Network
Expander
The stores you do not serve yet, the ones you already do, and the ones you are about to lose, on one ranked list. Network Expander builds the most complete picture of the market from internal and alternative data: store sales, store census, mobility data, geographic profile and socioeconomic factors. Three commercial plays run on that picture, each with its own models. One grows the number of stores you serve. One works out what each store you already serve is worth. One tells you which of them is about to go quiet.
You cannot see the whole market, rank it, or hold on to it
Your commercial strategy fails twice in the traditional channel, and this is the first failure. Plans get built on the stores you already sell to, because that is the only list that exists. It shows up in three ways, and Network Expander runs one play against each: one that grows the number of stores you serve, one that grows what each store you already serve is worth, and one that tells you which of them you are about to lose.
The stores nobody serves do not appear anywhere
A customer master lists who buys from you. It cannot list who would, so the white space never enters a plan and never gets a target. Independent stores do not report sell-out, and a panel reports the category rather than the store, so nothing from outside fills the gap either. What is left is a market you only know through the part of it you already sell to, and a share of your highest-potential stores sit on no route at all, with nothing in your reporting to say they are there. New territory gets found the way it always has: one rep at a time, on foot, in whatever order the streets happen to run.
The costGrowth that was always available, never planned for.
The stores you do serve are not ranked
Two stores order the same amount from you this month. One is already at its ceiling. The other could buy three times more, and nothing tells you which is which, because your records show what a store bought and never what it is capable of buying. So visit frequency follows convenience and habit, and the store with room to grow gets the same attention as the one that is already maxed out. The headroom sitting inside the customers you already have is usually the fastest revenue on the table, and it is the part nobody has a number for.
The costRevenue sitting inside the customers you already have.
The stores you are losing do not announce it
A store does not tell you it is leaving. It orders a little less, then a little less again, and by the time the drop is large enough to show up in a monthly report the business has usually already gone somewhere else. Your own records will not tell you whether a store is still buying at all: when a campaign finally goes out on that customer list, 30 to 50% of the stores on it have closed or already belong to a competitor. So retention becomes a reaction to a number that arrived too late, and the effort goes into accounts that were never coming back.
The costEffort spent on customers you have already lost, and revenue walking quietly out of the ones you still have.
It is not one or the other. More stores on the route, more out of every store already on it, and the ones about to leave flagged before they go.
Proprietary models
Not a platform we configured for your market. Models we built inside it, trained on what has actually happened across tens of thousands of stores in this channel, then calibrated to your business, your categories and your commercial priorities.
A report can only show you what is already in your records. Every store in it is a store you already sell to, and every number in it is a sale that already happened. That is the ceiling on a report, and it is exactly why the stores you are missing never appear in one.
A model works the other way round. We take what has actually happened across tens of thousands of stores in this channel, find the patterns that hold, and use them to answer a question about a store nobody has records on. A store that size, on that corner, with that foot traffic, those neighbours and that income around it, buys about this much of this category. That number exists nowhere. It is calculated.
The same machinery reads forward as well as sideways. The rhythm of what a store has already bought, how recently, how often and how much, is what says whether it is about to stop. So the same models that size an opportunity are the ones that see it slipping away.
And only at the stores you already have.
And leaves the thinking to you, usually on a Friday.
And hands your team the answer: this store, this much, this week.
Nobody on your team needs to be a specialist.
What arrives is a list, in order, with a number next to every store.
Pick the play. The model builds the mission.
Three commercial intents, each with its own models underneath. The intent is the decision you make: grow the base, grow the stores you already have, or keep them. The model is what does the work inside it, and each one answers a question most consumer goods companies cannot answer today.
Horizontal growth
Grow your customer base, profitably.
Which new stores, in which zones, will move the most volume?
One model runs this play, against the part of the market you have never sold to.
Scores every store in the territory you do not sell to by estimated billing potential, so the size of the opportunity is known before anyone drives to it.
We cross your portfolio against external sources, and AI entity resolution matches them into one record per store. What comes back is every active store in the territory, and the white space nobody is serving yet.
The complete universeEvery prospect is scored by estimated billing potential, so the size of the opportunity is known before anyone drives to it.
Qualified leadsUnderserved zones and channels are surfaced as lists, heat maps and routes that are ready to hand to the field.
Immediate executionA carbonated beverage company. The universe was built from data rather than street by street, prioritised by potential and activated in the field.
Vertical growth
Serve every store for what it is worth, not for what it sells today.
Which existing customers are buying below their potential?
Four models, four ways to rank the same universe. Which one runs depends on the play you are making, and it changes the moment your priorities do.
The largest absolute gaps between what a store buys today and the ceiling the model puts on it.
Headroom measured against each store’s own size, so a small shop buying far below what it should ranks above a large one already close to its ceiling.
Ranks the map by capturable volume, then takes the stores inside the densest cells, so a route earns more without covering more ground.
Highest foot traffic first, tie-broken by capturable volume, for when visibility in front of shoppers is the point.
The DNA of each store: geographic and socioeconomic profile, consumption patterns, foot traffic, dwell time, shopper preferences and sales potential.
A segmented universeWe maximise the probability of purchase by category or strategic focus, and the return on where you place people, investment and point-of-sale material.
Directed investmentVisit frequency and dedicated rep assignment follow the real value of the store, so the ones with room to grow get the attention and the ones already at their ceiling stop absorbing it.
Efficient coverageThe same play across Latin America, pulled through three different commercial levers: a targeted brand push, a price strategy, and prioritisation by potential.
