Digitisation before automation: why the order matters
AI automates well only what it understands, and it understands only what is clean. Digitisation → context → automation, explained on one real case.
Read →Analysis · How to digitise
A configurable ERP hands you settings and the job of deciding how you work. An opinionated workflow for your industry comes with statuses and approvals decided.
In short
Updated September 2026 · 10 min
You have received an offer for an ERP that "adapts to any business" and you are wondering why you would pay for a process decided by someone else. We took the opposite road in the platform: the workflow of your industry comes decided from the experience of firms like yours, and you fill in what belongs to the firm, not the structure. This analysis shows what you actually buy when you buy "configurable", why Linear grew by refusing settings, and where our decision ends and yours begins.
A configurable ERP delivers an empty frame, and the statuses, approvals and order of steps are decided by you, during implementation, on billed consulting hours. This is not an accusation, it is the promise on the product page: "extremely flexible and easy to adapt to the needs of any business", "modular architecture", "for companies of any size". The most honest wording we found is WinMentor's: the program can be used even by people with no accounting background "once the initial configuration has been completed" (WinMentor, 2026). The value starts after you have done the work.
That work has a cost that does not appear in the offer. Panorama Consulting counted, in its 2026 report, what happens in ERP projects: more than a quarter went over budget and almost a quarter over schedule, and the named cause is that "organizations discover fatal misfits late in the project", when they fix them with additional technology, scope expansion and custom builds (Panorama Consulting Group, 2026). Their sample is firms larger than yours, with dedicated ERP budgets. At a 60-person firm a misfit is not fixed with an extra module, it is fixed by going back to Excel.
Eurostat's figures say how far the "configure it yourself" model has reached at your size. Among Romanian firms with 10–49 employees, one in four has an ERP (enterprise resource planning system); among those with 50–249, two in five. In the EU the proportions are almost double.
| Romania · 10–49 employees | 25.65 % |
|---|---|
| EU · 10–49 employees | 41.08 % |
| Romania · 50–249 employees | 41.05 % |
| EU · 50–249 employees | 69.93 % |
Eurostat, isoc_eb_iip, enterprises using ERP software, reference year 2025
An opinionated workflow by industry comes with the statuses, the case structure, the order of steps and the approval circuit decided from the experience of firms in your industry. It is not a generic ERP you configure, and it is not custom software built for you alone. It is a third category, one the Romanian market does not yet have a word for: the workflow is decided once, for every mechanical contractor or every shopping centre, and each firm fills in what is its own.
The best-known company that grew this way is Linear, the application in which product teams track their work. They built for a single kind of team and decided on its behalf how work gets done. Jori Lallo, co-founder: "we design it so that there's one really good way of doing things", because flexible software "lets everyone invent their own workflows, which eventually creates chaos as teams scale" (Figma, 2024). Karri Saarinen, the other co-founder: "I don't think you can build the optimal tool for anything if it's very flexible or endlessly customizable" (First Round Review, 2025).
We go one floor below them. Linear decided a way of working for a segment: product teams, wherever they are. We do not decide a way of working for "small businesses", because the small business does not exist as an industry: a mechanical contractor does not quote like a distributor and does not approve materials like a shopping centre. We decide the workflow per industry, starting with construction, and put it into production at several firms before we call it decided.
| Configurable ERP | Opinionated by industry | |
|---|---|---|
| Case statuses | You define them, during implementation | Come decided, the same at every firm in the industry |
| Case structure | Fields and sections picked from a list | Decided: what a quote, a MAF case, an as-built file contains |
| Order of steps | You draw it, with the consultant | Quoting, material approval, procurement, as-built file |
| Approval circuit | Roles and rules to be built | Decided: site supervisor, designer, client; you add the names |
| Where AI steps in | Wherever you switch on a module | At fixed points, with a person confirming |
The estimating program comes with the forms, the order and the structure decided by the norm and by practice, and you fill in only quantities and prices. No builder asks an estimating program to be "configurable for any business". The F1–F5 forms are those of Government Decision 907/2016 (Romanian Government, 2016), the items carry estimating norms, the summary sheet is composed in a single order, and that is exactly what you buy: the structure decided, so that you do not invent it in Excel.
You know the alternative. The estimate in Excel is a web of formulas and shortcuts understood only by the person who built it, and changing a single cell breaks prices in ten others. The estimating program is an opinionated workflow for one step of the process. We do the same for the whole chain, from the quote to the as-built file, with the same rule: the structure belongs to the industry, the values belong to the firm.
What the estimating program decided for you
Items, estimating norms, the F1–F5 forms under Government Decision 907/2016.
Take-off, estimate by category of works, summary sheet.
