Josselin and I figured it could be really useful to share with you the most classic AI use cases we have deployed in companies, and their real impact.
We hear a lot about AI, but concrete details and real feedback are often missing. We sorted through it and identified 4 major use cases that speak to everyone, from freelancers to much more structured companies. A 5th classic one is lead generation and prospecting with AI; we will do a dedicated newsletter on it soon.
For each case, we detail the problem encountered, the proposed solution, how we deploy that solution, concrete examples and the final impact for the company.
Case 1: instant information retrieval (or intelligent search)
The goal? Find any piece of information in seconds, even across thousands of documents.
Problem
On average, we spend 30% of our time searching for information. At an individual level, that is huge. And this information is hard to access. Companies pile up data and documents all over the place, created by different teams and employees. This information is scattered, poorly structured and hard to find.
The result: a considerable amount of time is wasted looking for the right information in the right place, customer replies are delayed, and decisions are made without full context.
Solution
An interactive space where you can share your need and the AI takes care of finding the right information (and shares the source document with you). A kind of internal ChatGPT for the company, connected to the company's tools (CRM, ERP, Slack, email, ..) including the knowledge management tool, while respecting each user's access permissions.
It is a bit as if you could chat in a conversational format with any internal tool.
These assistants can instantly retrieve a precise piece of information while systematically citing their sources, without hallucinating.
How?
There are several options for deploying this type of solution. The one we use is Dust (a French tool 🇫🇷) with:
- Connection to drives (Google Drive, Notion, Confluence, Sharepoint)
- Link with the CRM (Salesforce, HubSpot, Attio)
- Access to messaging tools and internal documentation
It is a turnkey solution but one that requires custom integration to create assistants tailored to business needs (comparable to GPTs: assistants with custom instructions and equipped with specific tools), adjust the permissions per team, schedule assistants to run automatically, ..
And above all, there is a very strong change-management challenge: you need to train the teams to use this new tool (what can and cannot be done, ..) with follow-up on internal usage. There are always adoption gaps within teams and between teams.
Concrete examples
A few quick examples to help you picture it:
- “What is the customer refund process?” → The agent instantly finds the exact procedure while citing the source document and its last update date.
- “What were the exchanges with client X over the past 6 months and what came out of them?” → The agent compiles the emails, meeting notes and tickets to provide a complete summary with precise references.
A video is worth more than a list of examples
Measurable impact
Concretely, it makes teams more autonomous, avoids working in silos and smooths the flow of information internally.
In terms of numbers, we see search time divided by 10 on average across tasks (from several minutes to a few dozen seconds). Onboarding is accelerated for new employees, with a reduction in redundant requests between teams.
Practical information
The cost of this type of solution is around 29€/month per user. Implementation is fast (first results possible within a few hours); it is mainly upfront work to build the assistants, follow-up with workshops to train the teams, and continuous iteration (team feedback, process changes, …)
It quickly pays for itself when an employee saves at least 1 hour per month, or 15 minutes per week.
Spoiler: we get there fast! 😊
Case 2: a supercharged customer service
The goal? Answer simple tickets instantly and cut the handling time for the other tickets.
Problem
Support teams are overwhelmed by customer tickets coming in from every channel: Intercom, HubSpot, Crisp, Front, email, WhatsApp… Simple, repetitive requests monopolize valuable time that should be spent on complex cases. And by the way, teams often prioritize the simple tickets (usually only to realize how many complex tickets they have to handle).
The result: response times get longer, customers get frustrated, teams are overloaded and support costs skyrocket.
Solution
An AI agent connected to your ticketing system that automatically categorizes each request and answers it according to its level of complexity. The agent relies on an exhaustive knowledge base of your company and respects your tone of voice. It handles simple requests autonomously and prepares the ground for human agents on complex cases.
How?
Whatever your ticketing tool, we can deploy AI support agents. Depending on the volume, you can go with off-the-shelf solutions and complement them with custom work, or go 100% custom.
Spoiler: we go much further than the built-in solutions like the support agent at $1 per resolved ticket from HubSpot or Intercom.
The architecture:
- An AI agent connected to your knowledge base
- An automatic categorization system by level of complexity
- Level 1: instant autonomous reply
- Level 2: information gathering + handoff to an agent with full context
- Optional connection to other tools (CRM, product database) to retrieve context
- Automatic respect of the company's tone of voice
To do this, we build the knowledge base (relying on the existing one or not), we reproduce your business logic for categorization and define the agent's autonomy thresholds. The agent adjusts the formatting to your standards.
Concrete examples
A few quick examples to help you picture it:
- Simple ticket (Level 1): Question: “How do I reset my password?” The agent replies instantly with the detailed procedure, without any human intervention.
- Complex ticket (Level 2): Question: “I have a billing problem on my order from September 15th” The agent automatically retrieves the order history, identifies the nature of the problem, categorizes the ticket as “Level 2 - Billing” and hands it off to an agent with all the context already prepared + a draft reply to edit or validate.
Measurable impact
Concretely, we notice on average 30% of simple tickets in companies. That is the first wave we automate within a few weeks. The response time on these tickets drops to under 2 minutes, no matter the time of contact.
We iterate continuously to increase the categories handled by the AI agent. For an e-commerce company, we managed to reach 76% automatic handling.
We reduce the handling of complex tickets thanks to the context already collected and the draft reply.
And customer satisfaction goes up: shorter handling time, more complete answers, optimized formatting,… which may seem counterintuitive…
Practical information
In terms of cost, whatever the solution, there is the implementation cost, which varies with the complexity of the support service. Then a usage cost per ticket handled.
