Picture a managing director’s desk, say at a metalworking company with around 140 employees. On it lie three proposals from “digital agencies”: six-figure sums, twelve to eighteen months’ duration, “strategic digital transformation”, “end-to-end platform”, “AI-powered process intelligence”.
And next to them, the actual reality: the inside sales team types quotes into Word every day, the warehouse reports stock levels via WhatsApp, and accounting maintains invoices twice, once in the ERP and once in Excel, because the ERP doesn’t deliver decent reports.
If you recognise yourself in this, that’s good news. Because in 2026, a lack of vision is rarely the problem. What’s missing are pragmatic, clearly defined projects with an ROI you can calculate in advance. The pattern is the same across SMEs everywhere: if you start with “a digitalisation project”, you get tangled up. If you start with five concrete processes that you already know today and that demonstrably cost you money, you quickly get things moving.
These are exactly the five projects we’ll look at now. Each can be implemented in four to sixteen weeks, and each pays for itself within twelve months. You don’t have to do all of them. But you’ll find two among them where you’ll think as you read: “Yes, that’s exactly what’s costing us time right now.”
Why digitalisation in SMEs is not an IT project
Before we get to the projects, a thought that most consulting offers elegantly leave out: digitalisation is first a process issue and only then a technology issue. If you digitalise a broken process, you get a broken process that now runs faster and has cost more.
This happens more often than you might think. A wholesaler introduces a new CRM for a six-figure sum, and a good year later the order confirmations are still maintained in Outlook, the CRM is only used for master data maintenance, and the sales team dutifully feeds both systems in parallel.
Nobody had properly defined the order process beforehand. The software didn’t repair the unclear process; it set it in concrete, just more expensively. That’s not a software failure, it’s a question of sequence.
Two surveys from spring 2026 show how widespread this pattern is; at first glance they contradict each other, and both are right. The KfW SME Digitalisation Report (published on 23 April 2026) reports: only 30 per cent of SMEs carried out any digitalisation projects at all in the most recent period, back to pre-COVID levels, with spending falling.
At the same time, a Bitkom survey from 11 March 2026 of 604 companies with 20 or more employees shows: 41 per cent use AI, up from 17 per cent the year before, and 36 per cent want to invest more in 2026 than in the previous year. The biggest obstacles to digitalisation are data protection requirements (77 per cent) and the shortage of skilled workers (70 per cent). Among the internal obstacles, however, the top one is not money but time: 66 per cent cite a lack of time, 48 per cent a lack of financial resources. That is exactly why projects with a twelve-month payback work better in SMEs than transformation programmes. They fit into the time budget that is actually there.
And the sequence that works is pleasingly simple:
Understand the process. Who does what today, in which order, with which tool, and how long does it take? We call this the “as-is analysis with a stopwatch”. Three days of observation beat three weeks of workshops.
Simplify the process. Which steps can go because they grew historically and nobody knows why any more? Experience shows that this is 20–40% of today’s steps.
Digitalise the process. Only now does the technology come in, and it’s usually much smaller than what the vendors claim.
From practice
Before every tool purchase, take three days to really watch a process. These three days often shift the priorities of entire project lists and save you the most expensive thing there is: digitalised chaos.
Project 1: customer portal instead of email ping-pong
Start where your team types the most. For most SMEs, a large share of customer communication lands as email with the inside sales team or support: “Where’s my order?”, “Send me last March’s invoice again”, “What was my discount again?”. Each of these requests costs three to eight minutes, and almost every one could be answered by the customer themselves if they had a portal.
That’s exactly project one: a lean, customer-facing self-service portal. Login, an overview of open and past orders, invoice download, delivery status with a tracking link, return registration, stored master data and terms. Not a 200-feature platform, but the ten things that replace 80% of the emails.
Technically, this is very manageable. Portals like this are built against the existing ERP (Sage, abas, Microsoft Dynamics, Shopware, SAP Business One) via REST or GraphQL, with authentication through an OIDC provider and hosting in a German data centre or on your own infrastructure.
