The Valley of Despair: Why Digitalization Hurts Before It Works
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 →Guide · How to digitise
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.
In short
- Digitisation takes the process out of Excel and email and gives it a place, a status, an owner and a history.
- Automation carries out steps without a person, but only on the structured data that digitisation leaves behind.
- AI on top of chaos automates the chaos: it summarises, it does not decide.
Updated September 2026 · 8 min
You have heard of AI, you have a process that runs on Excel and email, and you want to jump straight to automation. We have put AI Agents on quoting and approval flows in the platform, and they worked only where the process had first been digitised. This guide separates the two words, puts them in order and shows, on one concrete case, what changes at each step.
Digitisation and automation: two words used as synonyms
Digitisation moves the process out of Excel and email into a system that leaves structured data; automation carries out steps on that data without people. Digitisation does not mean “we scan the paperwork” and it does not mean “we buy a tool”. It means a request, a quote or an approval has a place, a status, an owner and a history, and the process has left Excel, email and WhatsApp.
Automation means a system carries out steps of the process without a person: it fills in fields, forwards, checks, proposes a decision. It can be a simple rule or a model that reads documents.
They are steps, not synonyms, and the first produces the raw material for the second. Most “automation” failures in small firms are automations laid over chaos: the process stayed in email, with something on top of it moving files between folders.
The test is simple. If you cannot answer “how many cases are in approval right now” without opening a file, the process is not digitised, however many files you have in the cloud.
| Digitisation | Automation | |
|---|---|---|
| What it means | The process leaves Excel and email and moves into a system | The system carries out steps of the process without a person |
| What it leaves behind | Structured data: a place, a status, an owner, a history | Fields filled in, checks, proposed decisions |
| What it rests on | Decisions: who approves, which states exist, what is mandatory | The data left behind by the first step |
| Without the other | A useful record even if you stop here | It automates the chaos: moves files from one folder to another |
One case, step by step
The same request for a quote, digitised, has client, project, deadline, line items, owner and status; only then can an AI Agent work on it. Take a request for a quote that arrives by email with an Excel file attached, and follow it up the three steps.
Today, the person opens the email, reads the attachment, copies the lines into another Excel file, hunts for prices in old quotes, fills it in and sends a new version. What is left behind is a file and an email thread. Nothing that was decided is readable by anyone else without opening the file and asking questions.
Digitised, the same request becomes a case; the file stays attached, but the information inside it is now in fields. The person takes roughly the same steps, but the result is structured, and the question “where are we with the quote for X” has an answer without bothering anyone. What changed is not the amount of work, but what is left after it: a correct row in the record, instead of a file saved somewhere.
Only now does automation have something to stand on. An AI Agent can read the PDF and fill in the lines, match items against the product list, flag what is missing and say “three lines have no valid price”. Without the case, the same request would at best have produced a summary sent over chat.
Why AI on top of chaos automates the chaos
A model over an inbox produces summaries, an RPA (robotic process automation) robot over Excel fails silently, an assistant without data invents. These are the three failures we see at firms that skipped the first step, and none of them is a problem with the model.
The first is the summary. A model set to read an inbox produces summaries, not decisions: it has no way of knowing which quote is the approved one, because that is written nowhere. All that is written is that somebody replied “ok” to a thread with four attachments.
The second is the robot that fails silently. An RPA robot laid over an Excel file automates the position of the cell, not its meaning, and breaks at the first moved column. The file changes because a person maintains it, and the output still looks plausible, which is the worst kind of error.
The third is the assistant that invents. A conversational assistant with no access to the real data, asked for the price of a given product from a given supplier, answers something reasonable. It was trained to answer reasonably, not to look things up.
The conclusion is the same in all three cases: whoever has context solves the problem, whoever does not solves the symptom. It is not an argument against the models, but for the order you put them to work in: on a case with fields, the same tools do what they promise.
The right order and what you measure at each step
The order is digitisation, context, automation; at the first and the last step you measure the same thing: cases closed monthly by the same team. The middle step is not built; it accumulates, if the first was done well.