Quantities, unit prices, labour and transport rates.
Forms, order and structure: they belong to the industry, not to the firm.
Quoting, material approval, procurement and the as-built file are one case, in this order, with the AI Agent at three fixed points of the workflow. The item quoted from the bill of quantities is the same one that enters the material approval case, the same one ordered after approval and the same one that reaches the as-built file with its declaration of performance. Each step has decided statuses, a deadline and an owner, and the case does not move to the next step without the decision of the person at the current one.
The AI Agent (artificial intelligence) steps in where the work is reading and matching, not deciding. At quoting it reads the bill of quantities and matches the items with the firm's product list. At material approval it pulls the technical sheets and declarations of performance from the product library and flags what is missing. At the as-built file it assembles the chapters from the documents already filed along the way. At each of the three points, the result is a proposal that a person confirms before it becomes part of the case.
We do not put a model over the whole process. A model let loose on emails, spreadsheets and chats produces plausible results nobody checks, and in construction a plausible price or a plausible technical sheet costs more than the copying work it replaced. The points are fixed precisely so that every result has a person and a field in which it gets checked.
Decided are the statuses, the case structure, the order of steps, the approval circuit and the points where AI steps in; yours are the fields, the approvers, the thresholds, the library and the prices. The boundary is easy to remember: the structure belongs to the industry, the values belong to the firm. Two mechanical contractors have the same approval circuit, site supervisor, designer, client, but different people in each role, a different threshold above which they ask for a second signature and different products in the library.
That is why the first week is mapping, not configuration. We look at how the case travels today between Excel, email and WhatsApp not to rebuild the structure after it, but to learn who the site supervisor is, above what amount the owner steps in, which suppliers have negotiated prices and which documents already exist in the firm. Mapping fills in the values of a workflow that exists; it does not draw statuses and does not invent circuits.
What we discover during mapping goes back into the workflow, for the whole industry. When the third firm has the same need, it is no longer an exception, it is a gap in the workflow, and we fix it once, for everyone. When a single firm wants something else, we tell them it is something else; we do not build a setting for them.
At a mechanical contractor with more than 100 employees and several sites running in parallel, the first workflow put into production was material approval. The case structure and the approval circuit were the ones decided for the construction workflow: case generated from the product library, site supervisor, designer, client, status per item. Nobody drew a status.
Implementation time went on what belongs to the firm: importing the product library from the technical sheets they already had in folders, the names of the approvers on each site and the threshold above which the owner wants to see the case. The AI Agent stepped in at a single point at first, the technical sheets from the library, and the office person confirmed every case before sending.
The same workflow continues in quoting from the bill of quantities, in the material approval case and in the as-built file: three guides about the same item, quoted, approved, documented, in the order the workflow is decided.
A workflow decided in advance is useful as long as it is opinionated, not stubborn; the distinction is Dharmesh Shah's, from 2006, and we take it seriously. In three situations the industry's workflow is not yours.
At the next software offer, ask for a single document: the list of decisions the implementation leaves to you. If the list has statuses, circuits and case structures, you are buying configuration. If it has names, thresholds and products, you are buying a workflow.
Before any workflow, the process has to leave Excel and email; that order is the subject of digitisation before automation.
What to ask before you sign
The structure belongs to the industry, not to the company: the same statuses and the same circuit for every mechanical contractor. What differs between firms, who approves, above what amount, which products, which suppliers, is filled in during mapping, in the first week. If your process differs in structure rather than in values, it is not the right workflow and we say so in the first conversation.
It does anything once you have configured it, and configuring it is exactly the work you are paying not to do. Panorama Consulting's 2026 report shows that serious misfits are discovered late in the project and fixed with extra modules and custom development. A workflow decided for your industry has no misfits left to discover.
At fixed points of the workflow: it reads the bill of quantities, matches technical sheets from the library, checks what is missing from the case. At each point a person confirms. A model let loose over the whole process produces plausible results nobody checks, and a silent error costs more than the copying work it replaced.
Values you change yourself, any time: approvers, thresholds, fields, library. The structure we change, for the whole industry, when experience shows it is needed, the way an estimating program updates its forms when the norm changes. We do not build an exception for a single firm, because the exception is exactly the configuration you got rid of.
One workflow goes live on a real project, with a success criterion set together.
One email a month, only when we publish. Nothing else.
AI automates well only what it understands, and it understands only what is clean. Digitisation → context → automation, explained on one real case.
Read →Performance drops in the first weeks after go-live and climbs above the old level from month three. The stages, the critical moment and seven strategies.
Read →Linear and scalable growth are the two ends of one spectrum. Four questions show where your business sits and how automation moves it to the right.
Read →