Seed Production Software Problems: The 7 That Cost Real Money, and How to Avoid Them

Why does lot genealogy get scoped as a bill of materials?

Ask a development team to model traceability and they will reach for the pattern they know: components go in, a product comes out, and the genealogy is a tree. That model is correct for discrete manufacturing and it will fail at your first blend.

Seed conditioning is a graph, not a tree. A lot splits into size fractions on a screen. Two of those fractions recombine months later with material from a different parent lot, produced by a different grower on a different field, and sometimes from a different crop year. Then the result is rebagged into a different unit of measure. Standard lot traceability handles the merge poorly and handles the year boundary worse, because it assumes a product is assembled once and then sold.

The demo case is always simple: a single lot cleaned and bagged, where the tree looks fine and the awkward cases get described as edge cases. They are not edge cases. Blending and rebagging is routine plant work, and it is precisely where the value leaks.

The fix is to make the developer draw it before you sign. Ask them to draw the lot genealogy on a whiteboard, then ask them to draw a blend of carryover with new production, then a rebag. If they reach for a parent and child tree or a bill of materials structure, they have modelled discrete assembly and your first November blend will break it. The lot object needs parents rather than a parent, a class, a production field and grower, and an accumulated test history that travels with it.

What goes wrong when you bring legacy lot history across?

Every seed operation has history worth carrying: carryover inventory that will be sold next season, lots that may be questioned by a customer or a licensor, and the certification records that support both. Almost none of it exists in a form that maps cleanly.

The genealogy is the problem. In most operations the chain lives in a conditioning plant notebook, a certification file cabinet and one production manager's memory. Carryover is where it matters most, because a unit sitting in the warehouse right now needs a defensible parentage, a class, a current germination and a royalty status, and if any of those is reconstructed from recollection you have loaded an assertion rather than a fact.

The fix is to draw a hard line. Convert current inventory and anything with live commercial or regulatory exposure at full genealogy, and record everything older as legacy with its original documents attached rather than pretending it has a modelled ancestry. Then start the project immediately after harvest, so the genealogy model is exercised against a real conditioning and treating season before the next production year begins.

Why do scale, plant equipment and accounting integrations break after launch?

Three interfaces matter in a conditioning plant and all three drift. Scales and cleaners produce the weights that become conditioning yields. The accounting system holds grower contracts, settlements and log of purchases. And the certifying agency has its own forms and submission expectations.

The failures are quiet ones. A scale is recalibrated or replaced and the output format changes slightly, so weights keep arriving and are subtly wrong. A cleaner is swapped during a shutdown and the new controller reports in different units, so a conditioning yield that should be a loss reads as a gain and nobody questions it for a season.

The specific risk in seed is that these errors do not look like errors. Everything downstream is a weight, and a weight always looks plausible. So the fix is arithmetic rather than alerts alone. Reconcile mass across every conditioning operation: what went in, what came out as product, what came out as screenings and dust, and what is unaccounted for. A mass balance that does not close within a tolerance you set is an exception someone reviews before the lot moves on. Validate the shape of every inbound file, not just its values, and keep the raw capture so a corrected mapping can reprocess the period.

What happens when test expiry and label age rules are not covered?

This is the compliance gap that turns good inventory into unsaleable inventory, and it is routinely deferred to a second phase. Labelling limits how old a germination test may be, and the permitted age varies by state and for interstate shipment. A lot that is perfectly fine to sell in one market may already be stale in another, and the system usually finds out at the moment somebody tries to ship it.

The consequence is a forced retest under time pressure, during the weeks when the laboratory is busiest and your customer is waiting. Sometimes it is worse than a delay: germination declines in storage, so a retest run late can return a figure that no longer supports the sale at all, and carryover that was carried at full value on the books turns out to be worth substantially less.

The fix is to hold the label age rule as data per destination and warn weeks before a lot becomes unlabelable, not after. Schedule retests from that rule rather than from a person remembering. Carry treatment records in the same place, since treated seed brings handling and disposition restrictions that untreated seed does not. And build carryover valuation on the same data, so declining germination, retest cost and class risk are visible in the numbers your finance people use.

Should you build custom or configure what you already own?

If you produce a modest volume of public varieties, sell in bulk within one state, carry no licensed varieties and never blend across years, do not build. Your identity chain is short enough that a disciplined spreadsheet and a good tag file are honestly sufficient, and the capital belongs in cleaning capacity.

Before commissioning anything, look hard at what you already run. Grain accounting software handles bushels, bins, positions and prices well, and if your operation is mostly bulk with a small certified programme on the side, extending your existing grain system with a lot register and a tag file may be a fraction of the cost of a build. If you already run a manufacturing ERP (Enterprise Resource Planning) for another part of the business, its lot traceability will handle straightforward cleaning and bagging perfectly well; it is only the merge and the crop year boundary where it fails, so a narrow extension around blending may be all you need.

Build when two or more of these are true. You carry licensed varieties and accrue royalties. You blend or rebag. You produce across more than one certification class. You sell into more than one state and therefore live under more than one set of labelling rules. Or you have had a lot questioned by a customer, a licensor or a certifying agency, and reconstructing its history took more than a day.

How do hidden costs get into the quote?

A first release covering contracted production, harvest lot creation and full conditioning genealogy including blends and rebagging runs $65,000 to $140,000 across 12 to 16 weeks in our delivery experience. A full platform adding test scheduling with label age rules, certification agency reporting, treatment records, royalty accrual, grower settlement and carryover valuation runs $170,000 to $380,000 across 6 to 11 months. Four things push a quote beyond it.

Crop count is the largest, because a corn operation, a soybean operation and a small grains operation have genuinely different unit models and different conditioning steps, so three crops is closer to three builds than to one with options. Certifying agency count is the second, since each has its own forms and its own rules and a producer reporting to three agencies carries three reporting paths. Licensed variety count and the variety of royalty triggers is the third, because some agreements accrue at conditioning, some at bagging and some at sale, and carryover has to avoid accruing twice. The fourth is plant equipment integration if you want conditioning yields read from scales and cleaners rather than recorded by hand.

What holds the number down is doing one crop and one plant first, and starting immediately after harvest so the model meets a real season early.

What separates a build that works from one that fails here?

Enforcement. The builds that work make the system compute and refuse. The builds that fail present a dropdown and trust the operator.

Class is the clearest example. When two lots are blended, the result can only claim the lowest class of its inputs. In a system that works, that is derived and locked. In a system that fails, the class is a field somebody sets, and on a busy November afternoon it gets set to whatever the order needs. Same with unit of measure: bulk pounds at intake, units of a set kernel count for corn, bags for soybean, seed boxes and totes for delivery that come back and get refilled. Every conversion is a place where inventory drifts away from reality. The system should own the conversions rather than asking a person to do them.

The other separator is that a working build makes harvest lots attach to a production contract rather than appearing from nowhere. Acres by grower, by variety, by class, with the contract terms and settlement basis, so conditioning yields flow back into settlement and you can see what a lot actually produced after cleaning losses. Conditioning yield is a real cost driver that most operations estimate rather than measure.

When you interview a developer, ask how class is derived on a blended lot, and reject any answer involving a user selection. Ask how they will handle test expiry rules that differ by destination state. Then settle ownership in writing before kickoff: repository, cloud accounts and the unrestricted right to bring in another firm. At Digital Heroes the client owns the code from the first commit. There is no packaged product to fall back on in this category, so an inaccessible system is an inaccessible production history, and that history is what makes your inventory worth seed price instead of grain price.