AI in Business: 5 Applications for Swiss SMEs

AI for SMEs: Beyond ChatGPT

Your team may already use ChatGPT to draft text or support research. AI can also assist with individual steps in business processes. Its usefulness depends on the task, available data, existing systems, and review effort.

Customer histories, quote archives, and scheduling records can provide context for a trial. Use only data approved for that purpose, with access restricted to the people and systems that need it.

Here are five possible starting points for Swiss SMEs. Implementation time and returns need to be assessed for each business. The calculation below is explicitly hypothetical.

1. Automated Quote Generation

Writing quotes involves repeated steps: finding relevant items, checking current prices, and drafting text. AI can prepare a draft, while staff remain responsible for checking the work.

A possible approach is to retrieve relevant past quotes and approved catalogue data for a new request. Fixed validation rules must check access permissions, catalogue items, allowed quantities, and prices. Calculate totals and tax from the verified data. A responsible person reviews the draft and approves prices and the quote before external delivery. This pseudocode illustrates only the drafting step. The helper functions, authorization checks, and business validation still need implementation and testing for a real application.

interface QuoteRequest {
  customerType: 'residential' | 'commercial';
  serviceCategory: string;
  squareMeters: number;
  urgency: 'standard' | 'express';
  location: string;
}
 
async function generateQuote(request: QuoteRequest): Promise<Quote> {
  // Pull similar past quotes from the database
  const historicalQuotes = await db.quotes.findSimilar({
    category: request.serviceCategory,
    sizeRange: [request.squareMeters * 0.8, request.squareMeters * 1.2],
    limit: 10,
  });
 
  // Suggest a draft, with prices requiring verification before approval
  const pricing = await aiEngine.estimatePricing({
    historical: historicalQuotes,
    currentMaterialPrices: await fetchSwissPrices(),
    travelDistance: await calculateDistance('Zürich', request.location),
    urgencyMultiplier: request.urgency === 'express' ? 1.25 : 1.0,
  });
 
  return {
    proposedPositions: pricing.lineItems, // Not yet validated
    validUntil: addDays(new Date(), 30),
    status: 'draft', // No price approval or external delivery
  };
}

Retrieving past quotes as context does not automatically train the model. No fixed number of quotes establishes accuracy, and a model's confidence score is not a substitute for review or price approval. Evaluate representative and difficult requests against expert-reviewed test cases. OpenAI's evaluation guidance recommends task-specific tests and calibration against human judgments.

Measure preparation, review, and correction time against the existing workflow before claiming a saving.

2. Customer Data Analysis and Pattern Recognition

A CRM may contain useful patterns in order history, customer groups, or seasonal demand. Analysis can suggest relationships for staff to investigate.

Practical applications for Swiss SMEs:

• Churn indicators: A longer order interval may justify checking whether contact would be useful. It does not prove that a customer is about to leave
• Segment discovery: Compare customer groups using actual orders and margins, then check whether a proposed offer fits their needs
• Seasonal planning: Compare historical demand and test assumptions before relying on a workload forecast

A correlation does not establish why customers bought something or whether a campaign caused a change. Use an appropriate baseline before attributing retained customers or extra revenue to the analysis. The examples here describe possible uses, not a documented customer case.

3. Document Classification and Archiving

Invoices, contracts, permits, and correspondence need an accessible filing system. AI-assisted processing can combine OCR, or text recognition, with classification.

A possible workflow extracts text from a document and suggests a type, relevant dates, amounts, counterparties, and a project assignment. These suggestions need checking before they trigger further actions or affect records.

Accuracy depends on scan quality, language, handwriting, layout, and the system used. Test your own documents and measure errors and correction time. A general accuracy percentage would not establish performance on your documents.

Search can also use extracted information to suggest relevant documents. Check that results point to the right originals and respect existing access permissions.

4. Intelligent Scheduling

Field scheduling needs to account for employee skills, customer preferences, travel time, equipment availability, and working-time constraints. Check what your current planning software already supports.

A planning system can use these constraints to suggest schedules and routes. Generative AI is not required for every optimization task. Compare proposals with actual journeys and completed jobs before relying on them.

Useful checks include:

• Whether the proposed route actually reduces travel time
• Whether buffers cover realistic delays
• Whether staff skills, preferences, and availability are respected
• Whether there is demand and working time to use any additional capacity

Less travel time does not automatically create extra revenue. Measure the operational effect and review proposed changes before confirming appointments externally.

5. Automated Follow-ups

Rules or AI can prepare contact suggestions using approved signals. An email opening or a website visit does not establish purchase intent, and a missing signal does not establish disinterest. This pseudocode illustrates a proposal only. A responsible person reviews the suggested contact and channel before any external message or call.

async function evaluateFollowUp(customerId: string): Promise<FollowUpAction> {
  const signals = await gatherCustomerSignals(customerId);
 
  // A model score is a proposal, not verified purchase intent
  const engagement = await aiEngine.scoreEngagement({
    quoteViewed: signals.quoteOpenCount,       // Did they open the PDF?
    websiteVisits: signals.recentPageViews,     // Are they browsing our site?
    emailInteraction: signals.emailOpenRate,     // Do they read our emails?
    lastContact: signals.daysSinceLastContact,
    competitorActivity: signals.competitorMentions,
  });
 
  if (engagement.score > 0.7) {
    return { action: 'call', priority: 'high', reason: 'Contact suggestion for review' };
  }
  if (engagement.score > 0.4) {
    return { action: 'email', template: 'soft-reminder', delay: 0 };
  }
  return { action: 'wait', reassessIn: 48 }; // Do not pester cold leads
}

The score thresholds above are illustrative assumptions. Test whether the proposals help compared with the existing workflow. Measure qualified replies, completed sales, and unwanted contact.

A higher close rate is not established without an appropriate comparison. The example does not send messages or approve prices.

Hypothetical Calculation: Valuing Time Saved

This is a hypothetical calculation, not a measured customer result or a forecast. Assume an internal value of CHF 80 per hour. Also assume that the following monthly time savings are achieved after review and corrections, without counting the same saved time twice:

This calculated capacity value is not a cash saving and includes no extra revenue. Account for implementation, operating costs, maintenance, and whether the released time can actually be used.

Without measured net effects and known costs, a payback period or return on investment cannot be promised. Replace the assumptions with your own trial results before making an investment decision.

Conclusion

AI can assist with individual routine tasks. A business needs to test with its own data whether it saves time or reduces errors. Responsible staff retain technical judgment, price approval, and approval for external communications.

Start with one process, measure the results including review effort, and use that evidence to decide whether to expand.

If you want to evaluate which AI applications fit your business, get in touch to discuss the process, available data, and a useful first trial.