Digital Africa11 min read

AI search across 40,000 internal procedures in Montreal (2026)

Mohamed Bah·Fondateur, Kolonell
October 24, 2026
Share:
AI search across 40,000 internal procedures in Montreal (2026)

AI search across 40,000 internal procedures in Montreal (2026)

Digital Africa

The verdict in three sentences

For a 900-employee mid-sized company in Montreal sitting on 40,000 procedures and technical notes, an AI search assistant (RAG) costs 20,000 to 50,000 EUR to build, with running costs of 300 to 1,500 EUR a month. The project pays off as soon as a few hundred employees search for information several times a day, since the observed ballpark is 25 minutes saved per day per regular user. The non-negotiable condition is respecting existing access rights: the assistant must never quote a document the user could not open themselves.

Why classic search fails on 40,000 documents

An industrial company's procedures are spread across SharePoint, an old file server, a quality management tool and scanned PDFs. File names are inconsistent, versions overlap and keyword search returns 300 results with the right one on page 4. A technician searching for "tightening tolerance line 3" does not know the exact title of the procedure.

A RAG (retrieval-augmented generation) assistant indexes document content, retrieves relevant passages by meaning, then writes a short answer citing its sources with link and version. The employee checks in one click. The model only answers from indexed documents, which sharply limits invented answers.

2026 optionSetup costMonthly costRespects rightsHosting
Copilot built into the office suite0 to 5,000 EUR25 to 30 EUR per userYes, within the suite onlyPer vendor contract
Specialised SaaS solution5,000 to 15,000 EUR1,500 to 6,000 EURYes, standard connectorsRegional option
Custom RAG, model via regional API20,000 to 35,000 EUR300 to 900 EURYes, all systemsCanada or EU region
Custom RAG, self-hosted model35,000 to 50,000 EUR900 to 1,500 EUR (GPU)Yes, all systemsYour servers or sovereign cloud
Improved classic search engine8,000 to 15,000 EUR100 to 300 EURYesYour choice

What drives the budget

Three factors move the range. The first is the number of sources to connect: SharePoint alone is simple, four systems with different permissions need one connector each. The second is document quality: 15 to 25% scanned PDFs require character recognition and clean-up. The third is the level of data sovereignty required by management or by your key customers.

Line item2026 rangeShare of budget
Scoping, mapping sources and permissions3,000 to 6,000 EUR12%
Connectors and permission sync6,000 to 15,000 EUR30%
Document processing (OCR, chunking, metadata)4,000 to 10,000 EUR20%
Search interface and citations3,000 to 8,000 EUR16%
Evaluation on 200 real questions2,000 to 5,000 EUR10%
Training, rollout, privacy documentation2,000 to 6,000 EUR12%

The monthly AI cost depends on question volume. For 400 active users asking 8 questions a day, about 65,000 queries a month, the API bill for a regionally hosted model generally stays under 900 EUR. These are 2026 ballpark figures, to be recalculated on your real volumes.

Privacy, security and team trust

An internal assistant handles potentially personal data (meeting minutes, misfiled HR records) and industrial secrets. The minimum requirements are easy to state: hosting and processing in a controlled region, a contractual commitment from the model provider not to reuse data, query logs kept 6 to 12 months, exclusion of sensitive folders from the index and a privacy impact assessment if HR data is involved, as required under Quebec's Law 25. Involve your privacy officer and employee representatives from the start: an assistant perceived as a surveillance tool will not be used.

Mini case study

Need a professional website?

Kolonell builds websites that attract clients, optimized for the Sénégalese market. Free quote in 2 minutes.

Prefer a call back?

Leave your WhatsApp number and a Kolonell expert will get back to you within 1 business day. Free, no strings attached.

You are :

Thomas, CIO of a 900-employee industrial company near Montreal, identified 300 technicians and process engineers who consult procedures every day. Based on 25 minutes saved per day over 210 days, the gross gain reaches 26,250 hours a year. To stay prudent, he only keeps 30%, or 7,875 hours genuinely redeployed, valued at 38 EUR loaded cost: 299,250 EUR a year. The custom RAG project cost 38,000 EUR, plus 900 EUR a month to run, so 48,800 EUR in year one. Even dividing the retained gain by three, payback stays under six months.

FAQ

Can the assistant give a wrong answer?

Yes, it can, hence the need for systematic citations and an evaluation on 200 real questions before rollout. A rate of 85 to 90% correct, sourced answers is a realistic 2026 target.

How long does deployment take?

Allow 8 to 14 weeks from scoping to full rollout, including a 3 to 4 week pilot with 30 to 50 users.

Do we need to clean up the 40,000 documents first?

No, but obsolete versions must be identified. The assistant favours documents tagged "in force", and removing the 10 to 15% of obvious duplicates noticeably improves quality.

Does the data stay in Canada or the EU?

Yes, if you choose a model accessed by API from a regional data centre or a self-hosted model. Self-hosting adds roughly 600 EUR a month.

What happens when a procedure is updated?

Sync re-indexes the document within minutes to hours depending on the chosen frequency. Old versions are removed or tagged as archived.

Let's scope your project. We map your document sources and access rights, then cost a regionally hosted RAG pilot with an 8 to 14 week schedule. Detailed quote within 48 h. WhatsApp +221 77 596 93 33.

Tags:#AI assistant#RAG#Montreal#knowledge base#generative AI#data privacy
Share:

Mohamed Bah

Fondateur, Kolonell

Passionate about digital and entrepreneurship in Africa, Mohamed has been helping Sénégalese businesses with their digital transformation since 2020. Founder of Kolonell, he believes every SME deserves a professional and accessible online présence.