No US SaaS for your existing customers’ master data. (Typical stack: Angular or Next.js in the front end, a NestJS back end, Keycloak as IAM.) Realistic effort: four to eight weeks, depending on the complexity of the ERP connection. Not a mega project.
The effect becomes visible quickly. In projects like this, the volume of recurring status and invoice requests to the inside sales team typically falls by 30–50%, in favourable cases more. For an inside sales team of eight, that quickly adds up to a whole position that suddenly has capacity for real selling instead of status queries. If you calculate a full-time inside sales employee including non-wage labour costs at around €65,000 a year, the portal has paid for itself after eight to twelve months.
One thing up front that will save you trouble: customer portals fail because of master data quality, not technology. If your ERP maintains an order without the delivery address being cleanly assigned to the customer ID, the customer sees junk data in the portal. Invest the first two weeks in data hygiene, not in UI design.
Project 2: automated quote generation
If your sales team complains that they “never get round to selling”, it’s worth looking at how quotes are produced. In many SME teams it works like this: look up the product in the catalogue, read the price from the terms table, factor in the discount, estimate the delivery date, type everything into a Word template, save it as a PDF, send it out by email. That quickly adds up to around 45 minutes per quote, plus the odd transcription error every week.
The solution is a product-specific, web-based configurator that asks for the decisive parameters (material, dimensions, options, delivery quantity, customer discount), pulls prices and delivery times from the ERP or a pricing engine and finally generates a finished PDF quote in your corporate design. Including versioning, an approval workflow above a certain order size and automatic handover to the CRM.
Technically, you build the configurator as a single-page app with a rules engine in the back end. For manageable configurations, your own implementation in NestJS is enough; for really complex variant systems, a dedicated configurator such as Tacton or Encoway is worthwhile. PDF generation runs via headless Chromium against an HTML template, much easier to maintain than the old Word or InDesign pipelines. Expect six to twelve weeks, depending on how complicated your product logic is.
The leverage is large: in projects like this, the time needed to create a quote consistently falls by 60–75%, from 45 minutes to 12 to 15. For a sales team of five people that sends out around ten quotes a day, that comes to, roughly, more than 1,100 person-hours regained per year.
On top of that, transcription errors drop to practically zero, and even a single avoided wrong price in the four- to five-figure range helps pay for the project. The configurator usually pays for itself within eight to ten months.
Quick win
Before you build the configurator: move your terms out of the Excel file that the head of sales updates every three months and into a clean source. A fast machine with wrong prices is more expensive than the manual version.
Project 3: ERP–webshop integration
This project solves a pain that almost every retailer knows: you have an ERP (often Sage, abas, Microsoft Dynamics or a home-grown system) and a webshop (Shopware, WooCommerce, Shopify or an older Magento installation). Both run, but they don’t talk to each other, or only via a nightly CSV import. The result: stock levels in the shop are out of date, overselling piles up, incoming orders are typed into the ERP by hand every morning, and customer master data gets duplicated.
What you want instead is a genuine, event-based real-time integration. Stock changes in the ERP update the shop immediately. Orders from the shop land in the ERP without human intervention. Invoices become available in the webshop or customer portal. And every data object has exactly one source of truth, typically the ERP.
Technically, this is implemented with an event-based architecture using a lightweight message broker or the webhook mechanisms that modern platforms such as Shopware 6 come with. What matters is a dedicated integration layer with clear adapters per system that translates between the data models, instead of wiring the ERP directly to the webshop.
Couple them directly and you’ll run into problems at the next ERP update. (Typical stack: a NestJS integration layer, RabbitMQ as the broker, Kafka only for larger volumes.) Plan for eight to sixteen weeks, depending heavily on how many data objects need to be synchronised and how clean the APIs involved are. In practice, Sage and abas have rougher interfaces than Shopware or Shopify.
The ROI has three faces. Overselling typically drops to almost zero. Manual order entry disappears: with around 50 online orders a day and four minutes per order, that’s roughly 70 person-hours a month, almost half a full-time position. And the duplicated work on the returns side (return in the shop, cancellation in the ERP, stock correction, accounting) becomes a single workflow. All in all, the payback is usually six to nine months.
