Skip to content
Offra
FRSubscribe

The report

One tender. One document.
Five decisions inside it.

The report is about one notice and about your company: the price the contract should go for, your odds at the amount you file, and the things to settle before you bid. Everything else in the document is there to justify it.

This is what your subscription generates — 50 full reports a year, on the notices you pick out of the feed. When the answer is yes, Offra prepares the bid — same subscription, same notice.

SubscribeorBook a 20-minute demo

See a sample report

The forecasts in this example are real; only the company names are invented.

How it works

01

The notice is published

We read every Quebec public tender, the day it appears.

02

It arrives scored in your feed

Price, competition, likely competitors — then a verdict for you.

03

You open the report

On the ones that deserve it. In minutes, without leaving the feed.

04

You prepare the bid

The documents read, the requirements tied to their clause, and the list of what is left to produce.

We ask for neither your prices, nor your costs, nor access to anything: your company name is enough to find your own history, because it is already public. Whatever you choose to upload on top replaces nothing — it is an addition, and it is described on its own page.

Choosing the notice

The report is about one notice. You choose it here.

The feed is sorted by verdict, then by nearest deadline, and shows only favourable and mixed notices by default. Filter by region, open the one that matters, generate the report.

Offra portal · tenders
Regions · all

5 notices

  • Réaménagement d'une cour d'école

    Centre de services scolaire des Hauts-Vents · Montréal

    Favourable66 out of 100

    Closes in 2 days

  • Pavage de rues — secteur nord

    Municipalité de Val-Ferrière · Laurentides · Published estimate $1.1M

    Favourable68 out of 100

    Closes in 9 days

  • Réfection de la toiture — centre communautaire

    Ville de Sainte-Ombre · Montérégie · Published estimate $840K

    Favourable71 out of 100

    Closes in 12 days

  • Services d'ingénierie — réseau d'aqueduc, phase 2

    MRC du Lac-Ferrand · Capitale-Nationale

    Mixed55 out of 100

    Closes in 6 days

  • Remplacement de conduites d'eau potable

    Ville de Roche-Blanche · Chaudière-Appalaches · Published estimate $460K

    Mixed51 out of 100

    Closes in 21 days

The filters above work. The tenders and public bodies named are fictional; the layout, the sort and the filters are the portal's.

What you get

One document per tender. Five decisions inside.

Not another dashboard. A report you read before committing an estimating team, and reopen the night before filing. The extracts below come from the example you can open in full.

01

Bid or pass

A score out of 100 and five factors shown with their weight: your fit with the sector, the competition to expect, your usual price position, your relationship with the buyer, the size of the contract.

  • This is not a model: nobody publishes whether a company should have bid, so there is nothing to learn from
  • Unfavourable factors are shown as plainly as the others
  • When the information is missing, the grid says “not enough signal” instead of a middling number

Mixed

63.7 out of 100

  • Sector fit · weight 30261 bids in this family, 32 won
  • Expected competition · weight 204.5 expected, 4% chance of being alone
  • Price position · weight 2014.8% above the winner, over 1,451 observations
  • Buyer relationship · weight 158 bids with this buyer, 5 won
  • Contract size · weight 15$514,000, below your usual range
Extract — Recommendation section

02

Where to put the price

The winning price range in dollars: the 50% interval, the 80% one around it, and no single amount — the report publishes none. When the buyer has published an estimate, it is placed on the same scale and the gap is quantified — but a little over one notice in two publishes none, like the example alongside, and that is where the range matters most.

On 1.7% of notices there is no range at all. When the buyer publishes neither an estimate nor usable specifications, the interval we could produce spans a factor of about twenty — an order of magnitude, not a range. The report says so, shows the comparable contracts instead, and does not show a price to win either, since that is derived from the same distribution.

  • The intervals are measured on awards the model had never seen, not rounded upward
  • A wide range is an admission of uncertainty, not a defect
  • A range on 24% of notices is flagged as wide before you read its edges
  • The price to win is shown only where price genuinely decides the award
$197K$1.2M

50% range from $252K to $760K, inside the 80% one. The report publishes no single amount.

Extract — Price range section

03

How many across the table

The number of bidders to expect, the probability of being the only one to file, and the number of entrants no history can anticipate.

  • A lightly contested bid is not prepared the way a scrum of eight is
  • The chance of being alone is the forecast that changes a pricing strategy most
  • The uncertainty is shown, not hidden behind an average
Attendus
4.5
Chance of being alone
4%
Unpredictable entrants
≈ 0.0

Measured mean error of 1.9 bidder. The “unpredictable” ones are the companies we expect to file without being able to name them, for want of a record in SEAO.

Extract — How intense the competition is

04

Who, by name

The companies most likely to file, ranked by probability, with what qualifies them: already a supplier to this buyer, active in the last twelve months, specialist in the family. And for each, their record against you.

