Digitalisation in phases that fit your capacity: a scorecard, three gates, and a board-ready roadmap

Start with a process score, not a technology. In a company without a dedicated IT team, the digitalisation plan that survives is the one the team can absorb — capacity first, then ambition. The method is to map where data is re-keyed between the legacy ERP and Excel, score each candidate process on four factors, then run the winners through three gates: quick wins with zero integration, controlled ERP/Excel integrations, and supervised AI only after the first two are stable. We’re Niro Digital, and this is written from shipped client work, not from a content calendar. The framework is our professional judgement, not an independently validated benchmark.

Follow one real order — not a summary — from intake through production scheduling to invoicing. Mark every point where someone re-types a value that already exists in another system: an order number from email into Excel, a quantity from the customer PDF into the ERP, a delivery date from the ERP into the scheduling sheet. You don’t need software or budget for this; it takes an afternoon.

The 15-year-old ERP stays. Replacement is out of scope, because the reader who has been burned by big IT projects doesn’t need another migration project — they need the data flow around the ERP fixed. When the map shows the bottleneck is re-entry around a legacy ERP, a custom CRM is one candidate for the controlled-integration gate later, not a first move.

The scorecard is a planning heuristic, not an externally validated benchmark. Score each process from 1 to 5 on four factors:

• Frequency: 1 yearly, 2 quarterly, 3 monthly, 4 weekly, 5 daily.
• Rule-based potential: 1 every case needs human judgement, 3 mixed rules and judgement, 5 fully deterministic.
• Data readiness: 1 locked in paper or PDFs, 3 mostly structured but re-keyed, 5 already digital and clean.
• Absorption capacity: 1 no slack and no owner, 3 one capable owner part-time, 5 dedicated owner plus IT support.

Add the four scores. A 20 is the maximum.

The worked row scores daily order-intake data entry as high on frequency and fully rule-based, but only middling on data readiness because the values are re-keyed, and low on absorption because the work is shared with no single owner. Your rows will differ; write them one at a time.

Place every process in one of three gates, in order, by this rule: Gate 1 applies only where the workflow can be changed without an ERP or other system integration. Gate 2 applies where deterministic data exchange with the ERP, Excel, or another system is required. Gate 3 applies only where the task requires model judgement after Gates 1 and 2 have met their exit conditions. The total score ranks priority within the applicable gate, not the gate itself.

Gate 1 — quick wins, zero integration. High-frequency, rule-based processes with no system changes: document generation, spreadsheet validations, approval routing. Exit condition: stable for four to six weeks with a named owner.

Gate 2 — controlled ERP/Excel integrations. Connect the legacy ERP to the surrounding data flow with deterministic mappings and manual exception handling. Exit condition: stable for two to three months with no silent data loss.

Gate 3 — supervised AI. Only after Gates 1 and 2 are stable. A supervised agent system is one where model output is treated as a claim that must pass checks before it counts as done; a verification gate is a deterministic check that blocks bad output. We build these as durable workflows — steps that keep their state through restarts and partial failures — and every automation ships with budget ceilings and human approval gates for irreversible actions. Our published doctrine is: “We never trust model output blindly. Every automation ships with deterministic verification checks, budget ceilings, and human approval gates for irreversible actions.” This is what the AI automation service looks like in practice; it starts with a workflow audit, not a chat interface.

Two of our own published case studies show the gate pattern, not a promise for your processes.

One run of our content-agent platform produced a 3,046-word sourced article for $0.37 across 24 model calls. The pipeline has 14 stages, and 15 deterministic validators check citations, sources, quotations, links, and prohibited claims. The full case study documents the cost engineering that took one article from $8.40 to $0.37.

Lektobot inserted 250 reviewed comments into a real 31,204-word master’s thesis in 1 hour 24 minutes; 400 proposals entered the funnel and 250 verified comments came out. Eleven independent checks must pass, including byte-level inversion and a per-character formatting ledger.