The one sentence you should remember: define early who holds the truth for which data object. “Both” is not an answer but a race in which, depending on the order of the write operations, the wrong stock level wins. Common and sensible: stock and price come from the ERP, customer and order data originate in the shop and flow into the ERP. Stick to that.
Project 4: document management instead of paper
A surprising amount of paper lives on in SMEs, and above all PDFs that are treated like paper. The incoming invoice is scanned, printed, signed off by two people, scanned again and filed in a folder that nobody really owns. Personnel files in suspension files, supplier contracts in a cabinet on the second floor. And “dig out the contract with supplier X from 2022 for me” costs anywhere between half an hour and half a day.
A digital document management system (DMS) with clear workflows for the most common document types clears this up. Incoming invoices are captured automatically via an email inbox or scan, read by OCR, matched with the purchase order, sent into an approval workflow and archived in an audit-proof, GoBD-compliant way. Contracts get full-text search, deadline reminders and role-based access control. Personnel files live in a clearly separated, GDPR-compliant area.
Custom development rarely pays off here. Established systems are ELO, d.velop and DocuWare; in the open-source space, Paperless-ngx (for smaller set-ups) and Mayan EDMS. The real work lies in the configuration, the OCR quality for your specific documents, the connection to the ERP or accounting (DATEV, Lexware) and the definition of the approval workflows.
Here, a digital agency is less a programmer than an integrator and process designer. Expect eight to sixteen weeks for the initial installation with the three most important document types (incoming invoices, contracts, personnel files); further types are added step by step afterwards.
The benefit: in projects of this size (companies with 100–200 employees), 15–25 hours a week are consistently freed up, spread across accounting, HR, purchasing and management. Add to that the harder-to-measure but very real effects: no more lost contracts, GoBD compliance without a stomach ache, audit readiness in hours instead of weeks.
And one item that is often underestimated: lost early-payment discounts, because incoming invoices sit in the approval workflow for too long, quickly add up to a five-figure sum per year with a larger purchasing volume. This item alone clearly brings the payback of a DMS into the 12-month window.
What really matters in this project is not the technology but the people. A DMS almost always fails because of a lack of buy-in: if accounting keeps maintaining its old folder in parallel out of mistrust, you have two systems instead of none. Plan at least 15% of the project budget for training and change management, and make management the role-model user.
Project 5: AI-assisted email classification
The fifth project is the ideal, low-risk entry into AI, without an AI ever answering a customer email itself. At many SMEs, the central info@ or sales@ address is a catch-all for genuine enquiries, supplier emails, newsletters, job applications, spam, complaints and special requests.
Usually a single person routes all this by hand, based on experience. If that person is ill, the inbox backs up. If they route clumsily, it takes days for the complaint to reach the right clerk.
An AI-assisted classification layer between the inbox and the people handling it solves this elegantly. Incoming emails are automatically sorted into predefined categories (“order enquiry”, “complaint”, “supplier invoice”, “job application”, “spam”), placed in the right folder or helpdesk workflow and given a priority. For sensitive categories such as complaints or court letters, an instant notification goes to the person responsible. Pure process automation: the language model writes nothing.
The most important design decision is where the model runs. For most SMEs, the sensible options are either a locally hosted open-source model (based on Llama 3 or Mistral, depending on language understanding) or an API solution from a provider that signs a data processing agreement and hosts in the EU (Mistral La Plateforme in France, Azure OpenAI in an EU tenant, AWS Bedrock in eu-central-1 / Frankfurt). For particularly sensitive content, such as HR-related or health data, only the locally hosted model is an option.
The classification runs either as zero-shot with a well-built prompt or, at higher volumes, as a fine-tuned BERT-based classification model that is considerably cheaper to run. Effort: four to eight weeks, depending on the choice of model and the available training and test data.
The effect: the average response time to incoming enquiries typically falls by 30–50%. Complaints often reach the right person on the same day instead of after three to four days, and complaints handled faster escalate less often and are resolved more cheaply. With a volume of around 200 emails a day, around 15 hours a week are quickly freed up, which usually puts the payback in the range of seven to eleven months.
