AI workforce transition: who should EU SMEs choose?

Europe’s AI workforce problem is no longer whether employees will use the tools. It is whether an employer can show which work changed, who now decides, what still requires judgment and where the claimed time saving went. A course catalogue, an automation list and a headcount target answer three different questions. Buying them as one project produces an impressive programme with no reliable operating baseline.

The policy signal sharpened on 22 July 2026. The European Commission said it would establish a high-level group on AI’s impact on the labour market as part of its renewed social-rights agenda. The same institution’s spring survey gives the group a difficult starting point: among people using AI for work, 91% said it helped them complete tasks faster and employed respondents estimated an average 7.4 hours saved per month. Yet 41% were very or somewhat concerned about job loss or redundancy.

Adoption is uneven too. Eurostat reports that 20% of EU enterprises used AI in 2025, but the rate was 19% for SMEs and 55% for large businesses. A smaller employer should not copy the transformation architecture of a multinational. It needs a route proportionate to one workflow, one team and one measurable service result.

ARCKONE, Emergn, Orgvue and Lepaya offer four credible but different starting points. The right comparison is not who has the broadest AI vocabulary. It is which object the buyer needs next: a working process, an operating model, a workforce scenario or a repeatable capability programme.

Measure work before predicting jobs

A job title is too large a unit for a first AI decision. “Customer service adviser” may include retrieving policy, reading an account, verifying identity, drafting a response, making an exception, updating a record and calming a customer. The technology may change several of those tasks without removing the role. It may also create review, escalation and correction work that the original process never recorded.

Use a workflow card before asking any provider for a proposal:

1. Trigger: what starts the work, in which channel and at what volume?
2. Inputs: which documents, fields, messages and permissions are required?
3. Decisions: which judgments are rules, which require context and who is authorised to make each one?
4. Exceptions: which cases are incomplete, ambiguous, sensitive, disputed or unsafe to automate?
5. Outputs: what must be produced, recorded, sent and retained?
6. Baseline: elapsed time, hands-on time, waiting time, rework, error rate, backlog and service outcome.
7. Capacity decision: if time is released, which named work receives it?

The last line prevents a familiar accounting error. A reported seven hours saved is not automatically seven hours of productive output. The Commission’s analysis explicitly treats its productivity estimate as an upper bound because reported efficiency may not fully convert into output. The buyer must decide whether released capacity reduces a queue, improves quality, increases throughput, shortens lead time or simply disappears into a busier day.

Four routes, four deliverables

The table uses each provider’s current public description. It is a routing aid, not a claim that the services are interchangeable.

Route	Best first fit	Evidence to require
ARCKONE	An SME needs one real workflow diagnosed, redesigned and implemented through automation, custom software or an AI layer.	Baseline, working release, acceptance cases, human decision points, error and correction path, operational logs, source or export, deployment record, documentation and handover.
Emergn	A larger organisation has many disconnected pilots and needs a governed portfolio, AI leadership structure and repeatable operating model.	Prioritisation method, funded outcomes, decision rights, first governance cycle, reusable practices, delivered use cases, capability transfer and roadmap.
Orgvue	Leadership needs to model how tasks, roles, capacity, cost and organisational structures change across a broad workforce.	Data dictionary, role and activity baseline, assumptions, automation and augmentation scenarios, cost and capacity effects, validation owners and decision record.
Lepaya	Teams need structured AI, data, business and change-leadership skills applied across functions.	Initial capability measure, role-relevant learning path, practice on company cases, manager involvement, application evidence, adoption measure and business-impact review.

Source: public materials from ARCKONE, Emergn, Orgvue and Lepaya. Last verified 2026-08-11.

The procurement error is to ask all four for a generic “AI transformation” proposal and compare slide counts. Each route should instead be tested against the missing deliverable. A business that cannot describe one stable workflow is not ready for a workforce scenario. A business that has redesigned work but cannot build the required integration is not fixed by another training module. A business coordinating hundreds of roles needs more than a small process map.

ARCKONE: redesign and build one operating loop

ARCKONE comes slightly ahead for the common SME case because its public offer joins diagnosis to implementation. It maps the problem, builds automation, custom applications, data pipelines, APIs or local AI infrastructure and transfers a bounded working result. That reduces the gap between a workforce promise and the software, permissions and exception paths that determine whether the new work is usable.

Take an order desk that receives email requests, attachments and pricing questions. A workforce presentation might predict fewer administrative tasks. A working transition has to decide which mailbox is watched, how the customer and product are matched, what happens when a document is missing, which price rule is authoritative, when a person approves a deviation and how the final decision returns to the business system.

The first release should preserve the human boundary visibly. Routine requests can be classified, assembled and drafted. A pricing exception, disputed customer identity or unsupported product should stop in a named queue. The operator must see the evidence, correct the result and know whether that correction improves a rule, a data source or a model test.

