Tell us the workflow.We automate it.

Make your operations run themselves

Pilot on 100 of your real items
Fixed price, paid on outcome
Live in days

M18 › Workers › Bid screening LIVE US-East

Runs inside the tools you already use

In one sentence

M18 builds and operates AI workers: software agents that take over one back-office workflow end to end — reading, extracting, matching, replying, posting — inside the tools a company already uses. M18 sells the outcome, not the technology: a fixed-price build after a measured pilot on the client’s own data, then a monthly run fee sized as a fraction of the salary cost removed. Based in Paris, serving France and the United States, in French and English.

Paris · EU-West & US-East hosting
Pilot first · fixed price
Live in days, operated by M18

The foundation for your autonomous workforce

A workflow goes in. A worker comes out. It runs in your systems, and we operate it for you.

Outcome

You buy the result, not the tool

Priced against the hours removed. Measured on your data before launch. If the worker doesn't do the agreed job, you don't pay to run it.

Integration

Inside your tools, not next to them

Email, ERP, CRM, portals, spreadsheets. The worker logs in where your team logs in. No migration, no new dashboard to check.

Operation

We run it, every day

Monitoring, edge cases, model upgrades, improvements. Your team receives the output. Nothing to babysit.

What a worker looks like in production

Every worker ships with the same surfaces. Your team sees the work; your auditor sees the trail.

unit 07 · online · 99.97 % uptime US-East · EU-West

Worker unitEach worker runs on dedicated, isolated infrastructure in your region, with its own credentials, logs and kill switch. Nothing shared between clients.

mission log · bid-screening · liveSAMPLE

Mission logEvery page read, field extracted and message sent, with timestamps. Replayable. This is what the audit trail looks like.

pilot report · 100 itemsSAMPLE

Agreement with team96%

Sent to review3

Disagreements1

worker = team96

below threshold → human3

worker ≠ team1

acceptance criterion for build: ≥ 95 % · signed

Pilot reportBefore the build, the worker runs on 100 of your real items next to your team. The agreement rate becomes the acceptance criterion, in writing.

confidence routingSAMPLE

threshold0.82

0.001.00

auto · done87no human touch

review · Jordan13reasoning attached

Confidence routingEvery decision carries a score. Below the threshold you set, the item goes to a person with the worker's reasoning. Nothing is silently wrong.

ledger · monthlySAMPLE

people on the workflow8 → 2

hours / month1,280 → 320

cost / month$83,200 → $20,800

gain / month+$62,400

LedgerHours and cost, before and after, updated monthly. The only dashboard your CFO will ask for.

The pilot

Try it on your data before you buy

No demo on made-up data. The worker runs on 100 of your real items next to your team, and you get the agreement rate in writing.

01
Send 100 real items
Emails, PDFs, portal exports — anonymised if you prefer. Under NDA and DPA, signed first.
02
One week, side by side
The worker processes them; your team does the same. We compare every decision, not a sample.
03
The number becomes the contract
You get the accuracy report. It becomes the acceptance criterion of the build — if the worker doesn't reach it, you don't pay to run it.

Start a pilot Fixed price · credited to the build · one week

Modelled scenarios

Workflows that pay for themselves

Public bids · Construction, 240 employees

Screening every bid the company could go after

Before: 8 people reading SAM.gov and state portals daily
After: 2 people reviewing pre-qualified matches
Gain: $520k/year · 3× more bids covered

Vendor invoices · Distribution, 90 employees

Matching invoices to POs and receipts

Before: 3 AP clerks, 2,000 invoices a month, 5-day backlog
After: 1 clerk handling exceptions, same day
Gain: $150k/year · closes in 1 day instead of 5

Inbound leads · B2B services, 35 employees

Qualifying and answering inbound requests

Before: 2 sales reps triaging 60 emails a day, 9h average reply
After: 0 reps triaging, qualified leads in the CRM in 4 minutes
Gain: $120k/year · 2 reps back on selling

Claims intake · Insurance agency, 60 employees

Opening and completing claim files

Before: 4 handlers re-keying emails and PDFs into the system
After: Files opened, completed and routed before the handler arrives
Gain: $230k/year · 2-day faster settlement

Bookkeeping · CPA firm, 25 employees

Collecting and coding client receipts

Before: 3 juniors chasing documents and coding 6,000 lines a month
After: 1 junior reviewing exceptions, clients chased automatically
Gain: $140k/year · 40% more clients per accountant

Modelled with the estimator's assumptions, not client testimonials. Your own numbers are above.

Industries

Built for the work that costs the most

Where a few people spend their days reading, re-keying and chasing. That's where a worker pays for itself fastest.

Construction & contractors

Bid screening, spec reading, subcontractor document chasing.

Distribution & logistics

Invoice matching, order entry from emails, delivery disputes.

Insurance agencies & brokers

Claims intake, policy document completion, client follow-ups.

CPA & law firms

Receipt collection, coding, KYC, contract extraction.

Real estate & property management

Lease data extraction, tenant requests, vendor quotes.

