Technology and innovation, six simulated engagements.
Six companies with fictional names, in Europe and the United States, from two partners' idea to a digital services company with one hundred and eighty consultants. In each simulation, the AI answers, sorts, prepares and documents. Decisions remain with the teams, from the roadmap to incident response.
These simulations illustrate our method using typical cases. The companies and their exchanges are fictional.
What AI is already changing, from code to customer support.
of the software vendors surveyed for the Numeum and EY Top 250 rank agentic AI as their top choice.
of European information and communication businesses use AI, compared with 20% of businesses overall.
of developers use AI or plan to do so, and 51% of professionals use it every day.
complaints received by the FBI in 2025, a threshold crossed for the first time in twenty-five years. Among them, 22,364 mention AI.
Since 11 September 2026, the Cyber Resilience Act has required manufacturers to report any actively exploited vulnerability and any serious incident affecting their products, with an early alert within 24 hours then a notification within 72 hours.
Six companies, six starting points.
A quoting and invoicing software vendor whose support team is overwhelmed after every update, even more so since the e-invoicing reform.
Every user finds their answer without leaving the app.
- Héliographe develops quoting and invoicing software for tradespeople and small businesses, in France and Spain. Its twenty-five employees are split between Bordeaux, where the product and support teams work, and Barcelona, where the team looks after Spanish clients.
- Every update drives up ticket volumes, often with the same questions. Four people answer them, with several days’ delay.
- Since 1 September 2026, all businesses have been required to be able to receive electronic invoices, and clients write to ask whether their software is ready. Questions about issuing electronic invoices, which will be mandatory for SMEs and micro-enterprises from 1 September 2027, are already starting to arrive.
- An assistant is integrated into the app, in French and Spanish, and introduces itself as an AI. It answers from the help centre and the release notes, and it sees the screen the user is on to guide them step by step.
- When it does not know the answer, or when a client asks for a refund, it opens a ticket that summarises the exchange and the account configuration. Support takes over without asking the questions again.
What remains with support
Support alone handles refunds and disputes. The assistant does not modify any invoice and gives no advice on VAT, which remains the client’s accountant’s responsibility.
- Tickets opened after an update
- Time to first response
- Questions resolved without a ticket
- Clients ready for e-invoicing
A digital services company with one hundred and eighty consultants whose responses to tenders are often wrapped up at night, on the eve of the deadline.
Every response ready well before the deadline.
- Graphomètre brings together one hundred and eighty consultants in software development and data, spread across Lille, Brussels, Luxembourg and Amsterdam. Its clients are banks, insurers, public administrations and European institutions, which almost always select their providers through tenders.
- Tenders appear every week on the public portals of the four countries, and private buyers send theirs by email. The partners spot a portion of them, often late.
- Each response mobilises a partner and two consultants for whole evenings, copying references and reformatting CVs as requested. Some go out incomplete, others do not go out at all.
- An agent reads each tender as soon as it is published and produces a brief, with the requirements, scoring criteria, documents to submit and dates. The partners decide from this brief whether to respond.
- When they decide to bid, the agent prepares a first draft from past responses and the reference library, then a matrix linking each requirement to the page of the response. It formats the CVs of the consultants chosen by the partners to meet the tender requirements.
What remains with the partners
The partners alone decide whether to respond and set the price. They also choose the proposed consultants, and the agent neither filters nor ranks any consultant.
- Tenders spotted in time
- Time between publication and the decision to respond
- Responses submitted complete
- Hours spent per response
A manufacturer of air quality sensors whose two technicians answer installers in five languages.
Every installer guided through to the green light.
- Armillaire designs and assembles air quality sensors in Grenoble for schools and office buildings. They measure carbon dioxide and fine particles, then send their readings over the building network. The company has sixty people, with two sales offices in Munich and Milan.
- Sensors are installed by electricians and integrators, who call support whenever an indicator stays amber. Two technicians answer in French, German, Italian, English and Spanish, with colleagues’ help when they struggle with a language. Most calls are about the same settings and delay work on actual faults.
- European rules add further requirements. Since 1 August 2025, connected sensors sold in the EU have been required to meet cybersecurity requirements, and the Cyber Resilience Act has required since 11 September 2026 that actively exploited vulnerabilities be reported.
