum Performance", 2024 (vendor-authored). Cited for the clean-data prerequisite.
- SalesGTM, "How to Measure Territory and Capacity Planning Powered by Intent Data for Field Sales" (vendor-authored). Cited for the category practice of overlaying intent data on territory maps.
Frequently asked questions
Why does Territory show a table instead of a map?
Because roughly two thirds of signals name no location at all, and a map has no honest way to draw that. A choropleth renders missing data as empty countries, which reads as "no demand here" rather than "we could not tell". A ranked table can carry an Unknown row and a coverage line next to the regions it did resolve, so the gap is visible instead of implied.
What does an Unknown row actually mean?
That the signals in it are real and we could not place them. It is not an error state and it is not a filtered-out bucket: the signals are counted, grouped and one click from their source posts. Unknown is pinned to the bottom of the table regardless of size, because it describes our coverage rather than your market.
Why not use a model to infer the location?
Because a model resolves everything to something confident, and confident is the failure mode that costs money here. Research on inferring location from post text found roughly 52.8% accuracy at city level, which is close to a coin flip for the granularity a territory decision actually needs. A deterministic parser that names what it can name and abstains otherwise produces a chart you can act on.
Is a high Unknown share a problem with my listener?
Sometimes, and it is worth diagnosing rather than assuming. If your target region is known but most demand resolves as Unknown, the likelier reading is that your sources do not carry location cues for that audience, which is a search-plan question: keywords, exclusions and per-source queries are all editable on a running listener. A shrinking Unknown share as you tune is evidence the plan is improving.
Which plans include Territory?
Territory and Accounts are available on Pro and above, alongside the pipeline board, competitor intelligence and score tuning. They are grouped there because they are the views teams use for staffing and investment decisions rather than for daily prospecting.
Next
- Start the free trial — Creates a workspace. A card is taken up front and nothing is charged during the trial.
- Pricing — What it costs after the trial.
- All posts — The comparisons and the guides, grouped.