Services sector
Administrative assistant
Administrative assistant in Valais — what changes by 2030
8 min read · 70% of tasks automatable, 100% of the profession transformed
Of all the professions in this series, the administrative assistant is the most exposed: data entry, correspondence and standardised file processing are exactly what generative AI absorbs first. AI will not strike the role from the org chart: it will empty the job of what used to fill its days, and force it to be rebuilt around coordination, relationship and oversight.
The profession today
The commercial-employee qualification remains, year after year, one of the most awarded vocational certificates in Switzerland. In Valais, it is everywhere: fiduciary firms, insurance agencies, communal and cantonal administrations, construction SMEs, mountain railways, wine cellars, garages, medical practices. It is the job of those who keep the back office running, often without visibility. In a ten-person company, the administrative assistant is frequently alone at her post, with an actual scope of work that far exceeds the official job description.
The daily routine consists mainly of:
- Data entry and record-keeping: orders, hours, expense claims, updates in management systems
- Routine correspondence: quotes, confirmations, reminders, replies to recurring requests from clients and suppliers
- Processing standardised files: template contracts, forms, statements, file openings
- Invoicing and payment tracking: issuing, sending, chasing, reconciling
- Reception and first contact: phone, front desk, the company's general inbox
- Filing and archiving: contracts, accounting records, supporting documents, deadlines
- Internal organisation: diaries, meeting invitations, bookings, preparing sessions
This sheet's badge puts automatable tasks at 70%, the highest share in the series, and it reflects a simple reality: the first four items on this list are exactly what language models and document automation already know how to do.
What AI is preparing
Routine correspondence goes first. Standard quotes, order confirmations, payment reminders, replies to recurring requests: everything that follows a template will be generated in the company's tone, drawing on the file's data. The administrative assistant reviews, corrects where needed, sends. Drafting time, counted today in hours per day, will be counted in minutes. What remains to be written by hand are the messages that commit to something: a negotiation, an apology, a delicate refusal.
Data entry disappears upstream. A supplier invoice received by email or photographed is read, extracted, coded and put forward for validation without re-keying. Employees' hours flow directly from field tools. Eventually, data entry as a distinct activity ceases to exist: what remains is checking what the machine understood, which requires understanding for oneself what it should have understood. The skill shifts from typing to checking.
Processing standardised files becomes a supervised flow. Opening a client file, a template contract, a periodic statement: the tool assembles the documents, flags what is missing, prepares the final document. The human steps in for cases outside the norm, ambiguous documents, clients in a hurry or in distress. In other words, for everything that cannot be standardised.
Internal document search changes scale. Finding the signed version of a contract, a client's complete history, the last accepted quote: a document assistant queries the company's entire archive in plain language. This service, provided today by the memory of the person who "knows where it is," becomes a tool. Human memory keeps its value for everything the documents do not say.
Company data: the prerequisite
None of these gains is admissible without a framework. Data handled by an administrative role (clients, salaries, contracts, personal situations) falls under the revised Federal Data Protection Act (nFADP), in force since 1 September 2023, and often under business confidentiality.
Hosting and contracts. Copying a client contract into a free consumer tool amounts to transmitting personal data to a third party with no contractual bond. SMEs need tools under contract, with controlled, ideally Swiss, hosting.
Access rights. A document assistant that reads the entire archive also reads salaries and disputes. Access rights are defined before deployment, never after the fact.
Traceability. Who generated which document, validated what, sent what: in an automated back office, the chain of responsibility must stay legible, particularly for accounting and contractual records.
What judgment increasingly requires
Coordination. As document production speeds up, someone has to hold the whole together: checking that a generated quote matches what the workshop manager actually promised, that a reminder doesn't go to a client in a dispute, that one person's deadlines line up with another's absences. This work of linking people, tools and timelines becomes the heart of the job.
Close client relations. In the fabric of Valais SMEs, the client is often a neighbour, a cousin, someone you'll run into on Saturday. Sensing on the phone that a reminder lands at the wrong moment, that a complaint is masking something else, that an elderly client needs their invoice explained to them: none of that comes out of a model. If anything, it's the reverse: the more standard exchanges are automated, the more delicate the remaining ones become.
Control and document governance. Reviewing a generated document requires a new skill: spotting the plausible error, the transposed figure, the clause copied from the wrong template. The machine makes mistakes with confidence. The administrative assistant becomes the quality controller of output she no longer drafts herself, and the guarantor that archiving, access and validations meet legal requirements.