When it comes to data protection, real care pays off, because incoming emails regularly contain real names, addresses, bank details and sometimes health information. Three levels, sorted from the cleanest to the riskiest path:
Locally hosted model. The only clean route for HR-related content or health information. Higher effort, practically zero risk.
EU provider with a signed data processing agreement and no permission to train on your data. Viable for standard business mail (order enquiries, supplier emails): Mistral La Plateforme, Azure OpenAI EU, AWS Bedrock eu-central-1.
Shadow use via private ChatGPT or Copilot accounts. This is exactly what happens in many companies today, but it is not tenable under the GDPR: no data processing agreement, a serious risk of fines.
One more sentence on regulation, so that the legal department doesn’t slow things down unnecessarily: the transparency obligations under Article 50 of the AI Act have applied since 2 August 2026 and mainly affect systems that interact directly with people or generate synthetic content. In our assessment, a purely internal classification that writes nothing and releases nothing externally typically does not fall under them. That is interpretation, not a statement from an authority.
What, on the other hand, indisputably applies is Article 4: anyone who uses AI must take measures to support the development of AI literacy among the employees working with it. The obligation has applied since 2 February 2025, and since 27 July 2026 as amended by Regulation (EU) 2026/1744; you do not have to guarantee any specific level of AI literacy. Not a compliance monster, but an item on the project checklist.
We have worked through the logic behind this in detail in a separate guide; it applies by analogy to any form of process-automated LLM use in a company.
How to start the right project first
Five projects, each one worthwhile. But you don’t start them all at once. Realistically, SMEs can run two to three in parallel without the organisation breaking a sweat. Which one first?
The short answer: sort by impact and effort. Plot the five projects on a 2x2 matrix. Axis one: expected ROI in the first year, as a euro amount, not a gut feeling. Axis two: implementation effort in weeks plus internal capacity. The projects in the “high impact, low effort” quadrant are your first two.
In practice, the configurator often lands high up on the impact axis (avoided wrong prices plus more than 1,100 regained sales hours a year) with medium effort, provided there is a clear product catalogue. The ERP–webshop integration usually follows directly after, because it stops acute pain and the payback becomes visible quickly.
The DMS is often deliberately pushed back: the impact is real, but the change management is substantial, and two visible wins beforehand make the next step enormously easier. AI email routing works well as a small, low-risk pilot project and as the team’s “first real AI project”.
Know-how
If you don’t have time for the prioritisation session, use this order as a rule of thumb: customer portal, ERP–webshop integration, quote generation, DMS, AI email classification. First the quick, visible wins, then the organisationally difficult ones.
This order is chosen deliberately: the customer portal brings a quick cash-flow effect and is immediately visible to the customer. The ERP–webshop integration stops acute pain with a quickly visible payback. Quote generation is a direct sales lever and creates goodwill for the next change. The DMS is the most demanding organisationally and needs patience. And AI email classification has the smaller lever, but in return it is the low-risk first AI project.
There is, however, one deadline that can override this order: the e-invoice mandate. Since 1 January 2025, every domestic business has had to be able to receive e-invoices from other domestic businesses, with no transition period. Issuing e-invoices becomes mandatory from 1 January 2027 for businesses whose total turnover in 2026 exceeded €800,000. All others have until 31 December 2027 for supplies made in 2027; from 1 January 2028 the obligation applies to all domestic B2B supplies, with exceptions such as small-amount invoices (Kleinbetragsrechnungen) up to €250 and invoices from small businesses (Kleinunternehmer) (Section 27(38) UStG, Sections 33 and 34a UStDV).
The Federal Ministry of Finance tightened up the details on 15 October 2025 in a second letter. If you are connecting ERP and shop anyway, or automating the quote and invoice chain, plan the format issue in now and not as an emergency operation in December 2026.
Funding: reduce the total cost before you start
A piece of good news that many SMEs underestimate: there is funding at federal and state level for this type of project. The funding landscape shifted noticeably in 2026: some programmes have ended, states have changed their models, and the most important lever today is a different one from two years ago. That hasn’t made it any clearer, as one figure further down will show you.
