Farm apps in Kenya get pitched with a sign-up number, not a usage number. Zalisha, an agribusiness app built at Kenyatta University's Chandaria Business Innovation and Incubation Center, was working with about 500 farmers when it was profiled by the Kenya innovation agency's writeup on Zalisha. That number describes a pilot cohort. It says nothing about how many of those farmers were still opening the app six months later, and the source does not claim otherwise. This is the gap that most farm-app coverage in Kenya glides past: sign-up numbers get reported, retention almost never does.
The received idea is that a farming app closes the extension gap, the shortage of agronomists able to visit every smallholder plot. That idea is not wrong so much as incomplete. An app can carry information to a phone. It cannot make the information arrive at the moment the farmer needs to act on it, and it cannot survive being wrong, or slow, more than a couple of times before the farmer stops opening it. The strongest evidence Kenya has on this question is not about an app that failed. It is about one that worked, M-Farm, and it tells you exactly why.
The evidence that actually exists is about M-Farm, not about apps in general
M-Farm let farmers text a short code and get back crop prices. The Brookings review of mobile tools helping farmers in Kenya is one of the few pieces of research that measured behaviour change rather than downloads. Before M-Farm, 43 percent of surveyed farmers learned about prices from market buyers and 50 percent from middlemen. After M-Farm reached them, only 6 percent still consulted market buyers and 23 percent still spoke to middlemen. Those are the same categories, measured before and after, on the same survey. That is a real signal, not a marketing number.
What makes the M-Farm case unusually strong as evidence is the price. Ordinary text messages in Kenya cost around 1 KSH at the time. M-Farm's texts cost 10 KSH, ten times more. Farmers paid it anyway. Brookings treats that as a revealed preference test rather than a survey answer: people who did not believe the information was worth having would not have kept paying a tenfold premium for it. That is a stronger form of evidence than a satisfaction score, because it costs the respondent something to answer honestly.
None of this proves M-Farm is still in wide use today, and the source does not make that claim. What it proves is narrower and more useful: when an app answers a specific, recurring, high-stakes question (what will I get for this harvest, right now), farmers pay to keep it. That is the test to hold every other app against, including the newer ones being pitched at Kenyan farmers now.
Radio did not disappear, and that tells you something about timing
One detail in the Brookings research gets skipped in most summaries: a third of farmers kept using the radio after M-Farm arrived. Not because the app failed, but because radio and M-Farm were doing different jobs. Farmers who were still deciding what to plant were comfortable with radio. Farmers who were ready to sell wanted M-Farm, because the price information was current and, crucially, it stayed on the phone. Radio content is gone the moment the broadcast ends. A text message sits in the inbox until the farmer is ready to act on it.
This is the detail that the extension-gap framing misses. An app is not competing with radio or with a middleman as a general information channel. It is competing at a specific decision point, and it wins or loses depending on whether it is the right tool for that exact moment. A pricing app used at planting time is answering a question nobody is asking yet. The same app used at harvest, when the farmer is standing at the market gate deciding whether to accept the first offer, is answering the only question that matters that day. Retention, on this evidence, tracks stage of the crop cycle at least as much as it tracks the app's feature list.
For a passion fruit grower in Migori, cited in the same research, this played out concretely: 97 percent reported they could sell their harvest faster once they had price information at hand, cutting the time fruit sat around losing value. That is a sell-side result, not a planning-side one. An app pitched at planning decisions but abandoned by harvest is not necessarily a bad app. It may simply be aimed at the wrong end of the season.
Zalisha adds a feature list M-Farm never had, and that is a different bet
Zalisha, founded in 2016 and operational from 2017, is a broader attempt than M-Farm ever was. According to its own team, quoted in the Kenya innovation agency profile of Zalisha, it offers agronomy and weather advice tied to the farmer's geolocation, credit-score-based access to financial services, on-farm record keeping, and a direct route to buyers that removes the need to physically scout the market. That is a wider bet than a price SMS. It is also a harder one to keep a farmer opening every week, because more moving parts means more ways for one of them to feel unreliable.