  • The list is for recognising names, not for predicting the outcome
  • Your own rank in the assessed pool is shown
  • The share of bidders beyond reach is written beside the list
  • 1AMÉNAGEMENTS MIRVAUX[object Object] · [object Object] · [object Object]18%
  • 2EXCAVATION DUFRESNOY INC.[object Object] · [object Object]17%
  • 3LES PAVAGES MONTVAL INC.[object Object] · [object Object]14%
  • 4ORVILLE QUÉBEC CONSTRUCTION INC.[object Object]13%

Out of 656 companies assessed. On average 1.0 of these names will actually file a bid — the list is for recognising who circles this buyer, not for predicting the exact roster. Noms fictifs.

Extrait — section Concurrents probables

05

How this buyer awards

What they usually award, the gap between their estimates and their real contracts when they publish them, their share of repeat suppliers, their bidding windows, how often they cancel.

  • No forecast in this section — only their own public record, aggregated
  • A buyer who systematically awards below their estimate is not bid the same way as another

Ville de Saint-Lazare

Notices published
326
Bidders (median)
1
Repeat suppliers
32%
Notices cancelled
1.2%
Bidding window (median)
21 jours
Extract — Buyer section. No forecast here: these are the figures from this buyer's own public record.

And the report continues after the purchase.

Once the contract is awarded, the real outcome is added: the price obtained against our range, and who actually bid. The version you bought stays readable exactly as it was — which is also the only honest way to let you check whether we were right.

See a sample report

Optional

Add the specifications. The report tells you what is in them.

The specifications, the pricing schedule, your past bids: you can upload them against a report, and we pull findings out of them tied to their page — each one labelled according to whether it is quoted verbatim or inferred. These findings do not move the price range, and we explain why rather than letting you believe otherwise.

What we read in your documents →

What the report contains

Four models trained on the full history of Quebec public tenders, then evaluated on notices published after their training date.

The report's six sections, in its order and in its terms — so that nothing changes wording between this page and the document you buy.

  • A page that decides

    The price the contract should go for, your odds at the amount you file, and the things to settle before you bid.

  • A reasoned recommendation

    A weighted grid over five factors, each shown with its weight. Not a model: it refuses to rule when the signal is not there.

  • A price range, not a number

    Calibrated: it holds the awarded price half the time for the tight range, 80% for the wide one. No single amount is ever published — and on 1.7% of notices, no range either.

  • How many companies will show up

    The number of bidders to expect — mean error of 1.9 company — and your probability of being the only one to file.

  • The likely competitors, by name

    With their probability and their record against you. About 43% of the real bidders appear in the first ten names — 44% excluding companies with no SEAO history.

  • The report continues after purchase

    The real outcome is added once the contract is awarded. A report bought on its own stays readable indefinitely; a report generated on a subscription, as long as the subscription is active.

The range, read backwards

The model predicts the winning price. At a given amount, in what share of outcomes do you come in under it?

01-020-26

Travaux de resurfaçage de voies de circulation pour l'année 2026

Ville de Saint-Lazare · C02 — Ouvrages de génie civil · Montérégie

Example
50% range : $251,845 – $759,979. 80% range : $197,490 – $1,217,813. Your amount : $400,000.$252K$760K$197K$1.2M$400K$200K$500K$800K$1.1M

Drag to place an amount.

  • 50% of cases
  • 80% of cases
  • Your amount
Your amount
$400,000
Where this amount sits
50% range
You come in under the winning price
61 %des cas

Inside the tight range: half of comparable contracts settle here. This is the ground the decision is fought on.

The lowest bid wins 95% of the time on this kind of contract, which is the only reason this reading means anything here. This is not a price recommendation. It is the distribution of the winning price, read backwards — it knows nothing of your costs, your capacity or your margins.

This buyer published no estimate, which is true of a little over half of public tenders — and the half where forecasting is hardest. That is why the example is this one. The notice is a real SEAO notice; the forecasts in this example are real; only the company names are invented.

How to get it

Included in the subscription, or on its own.

The subscription generates 50 of them a year, on the notices you pick out of the feed — the same document, with no reduced version. A single report can also be bought on its own, at 199 $ CAD + taxes.

The one-off report and the subscription, compared
CriterionOne-offSubscription
Price199 $ CAD + taxes299 $ CAD + tax per month
ScopeOne tender, chosen by youEvery open notice that matches your company
What it coversThe report, and nothing around itThe feed, the report, then preparing the bid
How you get itA 20-minute call, then we produce the reportSubscribe in the portal, or a 20-minute demo first
Full reportsThe one you buy50 per year, then 99 $ each
Bids preparedNone — the report alone50 per year
How long access lastsIndefinitely — the report you bought stays yoursAs long as the subscription is active
PeopleThe link is freely shareable3 seats, then 49 $ per month
Your documentsUploaded against that one reportA company library, reused from one notice to the next

For the one-off report there is no buy button, and that is not an oversight: it is about one specific tender, and the payment unlocks a document already produced rather than starting one. So we have to know which notice it is before there is anything to pay for. The price is published; the next step is a conversation. No-questions refund within 14 days.