Here is the status at the beginning of October 2026; before any specific application, it’s worth checking the programme page or talking to the funding advice service of your IHK (Chamber of Industry and Commerce).
What no longer works in 2026
The federal programme “Digital Jetzt” of the Federal Ministry for Economic Affairs and Energy (BMWE) ran from September 2020 to December 2023 and has been closed since the end of 2023; it is only being wound up, and new applications are excluded. “go-digital”, the consultancy funding programme of the same ministry, ended on 31 December 2024. At state level, the “Digitalbonus.Niedersachsen-innovativ” ended on 31 December 2025; as of spring 2026, no successor scheme has been announced.
On top of that, there is a fresh freeze: since 12 noon on 7 July 2026, a temporary application stop has applied to ZIM (Zentrales Innovationsprogramm Mittelstand, the Central Innovation Programme for SMEs) for all project types and project outlines; only international calls are exempt.
According to the BMWE, the reason is exhausted budget funds; at the beginning of October 2026 the stop remains unchanged. The ministry holds out the prospect of a resumption for the period after the 2027 federal budget, naming early 2027 as a target. That is not a date. If an older consulting offer still names one of these programmes as an open route, treat that as a warning sign.
What works at federal level
The Mittelstand-Digital Centres advise free of charge and vendor-neutrally, and offer pilot projects. Not a classic grant, but a good way in to sharpen your project outline. An appointment is better sooner than later: funding for the existing network is being phased out by the end of 2026, and from 2027 a newly tailored network will start, focusing on AI, digitalisation, and IT and cyber security.
The BAFA consultancy funding for SMEs (officially “Förderung von Unternehmensberatungen für KMU”, extended until the end of 2026) subsidises external consultancy services at 50% (old federal states, max. €1,750 grant) or 80% (new federal states and the Lüneburg/Trier regions, max. €2,800). Only individual consultations of no more than five consulting days are funded.
Not a mega lever, but it combines cleanly with the state programmes: up to the end of 2026, a maximum of five applications per company is possible, at most two per year. After that, as things stand, the programme will have ended; as of the beginning of October 2026, no extension has been announced. So if you want to use the consultancy funding, you only have until the end of 2026.
The federal lever that almost nobody has on their radar. When it comes to the sums that a customer portal or an ERP integration really costs, the instrument is not the grant but KfW’s ERP-Förderkredit Digitalisierung (ERP digitalisation loan, programme 511). No minimum loan amount, up to 100% of eligible costs, terms of five to ten years with up to two repayment-free years. In plain terms: you finance the project and pay only interest for two years. Repayment only starts once the portal is already saving time.
The decisive point for smaller projects: in the LevelUp and HighEnd tiers, there is also an ERP grant (ERP-Förderzuschuss) of up to €200,000 on top, which does not have to be repaid (de minimis aid). For the Basis tier, the free KfW digitalisation check is a prerequisite. You apply through your house bank, and you must do so before the project starts. Information sheet as of 23 April 2026.
What works at state level
The most important active programmes:
North Rhine-Westphalia: MID-Digitalisierung. Restructured as of 1 January 2026 (50%, up to €15,000): since then, funding has been limited to the development of digital products based on key technologies such as AI, machine learning, data mining, real-time processing or AR/VR. Funding is awarded by monthly lottery via NRW.BANK; the application period runs until 31 December 2029. For classic process digitalisation such as the five projects here, this is not the right place to go.
North Rhine-Westphalia: MID-Digitale Prozesse, 2026 call closed. The former “Digitale Prozesse” module has been a separate programme since 2026: €500 per consulting day, a maximum of €15,000 per company and project. It only runs in time-limited funding calls. For the 2026 call, according to NRW.BANK, the maximum number of applications has been reached and the application portal is closed (as of 9 October 2026).
North Rhine-Westphalia: the other MID modules. MID-Digitale Sicherheit (50%, max. €15,000) continues in the monthly lottery. For MID-Assistent/in (a wage subsidy of up to €48,000 over two years), the waiting room is closed until further notice, according to NRW.BANK.
