Updated on 6 min read
Sébastien Nobour
Founder of DEVEDANOS
For a higher-education group, we worked on integrating AI into its platform for work-study placements, open only to its students. Automatic CV analysis cut manual data entry time by 80%, above all for the group’s staff.
In some processes, a person reads a document, then copies its content into a piece of software. A CV, an invoice, a purchase order. These are the first tasks to look at.
To find them in your business, look for what someone does by hand, over and over, several times a day or a week. For example:
All of these tasks can be automated. AI is needed for only some of them.
As long as the information arrives in the same form, a rule written into the software is enough to process it.
AI becomes useful when that form changes from one document to the next: an image, free text, a layout nobody anticipated.
The software then passes the document to an AI model, with instructions. The model extracts the information in a format you have approved. The software checks that the information matches that format before saving it. Where an error is costly, a person reviews it before it is saved.
In our view, this is the right way to start: a rule wherever a rule is enough, AI at the step where it no longer is, and an automated check after every response from the model.
Tools such as n8n, Make or Zapier connect existing software, step by step, without writing code. For a simple workflow between tools that already offer these integrations, they are enough.
Custom software becomes necessary when the process applies your business rules, when several people are involved with different permissions, or when the information has to be checked before it enters your systems. It then takes the form of a business application, in which AI handles one task.
Can this software be built by an AI tool, with nobody reviewing the code? It all depends on what happens if the application breaks down, produces wrong results, or leaks its data.
On a higher-education group’s platform, AI reads the image of the CV and extracts the student’s profile from it, in a format approved with the group’s staff.
Some offers promise you “your own AI”. But training a model costs tens of millions of dollars: unless you are handed the model itself, what is sold under that name runs on a provider’s model. So we don’t sell “your own AI”: we connect your software to the model of a company that develops AI (Claude from Anthropic, ChatGPT from OpenAI, Gemini from Google, Mistral, or Chinese models such as Kimi and GLM).
The AI model does not belong to you: its vendor decides its price, its terms and how long it stays available. Depending on a provider is nothing unusual. The risk lies in ignoring it.
A dependency can be managed when it is known, written down and prepared for before the provider changes anything. That means naming the model in writing, testing a second model and tracking spending every month. The five measures that prepare for this change are set out in our article on “your own AI”.
If a service provider offers you an AI integration, ask them for the name of the model in writing, and for the documents that prove it: the contract, the list of sub-processors, the provider’s invoices.
When the model is hosted by its vendor, the software sends the documents to be processed to that vendor. If these documents contain personal data, the UK GDPR (Article 28) prohibits the service provider from engaging the model’s provider without your prior written authorisation. You therefore know, in writing, who the software entrusts your data to.
When an employee pastes the same document into an AI assistant the company has not chosen, it is a form of shadow IT.
You can also choose an open model, one whose vendor publishes the model file, and run it on a server rented in the UK or the EU. A set of standard questions, whose right answers are known, then tells you whether it does as well as a large commercial model on your own use cases.
In your custom application, we hand an AI model one specific task from your business, such as reading CVs.
We develop your custom software, usable within a month, for a fixed monthly budget, and we adjust it continuously as your business evolves. The budget is built with you, from your figures and your goals: together, we set the amount and the problem your application has to solve. This budget covers development. Using the AI model is a separate expense, which depends on the volume processed. We track it every month.
Your application’s users work with it from the first month of development, and how they use it determines the next feature or features. Each feature goes into production as soon as it has been accepted and you approve its release. We measure the gain with you: the time spent on a task before, then after. We use this measurement to choose the next step.
And so that what worked yesterday still works tomorrow, we apply a zero-regression guarantee: every regression is fixed and covered by an automated test so that it does not come back, at no extra cost.
The source code, infrastructure, deployment pipeline and documentation are your intellectual property once the first monthly invoice is paid. A trial period of up to 15 working days in the first month: no invoice if you do not want the deliverables.
Our page on business process automation with AI sets out what we automate, and how.
Our guide covers the questions to ask before signing a quote for the development of your custom tool.
Do you copy your invoices or purchase orders into an Excel file? You can turn that Excel file into a web application, in which AI can read these documents for you.
Which task do you still do by hand, over and over? Shall we talk?
An Excel file shared between several users, errors, conflicting versions: when the spreadsheet has become an application, and what moving to a business web application changes.
Web application takeover: what to recover before changing software supplier, why tests come before any change, and when a rebuild is needed.
Finding a technical co-founder for your startup: what a technical co-founder brings, what you give up, how to choose one, and what to put in the shareholders' agreement.