Operational context is the side effect of digitisation: product lists, prices negotiated with suppliers, the real roster of subcontractors, approval flows as they happen and the decision history. It is clean because it is produced by the process, not reconstructed from exports. And it belongs to the firm, not to the software vendor: change the tool and the context stays, and without it every new tool starts from zero.
The measure is the same at the start and at the end for a simple reason. If automation does not close more cases with the same team, it has automated something other than the work.
The right order and what you measure at each step
- 01
Digitisation
You take one flow out of Excel and email and give it a place, a status and a history. You measure cases closed per month by the same team.
- 02
Context
You let the process run for a few months: product lists, prices, suppliers, an approval history. You measure how much you can find on your own, without asking.
- 03
Automation
You put AI Agents on the repetitive steps of a flow that is already digital. You measure, again, cases closed per month by the same team.
In practice
Construction quoting is the complete example: the bill of quantities becomes a case, the case leaves context behind, the AI Agent fills in the lines. At an MEP (mechanical, electrical and plumbing) contractor with over 100 employees, quoting was digitised first. The request becomes a case with lines from the bill of quantities, with an owner and a status.
After a few months, the cases had left behind an item list with prices and a quote history. Only on that context did the AI Agent come in: it reads the request, matches the lines against the item list and fills in the valid prices. The person checks only what is flagged as missing, and the number followed stayed the same as before any automation: quotes closed per month, by the same team.
The same case continues into material approval and into the as-built file: three steps of the same item, quoted, approved, documented. Each step leaves behind the context the next one needs.
Where it does not apply
Digitisation first, then automation, costs time and decisions; in three situations the effort is not worth it or not needed at all.
- Processes with a single person and ten cases a year: a well-kept file is enough, and a system would cost more than it saves.
- Processes already in an ERP (enterprise resource planning system) with an API (application programming interface): the data is structured, you can automate directly.
- Firms that want “a tool” without deciding who approves what: no system takes that decision for them, and digitisation starts with the decision, not with the tool.
What next
Pick the one flow that hurts most often and write its states on paper: who opens it, what it goes through, who closes it. Digitisation is mostly those decisions; the implementation is the short part.
The hard part comes afterwards, in the weeks when the team works in the new system and still feels that Excel was faster. That is what the valley of despair in digitisation is about.
Before automating a flow
- The flow has a place, not a file.
- Its states are written down, not assumed.
- Every state has an owner.
- You can answer “how many cases are in approval” without opening a file.
- It has run for at least three months in the system.
- You have chosen the number you measure before and after.
Frequently asked questions
Can't we jump straight to automation with a no-code tool?
You can, on a process that is already in a system: it has structured data, it has states, it has an API. On a process that lives in Excel and email there is nothing to build on – you automate the moving of files, not the work. The test is simple: if you cannot answer "how many cases are in approval right now" without opening a file, the process is not digitised.
How long does it take to digitise a process?
One flow, weeks rather than months – provided it is mapped first. Most of the time goes into decisions, not implementation: who approves, what statuses exist, which fields are mandatory, what happens when someone rejects. Firms that try to digitise ten flows at once finish none of them.
What do we keep if we stop after digitisation?
The context: the data, the product lists, the prices, the history of decisions. Even if you automate nothing afterwards, you already have more than you had in Excel – a record you can query, hand over to a new colleague and build on when you choose to. The first step pays for itself even if you stop there.
See which part of the platform fits your process.
Book a conversationOne workflow goes live on a real project, with a success criterion set together.
Get the new guide when it’s out
One email a month, only when we publish. Nothing else.
Related articles
- Guide
- Analysis
The Scalability Spectrum: Why Some SMBs Grow Faster Without Hiring More
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 → - Guide
The Complete Guide to Business Process Automation for SMBs
Which processes a company with 20–150 employees can automate, how to start in four weeks with a single process and what it costs, with our 2026 prices.
Read →