Across 200+ Inspectaman audits — most of them Chinese tier-2 supplier floors, increasingly South African mid-tier plants — the team keeps running into the same three reference vendors when buyers describe what they have shortlisted. WearCheck sits at one end. Cognite at the other. Foundry Cortex's pilots land in the gap between them. None of these vendors is wrong. They are bets on three different theories of how a 50-asset plant should run its reliability program.
This piece does the math without picking a winner — then names the operations each is actually the right answer for.
The three theories in one paragraph each.
WearCheck is route-based lab analysis. Sample collection on a schedule — oil, grease, coolant — sent to a lab, scored against trend, reported back to the reliability team. SA-native brand, 50+ years operating, well respected for what it does. The theory: detect chemistry-led failure modes early enough to schedule a planned intervention. It is a detection product, not a prediction one.
Cognite is enterprise OT-IT consolidation — historian unification, contextualised data fabric, embedded AI services, and an integration layer that turns the entire plant's data exhaust into a navigable graph. The theory: at scale, the binding constraint on reliability is data unification across systems that were never designed to talk. Cognite solves that. The reference deployments are Aker BP, Saudi Aramco, AB-InBev global. The bills match.
Foundry Cortex is wireless sensors plus predictive AI, sold per-asset, scoped for plants between 20 and 200 critical assets. The theory: mid-tier SA plants do not need a Cognite-tier data fabric and they are not served by once-a-month lab analysis. They need failure forecasting on the assets that actually break, priced transparently, deployed in 60 days. We have spent two years watching mid-tier plants get quoted out of contention by the enterprise vendors and under-served by the lab-only vendors. This is the missing middle.
The annual cost — at a 50-asset facility.
Holding the operation constant — a 50-asset SA mid-tier plant with twenty preventable failures per year and an average failure cost of R500,000 in unplanned downtime — the all-in annual cost of each vendor looks like this:
| Vendor | Year 1 cost | Sensors | Integration | All-in Year 1 |
|---|---|---|---|---|
| WearCheck (route-based) | R300k | R0 | R0 | R300k |
| Foundry Cortex (Y1) | R400–600k | R135k | R30–75k | R475–675k |
| Tractian | $45/sensor/mo | Included | R150–300k | R1.2–1.8M |
| Cognite | R18–90M | Customer-supplied | R9–36M | R27–135M |
Source: SA_Competitive_Punch_For_Punch.md, internal vendor pricing benchmarks May 2026. Cognite Year 1 band reflects mid-market deployments — full enterprise rollouts run higher.
Cost per failure actually prevented.
The annual cost is not the right number. The number that matters is cost per failure prevented — which forces an honest conversation about true prevention rate. WearCheck detects but does not forecast. Cognite has the architectural surface area to forecast but rarely runs at mid-tier scale because the cost-of-deployment dwarfs the cost-of-prevention. Foundry Cortex sits in the middle: Y1 prevention rate ~60%, Y3 ~75%.
| Vendor | Annual cost | Prevention rate | Failures prevented (of 20) | Cost per failure prevented |
|---|---|---|---|---|
| WearCheck | R300k | 20% (detection only) | 4 | R75k |
| Tractian | R1.5M | 70% | 14 | R107k |
| Foundry Cortex Y1 | R575k | 60% | 12 | R48k |
| Foundry Cortex Y3 | R1.7M | 75% | 15 | R113k |
| Cognite | R30M | 80% | 16 | R1,875k |
Source: SA_Competitive_Punch_For_Punch.md, Year-1 cost-per-failure-prevented calculation against 20 preventable failures/year baseline.
Foundry Cortex's Year 1 number is the best in the SA market — R48k per failure prevented. By Year 3, as sensor density and the model library mature, that number rises to R113k while prevention rate climbs to 75%. Cognite, at mid-tier scale, wastes money: the prevention is real but the cost of getting there is incompatible with a R500k average failure cost. The honest phrasing is "Cognite is correct technically and wrong commercially" for any plant under R30M reliability spend.
When WearCheck is the right answer.
Three conditions. First, fewer than 10 critical rotating assets — at that density, route-based lab sampling is more cost-effective than sensor capex. Second, dominant chemistry-led failure modes — coolant degradation, gear-oil contamination, water ingress — which lab analysis is uniquely good at flagging. Third, an operation that has not yet built reliability discipline and would not act on a real-time alert anyway. WearCheck's monthly cadence forces a meeting; that cadence becomes the program.
Where WearCheck is not the right answer: high-density vibration-driven failure modes (bearings, gearboxes, drives), where lab cadence misses the warning window entirely. The R300k spend buys detection of four of twenty annual failures. The remaining sixteen still cost R8M.
When Cognite is the right answer.
Two conditions, both rare for SA mid-tier. First, R50M+ annual reliability spend across multiple plants — at that scale, the data-unification thesis pays. Second, regulatory environments (offshore oil, petrochem, nuclear) where the audit trail requirements justify enterprise OT-IT architecture. Aker BP, Saudi Aramco, AB-InBev global are the canonical references. None of them look anything like an SA citrus packhouse or a Free State foundry.
For 95% of South African mid-tier plants, evaluating Cognite is a 9-month exercise that ends with a quote that triples the reliability budget. Useful only if the strategic answer is "we are repositioning as an enterprise operation." Otherwise: do not start the conversation.
When Foundry Cortex is the right answer.
Four conditions. First, between 20 and 200 critical assets — below 20 the per-asset model does not amortise; above 200 enterprise vendors start to make architectural sense. Second, an operation that has been quoted out of contention by Cognite or AVEVA and walked away from WearCheck because the prevention rate is too low. Third, a CFO who wants to see public pricing before authorising a procurement exercise. Fourth, a preference for SA-local hosting, ZAR invoicing, and a 60-day fixed-price pilot — not a multi-quarter discovery loop.
The honest framing: Foundry Cortex is engineered for the plants the enterprise vendors do not sell to directly. If WearCheck is detection-only and Cognite is unaffordable-at-mid-tier, the missing middle has been a structural gap in the SA market. We are building for that gap explicitly.
The verdict line.
WearCheck charges R300k a year and prevents 4 of your 20 annual failures. The other 16 still cost you R8 million. Foundry Cortex prevents 12 of 20 in Year 1 — total annual reliability impact drops from R8.3M to R4.58M. That is R3.7 million back in your CFO's pocket every year, verifiable against your general ledger. Cognite gets you to 16 of 20 prevented but the bill consumes the savings. The optimal answer at SA mid-tier is the missing middle. That is the entire reason the team built Foundry Cortex.
The 60-day Foundry Cortex pilot scopes this exercise against your actual asset estate, your actual downtime cost, and your actual failure history. R100,000, fixed price, credits 100% toward Year 1 on conversion. The pricing page has the rest of the detail.


