Agriculture: from a signal in the data to a working MVP
I found the next industry vertical hiding in our own signup data, designed the evidence to test it, and built the prototype that shows what it would be. [One line on the outcome of the decision gate]
Company: SafetyCulture
My role: [Confirm title]. I did the research, set the direction for UX and UI, and built the prototype myself.
Team: [Who you worked with: Data & Growth, Product, Engineering, other designers]
Timeline: [Start to finish, 2026]
Scope: Opportunity case, research design, and a working prototype for an Agriculture industry module
The case at a glance
17x less revenue than manufacturing. From the same front door. Agriculture ARR $2.50M against manufacturing $41.2M (FY26 close, AUD). [verify against the source sheet]
#1 industry people pick at signup. 6,417 free orgs self-identified as agriculture, against 6,366 for manufacturing, to June 2026. [verify]
757 active agriculture orgs mined for demand. Topic-modelled into four clusters of need, two well covered and two exposed. [verify]
[Outcome] What happened at the decision gate. [Add the result: approved, deferred, what it unlocked, with the date]
The situation and the problem
The situation. SafetyCulture is heading toward industry verticals: specialised modules that sit on top of the platform and tune what it already does for one industry. A vertical is bought by that industry, so each one has to earn its place with evidence that a specific industry wants it and will pay for it. Three verticals were already underway. The open question was which industry should be next, and what the first version of it should contain. [Edit to match how the brief actually reached you]
The problem. More businesses tell us they are in agriculture at signup than any other industry. Almost none become revenue: 336 paying accounts, 5% of that signup base, against 3,992 and 63% for manufacturing. The data pointed to productisation, not appetite. Manufacturing has a module, playbooks and a named go-to-market motion. Agriculture has none of them. The signup and revenue figures come from different systems, so I read the gap as a direction to investigate, not a funnel to optimise.
Three moves from a hunch to a bet I could defend
1. Signal: The disconnect. Same front door as manufacturing, a fraction of the revenue. The managed agribusinesses already paying (GROWMARK, AGnVET, Muirs, Delta Agribusiness) averaged about $22.5K, and they found us by accident.
2. Revealed demand: What they already build. Topic modelling of the template, action and issue names across 757 active agriculture orgs (24 categories) produced four needs: packhouse and produce quality, machinery and maintenance, chemical and biosecurity, seasonal workforce.
3. Priority: The map, the gate, one scheme pack. I prioritised the two exposed needs where no equivalent exists on the platform, and a certification pack, because certification is how this buyer procures. Livestock records were deliberately left out.
[Add the real story of how you got here: who you talked to, what surprised you, what you ruled out]
Agriculture isn't a weak segment. It's a mispriced one.
The reframe. A farm turning over millions can run on a handful of people, and its workforce triples at harvest and collapses after. Seats are the wrong unit of value. Sites, packhouses and depots are the unit of risk, compliance and audit, so the module is priced per property. That also made agriculture a live test of the per-location module model.
What would prove this was worth building, and what would kill it
Question 1: Is the demand real? Revealed from usage, not assumed: 757 orgs' own templates and issues. The method I chose could not be argued with in a room, because the customers had already written the answer.
Question 2: Who is the customer? Not hobby farms. 77% of agriculture accounts are self-serve at about $2.9K. The module targets managed, multi-site agribusiness, the $22.5K cohort, with four named design-partner candidates.
Question 3: What stops us? Willingness to pay is unproven until the end-of-year readout, so the ask was built as one pod, one 90-day cycle and a decision gate: stop with learnings, not sunk platform cost.
[Add the research you designed beyond the data: interviews, partner sessions, a survey, who ran them, sample sizes, what you learned. Not in the deck, so not claimed here.]
Zero to MVP: setting the direction and doing the work
Directing: strategy, research and design direction. Framed the opportunity and made the case for the fourth vertical. Chose the research method and the questions that could kill the idea. Set the design principles for the prototype: reuse before invention, compliance counted rather than scored, obligations that follow from tags. [Who you directed on UX and UI, and how: reviews, critiques, briefs]
Hands-on: the prototype, built by me. Built the working prototype end to end: a group console for the safety and compliance manager and a phone app for the operator. [Confirm tools and what you wrote yourself] Traced every screen to the platform pattern it reuses, so the build argument is visible on the screen. Modelled the real standards in the product: Freshcare FSQ5, HARPS and the LPA biosecurity plan. [Anything else you personally built or decided]
What it shows, screen by screen
The prototype follows one fictional customer, Callowa Ag Group: nine sites, a packhouse, depots and farms. [Confirm Callowa is a demo customer] Two people use it: Kate, who runs safety and compliance for the group, and Danny, an operator at a depot.