- An agent replies to installers on WhatsApp and by chat, in their language, and introduces itself as an AI. It asks for a photo of the indicator and the serial number, then guides troubleshooting using the model’s datasheet.
- When the diagnosis goes beyond the datasheets, or when a sensor seems faulty, it passes a complete file to the technician, with the photos, serial number, firmware version and latest readings.
What remains with the technical team
Technicians alone decide on a warranty exchange and on each firmware update. The security lead rules on each vulnerability report, and the agent triggers no remote update.
- Calls handled without a technician
- Time to complete a diagnosis
- Sensors returned with no fault found
- Technicians’ time on real faults
Two partners who want to sell an API that reads supplier invoices to accounting software vendors, and who have no clients yet.
An offering refined before the first meeting.
- Nocturlabe is the project of two partners from Toulouse. She is a developer, and he was a project manager at an accounting software vendor. Their API reads a supplier invoice, as an image or PDF, and extracts the data needed for accounting. They want to sell it to vendors, who would integrate it into their software.
- The prototype fits in a code notebook and a few test invoices. The partners have neither a prospect list nor proof that a vendor would pay.
- They hesitate between raising funds right away and waiting for a first client.
- In a few days, the prototype becomes an online demo and a page presenting the offer. Before any meeting, the offer goes before a panel of simulated buyers, agents built on profiles of product directors at accounting software vendors.
- The panel raises an objection the partners had not seen. With e-invoicing, invoices between businesses established in France will arrive with their data already structured, and the API will have nothing left to read there. The offering is refocused on invoices from suppliers established abroad, which mandatory e-invoicing does not cover, and on receipts attached to expense claims.
What remains with the partners
A simulated buyer never pays; real validation requires a vendor to sign and pay. The partners run every meeting themselves and set the price.
- Versions of the offer tested on the panel
- Meetings secured with software vendors
- Vendors engaged in a paid pilot
- Time between first contact and signing
A Texan planning tool vendor for maintenance teams, whose client feedback piles up without anyone reading all of it.
Every piece of client feedback is linked to the relevant feature.
- Based in Austin, Backstaff develops a tool that schedules the work of maintenance teams in factories and warehouses across the United States. Technicians receive their jobs on their phone, and managers track progress from a dashboard. The company has forty people.
- Client feedback arrives from everywhere, from support tickets to sales notes, via app store reviews and satisfaction survey responses. No one reads all of it.
- The product lead builds her roadmap around requests from the biggest clients, whose voices are heard most, while other clients' requests get lost.
- An agent reads every piece of feedback as soon as it arrives and files it under the feature it concerns, separating bugs from enhancement requests. It links each piece of feedback to the client’s account and keeps the client’s exact sentence alongside its categorisation.
- Every Monday, the product lead receives a per-feature summary, with client quotes and the week’s new feedback. She can trace each line back to the original feedback.
What remains with the product team
The product lead alone decides the roadmap and priorities. The agent works internally and replies to no client.
- Feedback classified within the week
- Feedback escalated through to a product decision
- Time between a bug report and ticket creation
- Clients informed of an enhancement they had asked for
A company that monitors US SMEs’ systems day and night, whose analysts are swamped with alerts.
Every alert reaches the analyst with its case file.
- Watchbill monitors the workstations, servers, email accounts and cloud access of US SMEs, from New York and Chicago. Its seventy people, mostly analysts, work in shifts round the clock. Some of its clients are financial firms subject to the State of New York’s cybersecurity rules.
- Detection tools produce thousands of alerts per week, and most turn out to be harmless. At night, two analysts sort the queue by hand and gather the details of each incident across several consoles.
- For a financial client, time is critical. They must notify the New York regulator within 72 hours after establishing that an incident has occurred, including when the incident affects a provider such as Watchbill.
- An agent takes each alert as soon as it drops. It groups those that relate to the same event and adds the context of the affected workstation and account. Classification follows the rules written with the analysts.
- For each probable incident, the agent prepares the case file, with the timeline, affected systems, indicators observed and open questions. For a financial client, it adds the facts the client will need if they must notify the regulator.
What remains with the analysts
Analysts decide each response, from isolating a workstation to blocking an account, and the agent never acts alone on a system. The client alone decides to notify the regulator.
- Time from alert to initial handling
- Alerts processed by analysts each night
- Incident case files complete on first review
- Time taken to deliver the file to the client