Preparing a bid costs a small business between 0.3% and 1% of the contract's value. On a median Quebec public contract that comes to a few thousand dollars — a fraction of the cost of the bid the report might tell you not to prepare.

SubscribeorBook a 20-minute demo

What we cannot see

A model that claims to know everything will cost you a contract.

We know neither your costs, nor your capacity, nor what gets settled outside the public process. What escapes the model is written here and in every report.

Companies with no history
About 3% of real bidders leave no trace in SEAO. We tell you how many unpredictable entrants to expect instead.
Your costs, and theirs
The model sees filed prices. Not order books, and not an urgent need to fill a summer. The range is never a price recommendation.
The smallest contracts
This is our weakest segment. Every report shows the level of information it actually had, rather than the same confidence everywhere.

The models learn up to the end of 2024 and are measured — once — on the following year. Every historical fact is reconstructed as it was known on the day of publication. The method in detail

Questions

What we get asked before signing.

Where does the data come from?
From the Quebec government's electronic tendering system (SEAO), published as open data under CC-BY 4.0. Every report states the date the data was loaded up to and the model version used. We neither buy nor resell any data about companies.
Can my competitors see my reports?

No. A report is produced for your company and is visible only through the link you are given, or to the members of your workspace if you subscribe.

The forecasts themselves contain nothing you told us: they are computed from public data, including your own bidding history, which is already in SEAO. If you upload documents, those are yours — they stay attached to your report or your workspace, are never shown to another customer, and you can delete them. What we do with them is explained here.

How reliable is the range?
Measured, and written into the report: on tenders the model had never seen, the 80% range held the real price 79.4% of the time. The report goes further — it shows you four of those tenders, with the range we would have published and the amount actually awarded, including one the range missed. It also carries an indicator of how much information it had: when the buyer publishes no estimate, the range widens and the report says so, rather than handing you a precise, wrong number. The measurements in detail, by band.
And if you cannot predict the price?
Then we publish no range, and the report says so instead. That is the case on 1.7% of notices — the ones where the buyer publishes neither an estimate nor usable specifications. The interval we could produce there spans a factor of about twenty: that is an order of magnitude, and showing it as a range would invite you to read its edges.

You get instead the comparable contracts already awarded with their real amounts, the observed spread between bidders, the rest of the report — competition to expect, likely competitors, buyer profile, requirements — and the reason, in writing. The price to win is not shown either: it is derived from the same distribution, and keeping the more confident of the two figures while refusing the other would be keeping the wrong one.

If that is not what you came for, the refund is immediate and needs no justification for 14 days. The measured breakdown is published.
Are the companies named really the ones who will bid?
No, and the report does not claim they are. The list is probabilistic, and its shortcoming is measured: out of ten names proposed, one or two actually file on average — simply because an open public tender attracts only 3.6 bidders in total on average. What the list does achieve is containing about 43% of those who really do file — 44% if you exclude companies with no SEAO history, which nothing can predict. So it is there to help you recognise who circles this buyer and this family of contracts, not to guess the opening of the envelopes.
Is the recommendation an artificial intelligence deciding?
No. It is a weighted grid over five measured factors, shown with its weights. There is no training data for “should this company have bid” — nobody publishes that — so there is nothing to learn from. You can contest the grid factor by factor, and every line points back to the section of the report that demonstrates it.
What if the tender changes after purchase?
Write to us and we redo the analysis for as long as the notice is open. The version you bought never changes: new information arrives as a new version, beside the original, which lets you compare before and after the addendum.
Why can I not buy this report directly online?
Because a report is about one specific tender, and the payment opens a document that has already been produced — it does not trigger its production. Until we know which notice interests you, there is nothing to put in a basket. That is what the twenty minutes settle: you name the notice, we look together at what the model says about it, and the report follows if it makes sense. The subscription, on the other hand, is taken online — because it is not about any one notice.

The next step

Should you bid?

The subscription opens the feed on your company and lets you generate a full report on the notices of your choosing. To see it first, the demo opens it on your data during the call.

SubscribeorBook a 20-minute demo

See a sample report

402,630 tenders loaded. Figures measured on open tenders published after 1 July 2025 — 12,076 for the price range, 8,540 for the competition to expect, 17,197 for the likely competitors — with forecasts produced using no information later than December 2024. SEAO data up to 24 July 2026. The sample report is about a real SEAO notice; the forecasts in this example are real; only the company names are invented. The companies named in the extracts on this page are fictional.