Discovery Pack

Before you decide anything, you have seen AI work or fail on your own data.

Three measuring cylinders filled with letters and documents to different heights; a hand points at the red one, which is fullest.

Most AI engagements start with a list of opportunities. The list is easy to make and hard to use. It says what you could do. Whether it works in your business, it does not say. So you are left with exactly the decision you asked for help with.

Discovery Pack ends somewhere else. Between the two days we try the strongest opportunities on a sample of your own data. When we meet again, the results are side by side, and you point.

How it works

Discovery Pack: two days with a trial in between

The recommendation rests on what the options produced on your own data.

Day 1
Mapping
Processes, data and systems. Three to five concrete opportunities.
→
In between
Trial run
The three strongest are tried on a sample of your own data.
→
Day 2
You point
Results side by side. The person who knows the process chooses.
→
Delivery
Recommendation
Go ahead, wait or leave it. With evidence.
The trial run sits next to your systems. Data is pseudonymised and deleted on delivery.

Day 1: Mapping. We go through processes, data and systems with the people who know them and find three to five concrete opportunities.

In between: Trial run. The three strongest are tried on a sample of your data: 200 customer emails, a quarter of orders or the last 50 quotes. It takes about a week.

Day 2: You point. Nobody has to write down what they know beforehand. The person who knows the process can see straight away whether the model has understood it.

Delivery: Recommendation. Go ahead, wait or leave it. With reasons, an assessment of compliance risks (GDPR, the EU AI Act, industry rules) and, if the answer is to wait, what has to be in place first.

What a trial run shows

What a trial run shows

Three opportunities, tried on a sample of the company's own data. Example.

Sorting customer emails200 emails from September
170 of 200
The 30 errors were all complaints. The model passes those on to a person.
Go ahead
Draft replies to quote requeststhe last 50 requests
31 of 50
31 could be sent after small edits. The rest lacked prices that only the sales manager knows.
With a person in the loop
Deviations in purchase ordersone quarter of orders
9 of 40
Only 9 of 40 known deviations were found. Item numbers are written differently in three systems.
Wait: Clean up the data first
Example with made-up numbers. The form is the one you get: numbers from your own data, side by side.

The numbers in the example are made up. The form is the one you get. Look at the bottom row: A trial run can also show that what is missing is order in the data. That is cheap to learn before a project and expensive to learn halfway through.

Your data

The trial run sits next to your systems and does not touch them. On day 1 we agree exactly which sample to use. Personal data is removed or pseudonymised where possible, we sign a data processing agreement, we only use model services that do not train on your data, and the sample is deleted on delivery.

What you provide

  • Four to six hours from the people who know the processes
  • A data extract within three working days after day 1
  • Read access to the systems being assessed
  • A contact person with the authority to decide who can be asked

Who it is for

The owner of a small or medium-sized business without an IT department who does not know where to start. The manager who needs a basis for a decision to take to the leadership.

Price

€20,000 excl. VAT. Fixed price for the whole engagement.

Write to lek@docujai.com or call +45 40 26 18 72.