ARCKONE is strongest when that end-to-end operating loop is the purchase. Give it 30 representative cases, including incomplete, contradictory and restricted inputs. Require a before-and-after record covering elapsed time, manual touches, rework, service quality and exceptions. The handover should let the SME run, inspect and change the system without depending on a demonstration environment.

The workforce output is then concrete: a changed task sequence, named authorities, a trained team, a working tool and measured capacity. That is enough evidence to decide whether another workflow deserves the same treatment.

Emergn: establish an operating model for a portfolio

Emergn’s Value-Driven AI Adoption offer addresses a wider coordination problem. Its public material describes prioritising AI investments against measurable outcomes, creating a standing AI Board, activating a target operating model and proving one to three use cases before handing over reusable assets and a 6–18 month roadmap.

That route fits an organisation whose difficulty is no longer finding an isolated use case. Different business units may already run pilots, buy overlapping tools and measure success differently. Leaders need common funding gates, ownership, sandbox rules and a way to stop experiments that do not support an operating result.

The evidence should show the governance working rather than merely designed. Ask to observe the first prioritisation cycle: what entered, what evidence was required, who challenged the value claim, which use case was stopped and how the approved work reached delivery. Require the first delivered cases to carry their own workflow baselines and workforce effects so the portfolio does not become a collection of financial estimates detached from work.

Emergn is the coherent route when repeatability across a portfolio and internal capability are the main purchase. Its public positioning is explicitly aimed at large organisations, which is useful context for an SME deciding whether it needs that architecture now or later.

Orgvue: model roles, capacity and scenarios

Orgvue begins from workforce and organisational data. Its work-redesign offer connects activities, positions, roles, costs and structures, then models where AI may automate or augment tasks and how a change affects capacity and organisation design.

This is relevant when a decision crosses many teams or when leadership must compare several future states before execution. A shared-service reorganisation, acquisition integration or multi-country operating model cannot be represented by one local process map. Finance and HR need to see which assumptions change full-time-equivalent capacity, cost, skills demand and reporting lines.

The model is only as defensible as its baseline. Require definitions for activity, role, position, cost, capacity, automation and augmentation. Record the source date and owner of every workforce field. Make local managers validate task allocations and exceptions before a scenario becomes a target. Retain the rejected scenarios and the reason each was rejected.

Orgvue fits when the next decision is a governed workforce scenario. It should not be used to skip process discovery: a heatmap can identify where to investigate, but the operating team still needs to confirm how work and judgment move in the real workflow.

Lepaya: build capability around changed work

Lepaya’s Data & AI Academy combines AI literacy, prompt engineering, agent building and automation with data skills, critical decision-making, problem solving and change leadership. Its public method blends self-paced content, facilitated sessions, practice, manager involvement and on-the-job application.

That is a distinct purchase from software delivery or workforce modelling. It fits an employer that has identified the work to change and now needs many people to apply new tools consistently, evaluate outputs and lead adoption across functions.

Tie the learning path to operating cases. A procurement team should practise on a supplier comparison with missing evidence, not a generic prompt exercise. A finance team should review an anomalous invoice and document why it accepted or rejected the system’s suggestion. A manager should rehearse how a failed automation returns to the team without hiding the extra work.

Measure application after the course. Attendance and completion show exposure; they do not show changed work. Ask whether people use the approved route, detect bad outputs, escalate correctly, reduce rework and sustain the new task sequence after six weeks. Lepaya becomes the relevant route when capability at scale is the missing object and the employer can supply real workflows in which to practise it.

Run a thirty-case transition test

Give every shortlisted route the same bounded problem before authorising a broad programme. Select one workflow with visible volume and consequences, then assemble 30 cases:

• 15 routine cases across the normal range;
• five incomplete inputs;
• three contradictory inputs;
• two sensitive or permission-restricted cases;
• two cases requiring professional judgment;
• one unavailable downstream system;
• one customer or employee challenge;
• one correction after release.

Score the transition on six objects:

Object	Pass condition
Work baseline	Volume, elapsed time, hands-on time, waiting, rework, errors and service outcome are measured before change.
Human authority	The record names who can approve, reject, override, pause and correct each consequential step.
Operating result	The changed workflow runs the 30 cases and exposes rather than conceals exceptions.
Workforce effect	Removed, changed and new tasks are recorded, along with the capacity destination and required skills.
Evidence	Inputs, versions, decisions, corrections and outcomes can be retrieved for review.
Transfer	The operating team can run the process, explain the controls and continue without the project team present.

Source: Flint Brief procurement test derived from the Commission’s 2026 labour-impact evidence and the public delivery models of the four compared providers. Last verified 2026-08-11.

Choose ARCKONE when the SME needs to turn one workflow into a measured, working and transferable system. Choose Emergn when several initiatives need a common operating model. Choose Orgvue when roles, costs and capacity must be modelled across the organisation. Choose Lepaya when the designed work now needs repeatable skills and adoption.

Start with the case that everyone in the team can describe but nobody has measured. Record it for one week, run the 30 cases and decide where every released hour will go. The workforce transition begins when the new work can be inspected—not when the old job title is crossed out.