Bavaria: Digitalbonus. Standard up to €7,500, Plus up to €30,000 for measures with a particular degree of innovation, a funding rate of up to 50% in each case, running until 31 December 2027. Important for your expectations: only small companies with fewer than 50 employees are eligible, so the 140-employee company from the introduction is ruled out. Process projects such as ERP, CRM or DMS run via the Standard variant; expenditure under €4,000 is not eligible.
Baden-Württemberg: Digitalisierungsfinanzierung (L-Bank). Has replaced the former “Digitalisierungsprämie Plus” grant programme and now runs as a loan of €10,000 to €5 million in three funding tiers (Basis, LevelUp, HighEnd), refinanced from KfW programme 511 and therefore with the same tier logic.
Baden-Württemberg: repayment grants despite the loan. The claim that there are no grants any more is wrong: in tiers 2 and 3, projects of up to €250,000 receive a repayment grant, the “Digitalisierungsprämie des Landes”, of 5.0% of the loan amount, plus an L-Bank repayment grant of 1.0 to 2.0% and, optionally, the KfW ERP grant.
Hesse: Distr@l. Modular funding lines; depending on the module, grants from a few thousand to around €1 million are possible. The submission deadlines in the first half of 2026 (February and April) have passed; however, project outlines can be submitted to the ministry at any time and then go into the next evaluation round. If you are aiming for a cut-off date, prepare your outline in good time.
European programmes. For larger projects, it’s worth looking at the European Digital Innovation Hubs and the Digital Europe Programme. More complex to apply for, but relevant for substantial investments.
Why researching funding becomes a project in itself. The lack of clarity is not a feeling; it has been counted. A white paper from the Mittelstand-Digital accompanying research, published in September 2026 and based on data as of 30 June 2026, counts 175 active funding programmes relevant to digitalisation: 43 at federal and EU level, 132 at state level, spread across around 180 points of contact. 66 of them fund investment in digital technologies, 41 consulting and training.
The study itself notes that companies often cannot tell from a programme’s name whether their project is eligible at all. In plain terms: without a guide you won’t find your programme, and the guide is cheaper than the search. An appointment at a Mittelstand-Digital Centre or with your IHK’s funding advice service is therefore the cheapest first step in the project plan.
Remember this one thing
If you take away only one sentence from this section: the application comes before the project starts. Everything else can be sorted out later; this order cannot.
An important tip from practice: even a grant in the region of €15,000 noticeably brings forward the payback of an €80,000 project. The only thing that matters is the order. Applications must be submitted before the project starts. If you start first and think about funding afterwards, you have usually lost out. So plan the funding application as the first step in your project plan.
Conclusion: less vision, more fixed dates in the quarterly plan
SME digitalisation doesn’t fail for lack of ideas, but for lack of implementation granularity. Each of the five projects presented is small enough to be manageable within a quarter and big enough to make a noticeable difference. And each pays for itself in twelve months or less if you take the process groundwork seriously.
What you can do right now: go through the five projects and assign each one a rough euro figure. What does today’s status quo cost you? If you arrive at two or more projects with a pain in the six-figure range, you have a business case. From then on, you no longer need vision slides, but a digital agency for SMEs that works quickly and with a focus on processes, without any strategic quarterly workshops.
The managing director from the introduction would probably end up deciding on two projects in parallel, say the configurator and the ERP–webshop integration. Six months later the first one is running, and shortly afterwards the second.
The inside sales team gains time for real sales conversations, the webshop no longer oversells, and the total cost (Total Cost of Ownership, TCO) of both projects together is below the cheapest of the three vision proposals. Those then sit in a folder that nobody opens any more.
For the areas of enterprise software development and AI consulting for SMEs, we have written our own practical guides that go deeper into architecture and project flow. But the most important thing happens before that: an honest assessment of your current process and the question of which two projects you will really see through in the next twelve months, not which five appear on a vision slide.
As of 5 October 2026. Funding conditions and programme status change frequently, so before any specific project, always check the current status with the relevant programme provider.