The company's own founder is candid about the gap between the pilot and the market. Ian Cheruiot, Zalisha's creator, says that once you are thrown into the market, the theories learned in class become very difficult to apply. That is not a criticism aimed at the app's features. It is an admission that a pilot cohort of about 500 farmers, working with a startup team still refining the product, behaves nothing like a national rollout where support thins out and each farmer's patience for a glitch is shorter.
There is no published retention figure for Zalisha, and none should be invented here. What the source does establish is a structural risk worth naming: an app that bundles agronomy advice, credit access and a buyer marketplace into one product is making a bet that all three features stay useful at once. If the credit feature disappoints a farmer once, does that farmer stop checking the weather advice too? The Brookings research on M-Farm never had to answer that question, because M-Farm did one thing. Zalisha's breadth is its main selling point and its main retention risk, and only time in the field, not a pilot count, will settle which it turns out to be.
The policy layer decides what these apps are even allowed to promise
Kenya's own Ministry of Agriculture draft policy on agricultural data and digital services sets out a framework meant to coordinate digital tools, data and services across agriculture, livestock, fisheries, irrigation and cooperative development. It does not name M-Farm or Zalisha, and nothing here should be read as if it does. What it signals is that Kenya is trying to build shared data standards rather than leaving each app to define its own. For a grower, that matters less today than it will in two or three years, when apps that can plug into a common data layer may hold an advantage over ones built as closed silos.
The FAO's profile of digital agriculture frames the promise in general terms: integrated digital tools can cut financial and labour costs, support management decisions, and lift the quantity or quality of what a farm produces. That framing is true and also unfalsifiable at the level most apps operate. It says nothing about which specific tool, at which specific decision point, actually earns a second look from a busy farm manager in Uasin Gishu or Kirinyaga. The gap between the FAO's general case for digital agriculture and the Brookings-measured case for one specific SMS tool is the gap this whole article is trying to point at.
What decides adoption in the first place, before retention even becomes a question
A 2020 study on the determinants of digital technology adoption among Kenyan farmers, referenced in Strathmore University's research on digital technology adoption factors, makes a distinction worth sitting with: the earlier Kiarie study looked at how a technology gets used, while this study asked what factors determine whether it gets adopted at all. Those are two different failure points, and conflating them is a common mistake in coverage of farm apps. An app can be technically excellent and still fail at the adoption stage, before anyone gets the chance to judge whether it earns a second month of use. Retention data, where it exists at all in Kenya, sits downstream of an adoption decision that has its own separate set of blockers: literacy, phone type, network cost, trust in where the data goes.
This matters for how a farm manager should read any app's marketing. A claim about features solves an adoption problem: it tells you what the app can do. It says nothing about a retention problem: whether the app keeps doing it reliably enough, at the moments that matter, for a grower to keep coming back. The two studies in the Kenyan literature that actually separate these questions, Kiarie's usage study and the Strathmore adoption study, are treated as distinct for a reason. Anyone pitching an app to you is answering the adoption question. You need to be asking the retention one.
The honest limit in all of this evidence
Brookings is explicit about a weakness in its own M-Farm findings: the research relies on self-reported survey data, and there is some evidence that the price increases farmers reported were somewhat exaggerated. That does not undo the broker-displacement numbers, which are a behavioural shift rather than a price claim, but it should stop anyone from treating M-Farm's results as a clean, independently audited outcome. If the strongest piece of Kenyan evidence on app retention carries that caveat, weaker or newer evidence should carry more, not less, scepticism.
There is no published table anywhere in the sources available here that ranks Kenyan farm apps by retention rate, churn, or months-active. That table does not exist, at least not publicly. A farm manager weighing whether to put a season's planning into a new app is being asked to trust vendor claims that have not been tested the way M-Farm's price-displacement numbers were tested. The honest answer is not to wait for that table to appear. It is to build a rough version of it yourself: ask the vendor what specific decision the app is meant to support, at what point in the season, and whether that decision is one you are currently making badly enough to be worth the switching cost.