Same engine, different levers. Brand, price and coverage all pull on the same ranking underneath, which is what makes it an engine rather than a campaign.
Results from Kin engagements across Latin America. Figures are per market, and what any one operation can expect depends on its own baseline, category and coverage.
Retention
Keep the stores you have already won.
Which stores are at risk of churning?
After four weeks without buying, half of those stores never buy again. That is the window this model works inside, and it is far shorter than the reporting cycle most teams find out on.
Scores every store you serve on how likely it is to stop buying in the next 30, 60 or 90 days, and how long you have before it does.
Every store’s purchase history is read for three things: how recently it bought, how often, and how much. A store drifting on any of them surfaces long before the drop is big enough to notice in a monthly report.
Stores at riskNot every loss looks the same. A store can stop dead, keep ordering but buy less each time, or buy the same amount further and further apart. The pattern tells your team what is actually going wrong.
The pattern behind itThe risk score is crossed with route, territory and segment, so a rep opens the week knowing which stores to visit now and which ones can wait.
A visit list, in time to actHow often the model was right about which stores were about to stop buying, checked against two years of purchase history in one market.
Measured on two years of purchase history in a single market, where the model also called 81.6% of every store that stopped buying before it went. What any one operation can expect depends on how much purchase history it can give the model to read.
Not every store is worth the stop.
Coverage and profitable coverage are not the same number. The score that tells you which stores are worth more than they buy today also tells you which stops cost more to serve than they return, and which ones are slipping away while nobody is looking. Visit frequency follows what a store is worth rather than how convenient it is to reach, so your team holds coverage without adding a rep, and growth is not bought at a loss one store at a time.
Why the alternatives stop short
Three things teams reach for when they need a picture of the market. Each of them gets part of the way.
A census samples what a team can physically walk.
Months of fieldwork, street by street, and it decays with every store that opens, closes or changes hands while it is still being collected. Network Expander builds the same picture from data in about four weeks, and it keeps updating from what your own team confirms on every visit.
It is the starting point, not the answer.
Your own records are the most valuable input here, and the reason the scoring reflects your business rather than a generic market. What they cannot contain is the part you have never sold to. What they will not volunteer is which of the stores you do serve is already slipping away. The external layers add the first. The models read out the second.
A map decays. A model improves.
A study is at its best the day it is delivered and worth less every week after. Because every visit your team logs confirms or corrects a record, this runs the other way: the universe gets more precise the longer it is in use.
That closes the first failure. The second one is still open.
A ranked universe does not move product on its own. Engine 02 turns it into what each rep does tomorrow, and reports back what actually happened.
Questions commercial teams ask
How do you build the store universe without a foot census?
By reconciling sources rather than walking streets. That is the first step of the horizontal growth play: your sales history and customer master are crossed against census records, mobility, geographic and socioeconomic data, and AI entity resolution matches them into one record per store, not three near-duplicates with different spellings. That produces a complete, ranked universe in about four weeks, where a census takes months and is out of date before it is finished.
What is the difference between the three plays?
They answer different commercial questions, so you pick the one that matches what you are trying to do. Horizontal growth finds the stores you do not serve yet and ranks them by potential. Vertical growth works out what each store you already serve could be buying, and gives you four different models to rank that headroom. Retention tells you which of those stores is about to stop buying. Most teams run all three, because winning a store, growing it and keeping it are three different jobs.
Why does vertical growth have four models?
Because where is my headroom has more than one right answer, and the right one depends on what you are trying to do. Hidden Share ranks by the biggest absolute gaps, which suits a volume push. Category Fit ranks by the gap relative to each store’s own size, which surfaces small shops performing far below what they should. High-Potential Clusters ranks geographically, so a route earns more without covering more ground. Traffic-Led ranks by footfall, for when visibility in front of shoppers is the point. Same universe underneath, four ways to point it, and you can change which one runs whenever your priorities change.
How far ahead does the churn model see?
It scores every store you serve on how likely it is to stop buying within the next 30, 60 and 90 days, and estimates how long you have before it does. That matters because the window is short: after four weeks without buying, half of those stores never buy again. A monthly report tells you afterwards. This tells you while the account is still recoverable, and it also tells you which kind of loss it is, because a store that stops dead needs a different conversation from one that is quietly ordering less each time.
How current does the universe stay?
It updates continuously. External layers refresh, and every visit your field team logs confirms or corrects a record, so the scoring sharpens with use instead of decaying. That is the difference between a model and a map.
What does potential scoring actually tell me?
What each store could buy by product category, against what it buys today. The gap between those two numbers is the revenue available without adding a single new store, and it is usually the fastest thing to act on. It also tells you which stops cost more to serve than they return.
Does this work with the sales app my team already uses?
Yes, and nothing gets replaced. Your app records what happened on the visit. Our models decide which stores are worth visiting and in what order, and the ranked universe can feed straight into the SFA or CRM your team already opens every morning.
How soon can my team act on it?
The universe arrives ranked and ready for the field, so the first routes can be drawn from it immediately. Commercial results follow the first execution cycle, because the model only pays once the field acts on it. We agree the success metrics with you before we start and report against them.
Start the week
with an answer
A 30 minute call, then an audit of one territory, so you can put a number on the opportunity you are not working yet.
Talk with an expert
Tell us which territory you are least sure about.
We will build a sample of your full store universe, including the stores no rep visits, rank it by potential and flag the ones already going quiet.