[Image: group dashboard] Readiness is the state you are in. Four counts, then audit readiness against the three standards that matter to this buyer, then a compliance benchmark by site. The auditor's binder is generated when needed, not assembled in a panic. Notice: the certificates expiring in 12 days sit in the same strip as the good news, so nothing is hidden by averaging.
[Image: site overview map] The map is the mental model. A property map with pins, blocks and a record of every change, because the standard asks for a kept-current, marked-up plan. Each pin carries the records that belong to it. Notice: the AI headline is labelled as an exemplar, not live inference, and the pin legend is the standard's own list (FSQ5 9.02 to 9.05).
[Image: gate log] Every entry is a record. Each gate entry captures who and what came on, with the visitor's signed acknowledgement and clean-down status, linked back to the property's biosecurity plan. Notice: expand a row and the signed acknowledgement, the declared items and the movement record are all in one place.
[Images: three phone screens] Notice: one big action, scan a QR to start a check, then only what is due this morning. Five plain-language report types, and a line saying it works offline. The QR gate sign doubles as the mandated biosecurity signage.
[Image: benchmarking] Compare like with like. Sites are rated on operational measures within a site tag, farm against farm, and compared with the same season last year. Compliance sits underneath, counted rather than scored. Notice: the group score sits beside a bar for every site, so the sites that slipped are visible without a click.
[Image: sites] Tags describe the site, obligations follow. One site object. A farm carries a biosecurity plan and a gate, a depot carries fleet and plant duties, a packhouse carries the scheme audits. Notice: the plan status column only says "Not required" where the tags make it so.
[Image: audits and certification] The daily work is the evidence. Readiness per standard, the audit rhythm, evidence that is about to expire, and non-conformances that live in Incidents with the clock running. Notice: the non-conformance ladder shows which band the site is in and what that does to the next audit.
[Image: standards library] The scheme, as a working product. The official documents, the working templates pushed to the customer, and the records they produce, held in three shelves per standard. When a scheme revises, the customer adopts the new pack. Notice: the US pack (SQF, WPS, FSMA) is shown as coming, to signal the same engine with different standards.
Three calls that shaped the product
Decision 1: Reuse before invention. Choice: every screen starts from an existing platform pattern and says which one. Why: the pitch was that most of this already exists, so the prototype had to show the reuse, not just claim it. Trade-off: it constrains the ambition of any one screen. [Add the one you most wanted to break]
Decision 2: Compliance is counted, never scored. Choice: mandated items are pass or fail and are excluded from the star rating. Why: averaging them would hide the one thing that fails an audit. Trade-off: a less flattering single number for the executive. [Add how it landed]
Decision 3: One object, obligations follow tags. Choice: a single Sites object whose tags decide what a site must do. Why: mixed enterprises are farm, depot and packhouse at once, and the module should not make them configure three products. Trade-off: [Add what you gave up]
An honest scorecard
A prototype earns trust by saying what is real. Labels below come from the prototype's own notes. [Confirm each]
Working: Group dashboard and audit readiness by standard
Working: Benchmarking list with operational criteria and exclusions
Working: Sites object with tags and plan status
Working: Gate sign-in, gate log and movement log
Working: Operator phone app: checks, report it, gate
Working: Internal audit running as a native inspection
Working: Incidents with classification clocks
Working: Standards library with pack adoption
Placeholder: Aerial base map. Placeholder imagery, a spec rather than a built engine
Exemplar: The AI headline is an example of the surface, not live inference
One pod, ninety days, a gate
Property map: built on the Spatial Canvas engine, skinned for property and paddock. Shows in the prototype as the site overview map, two views, with change history.
Gate check-in: built on Construction's Site Access module, extended with biosecurity capture. Shows as visitor sign-in, gate log and movement log.
Freshcare and HARPS audit pack: built on Inspections plus clause-mapped templates and a binder view. Shows as Audits & certification, the standards library and internal audit.
Fast-follows: chemical registers, seasonal inductions, Lone Worker. [Confirm which appear in the prototype] Livestock records deliberately out.
Read this before the numbers. The signup, active, paying and revenue figures come from different systems, so each pair should be read on its own, not as one funnel. Willingness to pay for a vertical is unproven until the end-of-year readout, which is why the ask was gated on it. [Add the result: what the gate decided, what has been built since, and what you learned]
Four things I would carry into the next vertical
1. Start from revealed demand. What customers already build in the product beats what they say in a survey.
2. Find the unit of value. The pricing and the product shape were the same problem. Seats were wrong; properties were right.
3. Build the prototype to argue. A working screen answers objections a slide can only anticipate.
4. Say what is not real. Labelling placeholders and exemplars made the real parts more believable.
[Rewrite these in your own words]
Next case study: [Next case study title]