Product shape
One app per stage of the plant, a promised integration layer, and agents reading across them.
Own appsAton (PLM), PERITO IBP (planning), TilliT and Lean System (MES, ERP), LOGIA and FidaWare (warehouse), Parttrap ONE (spare parts)
→
WIREIntegration layer for a "digital thread" from design to after-sales; no technical detail published
→
Aura AIAgent team per app: hypotheses, own SQL in a sandbox, evidence, confidence, $ value; weekly run
→
People10 prioritised findings a week; shift summaries; managers rate findings as useful, known or irrelevant
Alongside: implementation services on AVEVA MES, SCADA and PI (endorsed integrator in Finland and Sweden since 2021), PlanetTogether and DELMIA Quintiq planning, and bespoke machine-vision cells. Roima says it does not sell its own vision product.
Aura AI: an agent team that works while you sleep
"Surface hidden risks and opportunities, without a data scientist." Runs over the weekend so managers start Monday with findings.
Discovery board
- Findings from one weekly cycle, each tagged by the role it is for (operations, plant, quality, maintenance, CFO) with a confidence of 88-92%.
- Examples: two fillers at 16% and 21% availability because of under-scheduling, not breakdowns; a filler 31% blocked by a faulty case packer; a case packer whose time between failures collapsed fivefold.
- The CFO panel ranks investments: case packer rebuild $6.5M, rescheduling $1.42M with no capex, reject controls $880K, changeover SOP $520K. The cycle summary claims $9.3M a year recoverable for $101K.
| Host app | Agent team | Example finding (illustrative) |
|---|---|---|
| TilliT MES | Operations (symptom), Maintenance (cause), Quality (impact), CFO (money) | Night-shift changeovers 20% longer than day shift: operator process, not machine |
| Lean System ERP | Plant manager, operations, sales (on-time delivery, revenue at risk), purchasing, inventory, finance | Single-sourced, high-reuse parts |
| Aton PLM | PLM master, BOM engineer, supplier analyst, document steward | "91% of documents skip the review step"; a signed-off change may not have reached production |
| PERITO IBP | Demand, inventory balancing, supply and MRP, master data Q&A | Forecast and MRP risk flags |
Inputs on TilliT: production records, quality checks, downtime events, operator comments, KPI targets, labour and utility costs. Claims: live on existing TilliT data in a day, first findings within a week, no retraining. The LLM provider is not named. Every "See Aura in action" button links nowhere.
TilliT MES: cloud MES built around the operator
Bought from Australia in June 2025. AWS serverless, low-code, "action-oriented and operator-centred".
Operator station and line timeline
- Station view for Filler 1: the current order, start, suspend and complete, and a list of activities such as "volume check start-up", "enter stoppage reason" and "lab samples request" with a printed label.
- Timeline by shift for depalletiser, filler, labeller, packer and palletiser, with run rate and downtime.
- AI features: forms and SOPs built by an LLM, AWS Bedrock "heads-up" notes for operators, shift handover summaries, plain-language analytics.
Digital operations
Each line as a chain of activities: a blocked filler, a pre-start check running, a changeover planned, 12,000 bottles complete.
The pains they sell to
Paper everywhere, knowledge retention, audit risk, quality rework, disengaged operators, poor asset performance, lack of control, changeover issues. Most of these are Minder's pitch too.
The rest of the portfolio
Most revenue comes from here and from AVEVA services, not from AI.
| Product | What it does | Origin | Proof |
|---|---|---|---|
| Lean System | Factory ERP with MES, scheduling, warehouse, paperless production, a "digital twin" add-on (layout plus data, no simulation) | Finnish base | VAK lead time −32%; "100+ factories" |
| PERITO IBP | Demand and AI forecasting, inventory, supply and MRP | Bought 2023 (DK) | Thise Dairy scrap −20%, forecast accuracy to 86% |
| Aton | PLM: product data, BOM, change management | Own | Marel, STERIS Finn-Aqua, Nedap |
| LOGIA, FidaWare | Warehouse management and control | Bought 2019, 2022 | HiFi Klubben 3x packages a day; Jøtul +30% |
| Parttrap ONE | B2B spare-parts ecommerce, PIM, CPQ | Bought 2022 | Cibes Lift online orders 20% → 80% |
| AVEVA services | MES, SCADA, PI design and rollout | Partner since 2021 | Wipak 80 lines, 7 sites; HKScan; Boliden |
| Machine vision | Bespoke inspection cells on third-party cameras | Services | Bolefloor; an unnamed cleanroom maker |
Use cases
By industry, as Roima pages frame them. Values are vendor claims.