These are verification-pattern examples from different industries, not an answer to the comparable-manufacturer question. We do not have a published comparable metal-manufacturing, legacy-ERP roadmap with phase-level cost and effort to show here. They are Niro Digital’s own published case studies, not an independent audit. They demonstrate the verification pattern; they don’t promise a specific outcome for your process. For completeness on scope: Niro Digital provides four services — AI automation, custom software development, recruiting advertising, and lead generation advertising. This article concerns the first two.

Two external timelines help you sequence the work. The compliance dates in this article are sourced from official BMF and European Commission materials as of 27 August 2026.

Germany introduced mandatory e-invoicing for domestic B2B invoices with effect from 1 January 2025. Since that date, every business in Germany has a statutory obligation to be able to receive e-invoices. From 1 January 2027, companies with prior-year total turnover of at least €800,000 must issue e-invoices to business customers. From 1 January 2028, all remaining companies, with few exceptions, must issue e-invoices.

An e-invoice must be in a purely structured electronic format conforming to the European e-invoicing standard EN 16931 syntaxes; hybrid formats that pair structured data with a PDF are also recognised. Under the German finance ministry’s updated guidance of 15 October 2025, a file that fails the structured-format requirements counts as an “other invoice” — a non-compliant invoice category with VAT-law consequences. The finance ministry states the reform’s aim is that recipients no longer re-key invoice data, avoiding duplicate work and errors, and that it prepares a later transaction-based electronic reporting system.

On the AI side, the EU AI Act entered into force on 1 August 2024, with obligations applying in stages. Since 2 February 2025, the general provisions — including the AI literacy duty — and the prohibitions on prohibited AI practices apply. From 2 August 2026, the majority of AI Act rules come into force and enforcement begins; Article 50 transparency obligations apply from that date. Standalone high-risk AI systems under Annex III must comply from 2 December 2027, and AI embedded in regulated products from 2 August 2028. Regulation (EU) 2026/1744, the Digital Omnibus on AI, is in force and defers those high-risk deadlines.

Use these dates as anchors for invoice-data quality and AI governance, not as a full compliance map. If you invoice German customers, confirm cross-border treatment with your tax advisor.

No reliable, current public data was found in this research for typical SME digitalisation phase costs, so we present no general benchmarks. The article cannot provide a typical SME phase benchmark. Niro Digital publishes indicative price ranges on the pricing page for typical projects; those are the only cost figures we would put in a plan, and they are indicative, not a quote for your company.

To cost the board plan, enter the applicable published /pricing range for each proposed Gate 1 and Gate 2 item, total those ranges against the annual €30k budget, and mark any remaining amount as contingency. Do not budget a Gate 3 item until its preceding exit conditions are met.

Cost here is not a decision to judge by ROAS — return on ad spend — because that metric belongs to paid campaigns, not to whether an internal process gets cheaper, faster, or more accurate. One published value matters more: you own your ad accounts and your code. No hidden fees, no padded metrics.

Take the top-scoring Gate 1 process and make it the next quarter’s pilot. Define a success metric with a target and date, name one owner, and set a go/no-go review. Even after automation, irreversible actions still require human approval.

Three risks to test before approving the pilot are: a CRM nobody adopts, no internal maintenance capacity, and over-ambitious AI. The scorecard tests the first by forcing the absorption score up front — a process with no owner stays out of the pilot. The gate sequence tests the other two: Gates 1 and 2 build stable, owned processes before any model is allowed near them.

Copy this template into your board deck.

No reliable public cost benchmarks exist, and this article cannot provide a typical SME phase benchmark. Budget the plan against the annual €30k budget: enter the applicable published /pricing range for each proposed Gate 1 and Gate 2 item, total those ranges against that budget, and mark any remaining amount as contingency. Do not budget a Gate 3 item until its preceding exit conditions are met. Copy the template, fill in your top three scores, and you have a draft your board can interrogate. The remaining uncertainty is fit: whether your absorption scores are realistic, whether the pilot is sized to your team, and whether the indicative ranges apply to your situation. If you want the plan stress-tested before the quarter-end submission, book a strategy call.