Between a company's major IT systems (ERP, MES, CMMS) and the reality of the shop floor — paper, verbal handoffs, undocumented know-how — there's often a gap. That's the gap Cedric Geslin, Senior Account Manager (Poka), a Quebec-based connected-worker platform, set out to address at the Tech For Industry Show.

Poka employs around 200 people worldwide and works mainly with large accounts in process industries — food and beverage, cosmetics — as well as in automotive, with ambitions to also expand into discrete manufacturing. The company claims three differentiators: a complete platform, combining a training component (learning & development) and a daily management system; an open platform, designed to integrate with clients' existing systems (MES, ERP, CMMS) via APIs rather than trying to replace them; and a roadmap heavily oriented toward AI agents, with custom development for clients who request it. Deployment is exclusively SaaS on AWS — a constraint that currently excludes the aerospace and defense sectors.

According to the speaker, most large companies have invested in successive technology priorities — ERP, then MES, then CMMS — without ever really equipping the intermediate layer: the operators themselves, often still dependent on paper, whiteboards, and the verbal transmission of informal know-how ("tribal knowledge"). Yet it's precisely these operators who produce and who step in when something breaks — hence the importance, according to Poka, of dedicating resources and tools to them.

The speaker particularly stresses the risk of knowledge loss tied to mass retirements in the coming years — Germany, for example, is expected to see around 4 million retirements by 2030, according to figures cited. Without documenting this tribal knowledge, companies lose valuable skills, sometimes at a steep price (recalling retired staff, resorting to costly external contractors).

Rather than starting from scratch, Poka digitizes existing content (PDFs, Word documents, videos, photos) and automatically restructures it into detailed, step-by-step work instructions viewable on a tablet. Procedure updates are versioned, with notifications to relevant staff when a change makes new training necessary.

Replacing the Excel spreadsheets traditionally used by HR, Poka offers a per-operator skills dashboard, tracking training paths, certification deadlines, and assigning courses through a feature similar to internal messaging. Access to a machine can be conditioned on prior completion of a training path.

The platform makes it possible to share best practices across sites within the same group, with AI capable of automatically translating into 39 languages: an instruction written in French can be deployed immediately, and translated, at a site in Mexico.

Forms and checklists (for example, for starting up a production line) are digitized, electronically signed and logged — making it possible to know precisely who started a line, when, and with which validated skills.

By scanning a QR code on a broken-down machine, the operator immediately accesses the associated knowledge base. Connected to an existing CMMS, Poka's AI can help quickly find similar past failures and their resolutions.

The speaker frames this proposition within a broader competitiveness challenge: facing countries like China on labor costs, French and European industry cannot win a race to the bottom. Differentiation levers lie elsewhere — consistent quality, speed, flexibility, and specialized know-how (aerospace, luxury, among others) — strengths Poka positions itself to reinforce by capturing and spreading know-how.

The current Poka assistant draws on all of the platform's data (knowledge, incident history, forms, skills) to offer information retrieval, translation, diagnostics and problem analysis — with the constant goal of saving the operator time. Poka's roadmap plans a gradual scale-up: after the current assistant, the company is preparing autonomous AI agents (connectivity via MCP servers, root-cause analysis agents, an agent to find qualified staff for a given line), then a "studio" letting clients build their own custom agents — or have Poka develop them.

Unlike implementing an ERP or MES, which can take several years, Poka claims an initial rollout within 90 days, with ROI typically observed between 8 and 9 months. The deployment method proceeds in waves: one or two use cases are first tested at a small number of plants, then gradually rolled out to more sites once their effectiveness is validated. Some Poka clients deploy the solution across 50 to 60 plants, with gains adding up at the group scale rather than at a single pilot site.

One client that deployed Poka across 12 divisions reports a productivity gain of more than 8%. A second client example details several metrics communicated directly by the company: around 70% time saved on creating and accessing work instructions, 60% time saved on incident resolution, and an estimated 40% gain in new operator training time — with Poka having become, according to this client, the plant's second most important tool after the MES.

Another highlighted feature: the "huddle board," a digital summary dashboard replacing the traditional morning production meeting, where each manager previously had to manually gather their numbers (sometimes taking two hours of preparation). The tool offers an immediate summary view (for example, the number of incidents from the previous day and their resolution status), with full traceability of decisions made and their actual follow-up.

Poka's proposition rests on a simple but rarely systematically implemented idea: invest directly in frontline operators' tools, rather than concentrating digitization efforts solely on management systems. By connecting this layer to existing systems (MES, ERP, CMMS) rather than trying to replace them, and by capturing know-how before it retires, the company advocates a fast, progressive deployment approach designed to generate a measurable ROI in under a year.

Watch the full session, along with the other Tech For Industry Show replays, on our dedicated page.