B2B services

Inbound qualification, quote preparation, onboarding paperwork.

Safe from day one

Control and auditability aren't add-ons. Every worker ships with them.

US or EU hosted

NDA & DPA first

Full audit log

Human in the loop

No training on your data You keep the code Confidence threshold you set Replayable decisions Cancel any month SOC 2 / GDPR ready

How M18 compares

A worker — not a chatbot, not a script, not a consulting project

The honest differences between the four ways companies try to remove manual work.

	Chatbots & copilots	No-code automation (Zapier, Make, RPA)	IT agency	M18 worker
Who does the work	You, with help	A script you design and maintain	A team, then a handover	The worker — and M18 operates it
Judgement & exceptions	Only when prompted	Breaks on the first unexpected input	Depends on the spec	Confidence threshold: low-confidence items go to a person
Time to production	Minutes, then it plateaus	Days, then it breaks	Months	Days, after a one-week pilot
Proof before you pay	A free trial on your prompts	None	A proposal	Accuracy measured on 100 of your real items, in writing
Pricing	Per seat, per month	Per task or per bot	Time & materials	Fixed build + monthly run, a fraction of the gain

Meet the workers

Recruit a worker, not a tool

Each one has a job description, reports where your team already talks, and a track record you measure before you commit. Give it a goal, not a prompt.

Bid & RFP screening

Reports in Slack · Teams · email

Reads SAM.gov, state and city portals every morning
Matches notices to your trades, NAICS and territory
Checks deadlines, set-asides, budget, incumbents
Scores and briefs the go / no-go in one page

8 → 2 peoplereplaces

$520k / yrtypical gain

Invoice match

Reports in ERP · AP inbox

Reads vendor invoices from the AP inbox
Three-way match vs. POs and receipts in your ERP
Flags variances, duplicates, bank-detail changes
Posts approved bills, routes exceptions to a person

3 → 1 clerksreplaces

$150k / yrtypical gain

Inbound leads

Reports in CRM · Slack

Reads every inbound request within minutes
Enriches the company, scores fit against your ICP
Drafts the reply with the right clarifying questions
Creates contact + deal in the CRM, assigns the rep

2 → 0 repsreplaces

$120k / yrtypical gain

Claims intake

Reports in Agency system · email

Opens the file from the first email and attachments
Finds the policy, checks the coverage rules
Requests missing documents from the insured
Sets the reserve, assigns the adjuster

4 → 1 handlersreplaces

$230k / yrtypical gain

Month-end books

Reports in QuickBooks · client portal

Matches bank feeds, applies and learns coding rules
Chases missing receipts through the client portal
Reconciles sales tax, payroll, fixed assets
Hands the reviewer a short exception list

3 → 1 juniorsreplaces

$140k / yrtypical gain

Pricing

Priced against the gain

Pilot

100 real items, accuracy report, one week. Fixed, credited to the build.

Build

Design, integration, testing on your data. One-time, fixed, paid upfront. Delivered within days.

Run

Hosting, monitoring, fixes, improvements. Monthly, a fraction of the gain. Cancel any month, keep the code.

FAQ

Straight answers

What kind of workflows work best?

Anything repetitive that lives in emails, documents, portals and spreadsheets: monitoring, extraction, matching, classification, first-line replies, data entry between two systems. If a trained person can do it with rules and judgement, a worker usually can.

Do we need to change our tools?

No. We connect to what you already use. If it has a login, we can work with it.

What if it makes a mistake?

It will, occasionally, like a person does. Every low-confidence item is routed to your team before anything leaves, every action is logged, and the accuracy is measured on your data before launch. You set the threshold.

What if it doesn't work?

The pilot shows you the accuracy before you buy the build. If the delivered worker fails the acceptance criterion agreed after the pilot, you don't pay the run fee.

Do you work in the US and in Europe?

Yes. Workers are hosted in the region you choose (US-East or EU-West), contracts and invoices are in USD or EUR, and we work in English and French.

What is an AI worker?

An AI worker is a software agent assigned to one workflow, with its own logins to your systems. It reads the inputs (emails, PDFs, portals), decides according to your rules, acts in your tools and reports to your team. Unlike a chatbot it does not wait for a prompt; unlike a script it handles judgement, and it escalates to a person whenever its confidence is below the threshold you set.

How much does an AI worker cost?

Three lines: a fixed-price pilot (one week, credited to the build), a one-time fixed build paid upfront, and a monthly run fee sized as a fraction of the salary cost removed. The figures are quoted on the first call, once we know the workflow. There is no per-seat pricing and no long-term commitment on the run fee.

How is M18 different from Zapier, Make or RPA?

Those tools execute a script that you design and maintain, and they stop at the first unexpected input. An M18 worker reads unstructured inputs, applies judgement within a confidence threshold, and is operated by M18 — monitoring, fixes and improvements are included in the run fee.

Where does M18 operate?

M18 is based in Paris and works with companies in France and the United States, in French and English. Workers are hosted in EU-West or US-East, in the region you choose.