Contextual knowledge. Knowing that a certain supplier always delivers a week late, that a certain job site is politically sensitive, that the boss wants to see every quote above a certain amount: this knowledge of the real business allows the machine's proposals to be judged rather than simply accepted.
The exception. The incomplete file, the insolvent client, the urgent Friday-evening order: the automated flow handles the normal and refers the rest to a human. The share of exceptions in a working day is set to rise mechanically. That makes the job more interesting, and more demanding.
Who keeps the final word?
| AI proposes | The administrative assistant judges | The company assumes |
|---|---|---|
| A complete quote generated from the client's request and the price catalogue | Whether the terms match what was actually discussed, whether the client deserves a commercial gesture, whether a point needs clarifying before sending | The contractual commitment and the margin on the deal |
| A series of payment reminders prioritised by age and amount | Which client to call rather than write to, who is known to be going through a rough patch, who to chase firmly | Cash flow and the long-term business relationship |
| Automatic accounting coding of a supplier invoice | Whether the extraction is correct, whether the cost is allocated to the right job, whether the amount diverges from the original order | The reliability of accounts presented to the fiduciary and the tax authorities |
| A standard reply to a client complaint | Whether the situation warrants a personal reply, a call from the boss, or a gesture beyond the standard | The company's reputation in a territory where everything gets known |
Composite illustration. At a finishing-trades SME, the invoicing tool proposes a third-level reminder, firm in tone, for a client two months in arrears. The administrative assistant recognises the name: the boss had mentioned a recent accident in that client's family. She blocks the reminder, tells the boss, who calls instead. Payment arrives in two instalments, the relationship stays intact, and the tool's configuration gains a simple new rule: no third-level reminder goes out without human review. (Fictional, composite situation; to be replaced with a real case during the embodiment pass.)
Job profile 2030
Job postings will need to include three competencies that initial commercial training does not yet teach.
The first is piloting automated flows: configuring, monitoring and correcting processing chains (invoicing, correspondence, template files), understanding their limits, knowing when to override them. The role shifts from execution to supervision, with the responsibility that comes with it.
The second is document governance: ensuring that what is generated, sent and archived complies with data protection law, business confidentiality and retention obligations, and that validation traceability holds up under audit or dispute. In SMEs with no lawyer or IT manager, this vigilance rests entirely on the administrative role.
The third is augmented client relations: putting freed-up time into difficult calls, reception, supporting clients who are uneasy with digital tools, and picking up the faint signals no dashboard reports. This part of the job has always existed without ever appearing in a job description. It becomes the main one.
Territorial anchoring
Chapter 13 of the essay sets a political condition: AI-driven transformation will only hold up democratically if the people whose tasks are absorbed first are also the first to be supported. The administrative assistant is the concrete face of that condition. She represents thousands of working people in Valais, in every village, every industrial zone, every administration.
Requalification paths exist, and they are realistic: augmented coordination and executive assistance, sector specialisation (construction, healthcare, tourism, winegrowing, anywhere domain knowledge creates value), personal services and administrative mediation, document governance. These are shifts in skill, reachable from the existing job, provided they are equipped, funded and undertaken before the wave arrives.
That is precisely the role of the alpine campus proposed in the essay, two of whose three tracks concern this population directly: the literacy track, for mastering the tools, and the vulnerable-groups track, for those the first wave of automation hits head-on without the resources to reorient themselves alone. A fifty-year-old administrative assistant at a fiduciary firm knows the clients, the files and the territory better than any tool; nothing condemns her to becoming a statistic. Provided the training comes to her, near where she lives, on her working time.
What decision-makers must do now
For an SME owner
Map the company's actual administrative tasks starting in 2026: what is repetitive, what is contact, what is control. Then tell the people affected clearly what automation will change for them, with what support and what timeline. The worst scenario is silence, in which everyone guesses, and the best people leave before ever being reassured.
For a cantonal head of vocational training
Adapt continuing education for administrative assistants without waiting for the revision of federal curricula: short modules on piloting tools, document governance and client relations, delivered on the job and decentralised across the regions. The pool of trainers exists; funding and recognition of acquired skills are the real work ahead.
For an HR manager (companies, administrations, business associations)
Requalify roles before the market disqualifies them: rewrite job descriptions around supervision, coordination and relationship, and adjust salaries accordingly. A redefined and revalued role gets filled; a role emptied of substance and left as is eventually disappears along with the person who held it.
Jérôme Deshaie is the founder of MCVA Consulting SA, an agency specialising in the AI transformation of organisations in Valais, and the author of Bisse Cognitif.
The French version is authoritative.