This ranges between 0.20€ and 1€ per ticket depending on the complexity, the volume and the tool used.
Regarding the return on investment, we calculate the cost per ticket upfront based on the team structure to share the savings achieved. Here, for example, we land on a cost of about 20€ per ticket.
For the setup time, it all depends on the complexity of the support, but on average, count a few weeks for a v0 (categorization, context retrieval) to a v1 (handling simple tickets). And there is continuous iteration work to broaden the range of tickets handled.
I didn't mention voice agents for the support team, which are very useful for taking inbound calls, understanding the request, handing off to the teams, autonomous handling, creating tasks in the CRM, and so on. I have written several editions for those who want to dig deeper, including this one:
AI voice agent: what you can (really) automate in 2025. A field report
Case 3: automated and intelligent monitoring
The goal? Capture 100% of strategic information (market, competitive, regulatory) without spending a single minute a day on it.
Problem
Teams spend hours every week manually monitoring their competitors, regulatory news and market developments. They browse dozens of websites, scan the media and check government sites to detect regulatory changes.
This time-consuming work is often incomplete: critical information slips through the cracks, and the summary takes time to produce.
Solution
An automated monitoring system that collects, filters and summarizes the relevant information for you. You receive a structured recap daily, weekly or monthly by email (or otherwise), with all sources cited and ranked by priority. Here is an example for a call for tenders:
How?
There are several approaches to collecting information:
- passive, via an RSS feed for example: as soon as content is available, the information is processed
- active, via scraping (Linkup, Firecrawl, Apify, ..)
Once the information is collected, we process the data (filtering, categorization, extraction, …) to store it. Here on Supabase, for example:
The advantage with Supabase is that you can then do RAG to retrieve information on a topic in a conversational mode.
The complete process:
- Automatic data collection from the configured sources
- Filtering by AI agents: relevance, importance, category
- Structuring and formatting of the information
- Storage in Google Docs, Notion, a database or another tool
- Automatic generation of periodic summaries
- Sending the recap by email or updating the monitoring document
Concrete examples
The use cases are numerous, from competitive monitoring (products, prices, communication, hiring) to regulatory monitoring (laws, standards, decrees) and market monitoring (trends, innovations, new entrants, mentions, reputation, sector news). Concretely:
- Competitive monitoring: Daily monitoring of 5 competitors: automatic detection of new product launches, price changes, LinkedIn posts, job openings. A weekly summary sent on Monday morning with an analysis of strategic moves.
- Regulatory monitoring: Monitoring of official sites (Journal Officiel, CNIL, sector authorities). An instant alert as soon as a new text impacting your business is published, with a summary of the key points and implications for the company.
- Market monitoring: Aggregation of articles from 20 specialized media, filtering the top 10% most relevant, categorization by theme (technology, finance, HR, etc.), and generation of a monthly newsletter with emerging trends.
Measurable impact
We find that in general, there is at least one employee doing this type of monitoring internally by hand (5-10h/week). With this system, the time is limited to proofreading, with coverage broadened to 100% of the configured sources monitored 24/7. No critical information is missed + a structured, actionable summary.
In some cases, you can set up real-time alerts on events deemed “critical”.
With a complete, searchable history of all the collected information.
A quantified example: For a team that used to spend 8h/week on manual monitoring:
- Time saved: 35h/month, or 0.2 FTE
- Volume of information processed: 5x to 10x
- Responsiveness: instant alerts vs. random discovery
- ROI: pays for itself in 1-2 months
ROI: Immediate if you spend at least more than 2h/week on manual monitoring
Case 4: automatic data extraction and entry
The goal? Eliminate 100% of the copy-pasting from documents into your business tools.
Problem
Every day, companies receive documents via forms or email: quotes, purchase orders, invoices, contracts. These documents contain critical information that has to be extracted manually and then entered into different tools (Excel, accounting software, CRM). It is really repetitive and time-consuming, generates data-entry errors and delays the processing of files…
Solution
A system that automatically detects the arrival of new documents, extracts the relevant information via OCR, structures it according to your needs, and sends it directly into the right fields of your business tools. Zero human intervention for standard cases.
How?
In general, it is more of a custom workflow where we reproduce the business logic. We use n8n for orchestration and automation (with the option to host it on your servers), an OCR like Mistral for data extraction, and a small AI model like Gemini 2.5 Flash for processing.
The complete process:
- Automatic detection of new documents (email, form, shared folder)
- OCR to digitize and read the content of the documents
- AI agents to extract the precise information
- Structuring and formatting of the data according to your business rules
- Automatic sending to the target tools (Excel, CRM, ERP, accounting software)
Concrete examples
This is a classic case in companies, and the range of documents that can be handled is very broad: quotes or sales proposals, purchase orders, supplier and customer invoices, contracts and amendments, administrative forms, bank statements…
Concretely:
- Supplier invoice: An invoice arrives by email. The system automatically extracts: supplier name, invoice number, date, amount before tax, VAT, total amount, due date, and sends this data directly into your accounting software with the PDF attached.
- Customer purchase order: A purchase order is received via a form. The system identifies: customer, ordered products, quantities, unit prices, delivery address, and automatically creates the opportunity in the CRM with all the information structured.
- Contract: A signed contract arrives by email. The system extracts the key clauses (duration, amount, important dates, stakeholders) and updates the contract database with automatic alerts on deadlines.
Here is an example of a workflow in the Microsoft environment for extracting information from invoices:
Measurable impact
The processing cost is very low (<0.05€ per doc) and copy-paste tasks are eliminated. The idea is to free up teams for higher-value tasks.