| Industry | Problems named | Roima offer | Claim |
|---|---|---|---|
| Food and beverage | Food safety, traceability, demand balance, downtime, multi-site scale | TilliT, AVEVA MES, PERITO, recipe management, vision for fill level and labels | Thise −20% scrap; Carlsberg 80+ sites |
| Discrete and machinery | Demand swings, engineering to after-sales, lead times, spare parts | Lean System, Aton, PERITO, Parttrap ONE | VAK −32% lead time |
| Automotive parts | Paper job sheets, defects, visibility | TilliT | Ibero rejects ~10% → under 2% |
| Forest and paper | Quality, wood grading, emissions | Custom MES, vision, planning | Bolefloor >80% yield |
| Process, chemicals, plastics | Scheduling, SCADA, quality | AVEVA stack, APS | Qualitative |
| Metals and mining | Optimisation, energy, predictive maintenance | AVEVA PI, Maximo, Quintiq | Boliden "substantial savings" |
| Life sciences | Compliance, serialisation, document control | TilliT (21 CFR Part 11), Aton | No metric |
| Logistics and retail | Warehouse efficiency, automation | LOGIA, FidaWare | +30% to 3x |
Proof
71 customer stories; about half are warehouse projects. None involves Aura.
With numbers Fact
- Ibero (Brazil, TilliT): final-inspection rejects from ~10% to under 2%. The case says 75%, 97% and 98% in different places.
- Unnamed kitchen-sink maker (TilliT): raw material −25%, finished goods −10%, output +4%, changeover −15%.
- Thise Dairy (PERITO): scrap −20%.
- Wipak (AVEVA MES template): 80 lines at 7 sites, each rollout under a month.
Customers named Fact
- Carlsberg, Suunto, Boliden, HKScan, Alfa Laval, Marel, Metso, Valmet Automotive, Arla, Refresco, Westinghouse, Schneider Electric (TilliT).
- Most manufacturing cases date from 2021-2024 and give no metric.
Analysis
Fact = from Roima's site, registries or press. View = our assessment.
Company Fact
- Nine acquisitions 2014-2025; Corten Capital majority owner since Aug 2024.
- New CTO (Jan 2026) with a mandate for AI and cloud products; an AI Center of Excellence lead.
- Finnish operating company: revenue €29.6M (2021) to €39.3M (2025), flat ~210 staff.
Moat View
- Real: an installed base of 500+ Nordic manufacturers and distributors; AVEVA sites; data already inside their own apps for Aura to read.
- Thin: Aura is new and unproven; the "unified platform" promised in 2024 has no visible technology; growth is by acquisition.
Weak spots View
- Aura is a weekly batch analyst, not a live agent; it depends on Roima's own data models.
- No plant model, physics, world model or fault-diagnosis product; root cause is SQL correlation.
- The flagship TilliT number contradicts itself.
For Minder View
- Take: findings tagged by role, evidence notebook per finding, a CFO view that ranks fixes by money, a Monday-morning cadence.
- Win on: live conversation, capture in Vietnamese, a plant model, and presence in SEA where Roima has none.
Roima vs Minder
Where the two products start, and where they meet.
| Roima | Minder | |
|---|---|---|
| Starting point | Its own ERP, MES, PLM, planning and warehouse apps, plus AVEVA | First-party event capture from workers and signals; legacy systems as adapters |
| Frontline | TilliT stations: work orders, stoppage reasons, checks, AI notes | Workers log tasks, decisions and outcomes by voice in their own language |
| Agent | Aura: weekly agent team with SQL evidence and $ value | Live agent over the operational graph, cited answers, routines |
| Plant model | Layout plus data ("digital twin" add-on) | Plant model with machines, lines and live properties |
| World model | None | Trained on the plant's decision record |
| Sales motion | "See it, prove it, scale it" pilots | Paid pilot, one or two KPIs on the customer's baseline |
| Geography | Nordics, India delivery, small US and Australia | Vietnam proving ground, Europe next |
Sources
- Product screens: Aura AI, Aura for TilliT MES, TilliT MES (captured 29-30 Oct 2026). No Aura video or screenshot set is published; the board image is from the product page.
- Site: home, our story, Aura for Lean System, Aura for Aton, TilliT architecture
- Cases: Ibero, Thise Dairy, Wipak
- Third party: Tech-Clarity on TilliT (Jul 2025), Asiakastieto financials, AVEVA partner page
Conflicts left visible: group revenue "about €60M" vs "about €50M" (2024); headcount 400+ vs 450+; Ibero defects 75%, 97% or 98% fewer. Aura's launch date and LLM are not published.