Many Swiss SMEs are investing in AI but see no business impact. The reason is rarely the technology itself, but almost always the lack of a driver tree. Those who derive AI ROI in a structured way using a driver tree measure benefits in CHF rather than in promises.
In my work with growing SMEs, I regularly see this: reports by cost center are accurate, but there’s no visibility at the order level, into lead times, or rework rates. AI models then end up calculating based on noise. In such situations, I first pause, define the value levers, and integrate a few, tightly scoped applications into existing processes.
Regardless of the technology stack, the following applies: First, the business logic; then, minimum data standards; then, small models. Results are open to debate, but liquidity is not.
Key Points at a Glance
- A driver tree directly links revenue, contribution margin, cost of goods sold, lead times, and DSO to EBIT and cash. Without this logic, AI optimizes in the wrong place.
- Best practice: 4–6 weeks for value levers, data, and the measurement framework; then integrate three lean applications into existing processes and measure them weekly in CHF.
- Sample calculation: CHF 20 million in revenue and a DSO of -6 days result in approximately CHF 0.33 million in freed-up liquidity; a -0.8 percentage point decrease in the cost of goods sold ratio, bringing material costs down to CHF 12 million, yields approximately CHF 96,000 in EBIT.
AI ROI via a driver tree: What does this mean for SMEs?
A driver tree is a condensed representation of how your company generates profit and cash. It shows which input variables (e.g., order size, variants, material usage, lead times, payment terms) have what effect on contribution margin, EBIT, and liquidity.
The goal is not to model everything. The goal is to identify the 3–5 levers that drive EBIT and cash flow. This can only be achieved if the driver tree is established first and every AI application serves a clear CHF objective.
From Cost Center Reports to the Value Driver Model
Cost center reports are necessary, but they are usually too general for AI-driven decisions. A value driver model goes deeper and operates at the object and process levels (order, item, batch, ticket):
- Order Size and Pricing Logic: At what order size does the contribution margin tip? Which discounts are worthwhile, and which aren't?
- Variants and Complexity: Which product or service variants result in above-average rework or longer setup times?
- Material and Goods Cost Ratio: How Do Scrap, Express Procurement, and Lot Sizes Affect COGS?
- Processing and Lead Times: Where are orders piling up? Which bottlenecks are driving overtime or outsourcing?
- Days Sales Outstanding (DSO): Which customer segments pay late? Which dunning stage is effective? How accurate is the matching of QR references?
These drivers must be backed by numerical data. The data sources include ERP/MES data, financial accounting (including accounts receivable and accounts payable), bank statements (ISO 20022 CAMT) from the Swiss primary bank, and valid QR invoices. Stability is key: Only stable metrics are suitable as objective functions.
Step-by-Step: Building and Measuring a Driver Tree
Step 1 – Set the Target Function: Define 1–2 measurable targets in CHF, such as a -0.5 percentage point reduction in the cost of goods sold ratio, a -5-day reduction in DSO, or an increase of X CHF in contribution margin per order. Establish the baseline based on the last 3–6 months.
Step 2 – Capture the minimum required data: For each order, you need item prices, material withdrawals, timestamps for each process step, quality codes, and complaints. For DSO: invoice date, payment terms, payment receipt (including QR reference and fees). The data doesn’t have to be perfect—it just needs to be consistent with business logic.
Step 3 – Map out the driver tree: On the whiteboard, start with revenue, then move to Contribution Margin I/II, fixed costs, EBIT, and finally to cash (working capital: inventory, accounts receivable, accounts payable). Assign a formula and an owner to each branch.
Step 4 – Measurement Framework: For each intended AI application, define metrics (e.g., MAPE in sales planning, rework rate, DSO), control groups, and decision rules. Measure the effects weekly in CHF and document your assumptions.
Step 5 – Implementation: Start with three small applications that integrate directly with existing processes. Avoid “big bang” integrations; Excel or a lightweight BI tool is often enough to get started.
One-line ROI calculation: ROI = (Benefits in CHF – Costs in CHF) / Costs. Benefits include, for example, savings on material costs, avoided overtime, reduced write-offs, or relief on a credit line. Costs include internal expenses, external services, and licenses.
Three streamlined AI applications with a direct impact on cash flow and EBIT
1) Sales planning that reduces inventory and express shipping costs: A simple time-series model, supplemented by basic drivers (seasonality, promotions, restrictions), is often sufficient. Measure MAPE and translate improvements into CHF through changes in inventory, availability, and express shipping costs.
2) Risk of rework per order variant: A classification model identifies high-risk combinations of variant, material, and capacity. Rule: High risk is typically associated with senior setup, inspection plans, or changes in lot sizes. You can measure the impact through rework, scrap, and overtime—all of which are directly reflected in COGS.
3) Prioritizing Accounts Receivable for Faster Payment Collection: A payment score takes into account payment history, response to reminders, and dispute rates. This results in call lists, reminder channels, or adjustments to payment terms. DSO decreases, discount utilization increases, and liquidity becomes more predictable. This is based on accurate QR invoices and reliable matching in accounting. If capacity is lacking in this area, accurate accounts receivable management through the accounting department is often the most effective lever.
All three applications operate in close alignment with existing processes: purchasing and scheduling meetings, production approval, and accounts receivable. The decision rules are the responsibility of the business unit; the model merely provides priorities.
Governance: Responsibilities, Data Protection, Reporting Frequency
As a member of the Board of Directors, I ensure that responsibilities are clearly defined: The CFO is responsible for the driver tree, data definition, and measurement in CHF; the business units own the decision rules; and IT ensures availability and security. External parties may assist—but ownership of data and models remains internal.
Data protection requirements under Swiss law must be observed. Whenever possible, use pseudonymized data relevant to the process. Bank data (CAMT files) and customer account data must be given special protection; access is granted on a role-based basis and logged.
You acquire licenses incrementally based on measurable impact. Each milestone is a gate: The next phase of expansion only proceeds if the documented financial impact is satisfactory. If there is a lack of internal CFO capacity, a “CFO as a Service” can temporarily help establish the driver tree, measurement framework, and reporting.
Conclusion: Don’t start with AI—start with the drivers. Stick to a 4–6-week timeline, define value levers, data, and measurement criteria, launch three small applications, and report weekly in CHF. This is how AI evolves from an experiment into a tool for driving EBIT and cash flow.
Frequently Asked Questions
As an SME, how do I build a driver tree for AI?
Start with the target in CHF (EBIT, cash) and work backward to the operational drivers (order, variant, time, material, DSO). Outline formulas, responsible parties, and measurement points. Use existing ERP/financial accounting data and fill in any missing timestamps in a pragmatic way.
What data do I need to measure AI ROI?
You need accurate order and position data, material withdrawals, process timestamps, quality codes, and customer data with QR references and payment receipts. In addition, bank statements (CAMT) from your Swiss bank help ensure robust matching. Perfection isn’t necessary—consistency and stability are.
How do I calculate the ROI of an AI project in CHF?
ROI = (Benefits – Costs) / Costs. Quantify benefits in terms of lower COGS, less overtime, fewer write-offs, lower inventory levels, or faster payment collection. Convert each metric to CHF, document your assumptions, and use control groups to rule out spurious effects.
If you'd like to implement this approach in your company, I'd be happy to schedule an initial consultation: tf-fs.com/contact/.