What separates a kept app from an abandoned one, on the evidence available
Three things emerge from the material, not as a checklist to tick but as a pattern. First, the app should answer a question the farmer is asking right now, not a general category of question the farmer might ask eventually. M-Farm won at harvest time and lost, in effect, to radio at planting time, because the two moments call for different things. Second, a narrow tool that does one job precisely has an easier retention story than a broad one bundling several. Zalisha's ambition, agronomy advice, credit access, and a buyer marketplace in one app, is a harder thing to keep all working well than a single SMS price feed. Third, willingness to pay is a better signal than willingness to download. The 10 KSH M-Farm charge, against an ordinary 1 KSH text, separated the farmers who valued the service from the farmers who merely tried it once.
None of this is a case against apps. It is a case against treating sign-up as the finish line, when the evidence Kenya actually has says sign-up is closer to the starting gun. A previous piece on this site, comparing farm tech options across apps, sensors, extension and automation, makes a related point from a different angle: each of these tools solves a different part of the information problem, and none of them replaces the others outright. An app answering a market-price question and a sensor network answering a soil-moisture question are not competing products. They are covering different gaps, and a farm manager should judge each against the specific decision it is meant to support, not against a general promise to modernise the farm.
Before you put a season into an app, ask what it replaces
The clearest test the evidence offers is not a feature comparison. It is a question: what did the farmer do before this app existed, and does the app do that thing better, at the moment it matters, than the old method did. For M-Farm, the old method was asking a middleman or a market buyer, both of whom had an incentive to underquote. The app beat that comparison cleanly, and the broker-displacement numbers show it. For a bundled app promising agronomy advice on top of a marketplace, the old method might be an extension officer's occasional visit, a neighbour's opinion, or nothing at all. Whether the app beats that comparison is not yet documented anywhere in the public Kenyan research, and a farm manager should not assume it does simply because the feature list looks complete.
This is also where a sensor network earns its place, not as a replacement for a farm app but as a different kind of instrument answering a different kind of question. A grower deciding whether farm sensors are worth the outlay for their own acreage can work through the arithmetic in a separate piece on when farm sensors pay for themselves, which treats the payback question on its own terms rather than folding it into an app comparison. The two decisions, which app to trust with market information and whether to add ground-truth data from sensors, are not the same decision, and conflating them is another way retention data gets muddied in casual advice.
What this means for the next app pitched to you
When the next app arrives with a demo and a sign-up sheet, the useful question is not whether the features sound comprehensive. It is whether the app is trying to answer one question you actually have, at the point in the season when you have it, and whether it has done that reliably enough for someone, anywhere, to have measured a change in behaviour rather than a change in opinion. M-Farm has that evidence, imperfect as it is. Most apps pitched in Kenya right now do not, and saying so plainly is more useful to a farm manager than pretending the retention data exists when it does not.
NuaSense's own view of where digital tools fit into a Kenyan farm's overall decision-making is laid out in a broader look at smart farming trends in Kenya, which covers mobile advisory platforms alongside sensor-based approaches rather than treating either as a complete answer on its own. The point that carries through both that piece and the evidence reviewed here is the same one: a tool earns a place on your phone by answering a specific question well and often, not by promising to answer every question eventually.
The question worth asking before you sign up
Ask the vendor, plainly, what decision this app is built to support, and at what point in your season you would use it. If the answer is vague, general, or covers everything from planting to sale to credit in one breath, treat that as a retention risk rather than a selling point. If the answer is narrow and specific, like knowing today's price before you drive to market, that is closer to the M-Farm pattern that Kenya's own evidence actually supports. The gap between an app that gets downloaded and an app that gets opened in month six is not a mystery. It is, on the evidence available, mostly about whether the app kept showing up exactly when the farmer needed it, and nowhere else.