Case study — Lead to Order
Don’t automate the waste.
Lead to Order runs through the same six steps in every company: capture, qualify, specify, price and quote, negotiate and approve, order and handover. AI agents can make it several times faster, but only in the right order.
A case by Tom Mekenkamp of truck8.ai and Dagmar Ritterbex of The Lean Kitchen.
Follow one fictional company from challenge to impact:
Read a day in the lifeDownload the case (PDF)
Scroll, or use the arrow keys to step through. Press F for full screen.
Lead time, one hospital tender
24 working days
26 hours of people’s work
- Waiting
- People’s work
- Agents, minutes per step
- A person decides
- Nobody decides
Illustrative figures for a fictional company. Under each step: working days the case waits, then hours of people’s work.
Challenge
A tender takes 24 working days. Three of them are work.
From publication to first order, a hospital tender at Aurelis waits for the weekly portal check, for supply planning, for a meeting slot, for finance and for three sign-offs in a row. With a bid window of five weeks, there is no slack, and mistakes surface after the deadline.
The biggest wastes are waiting, re-typing between systems, and specialists doing admin.
Approach, 1 of 3
The trap: automate the old process
Add an agent to every step and change nothing else. The tender is found the same morning and the brief is ready in an hour. It still waits for the go/no-go meeting, for finance’s Thursday slot and for three sign-offs in turn.
24 days become 18. Same waste, faster.
Approach, 2 of 3
Lean first: take out the waste
No AI yet. The tender desk checks the portals daily. Go/no-go moves into the morning stand-up, with supply planning in the room. Margin rules are agreed once, so finance only sees exceptions. One approval pack goes to all approvers at the same time.
24 days become 9, with the same people doing the same reading.
Approach, 3 of 3
Then agents, with decision points
Now agents take the reading, checking and drafting, on a process that no longer makes them wait. People keep four decisions: go/no-go, regulatory sign-off, final price and submission. Spotting tenders and booking standard orders run on rules.
9 days become 5.
Impact
What Aurelis gets
Lean first, then agents, against the classic process.
Impact: the guardrail
Could it be faster? Yes, if nobody decides.
Let the agents decide alone and the tender is done in under a day. It also means five commitments nobody signed: a bid on a lot the plant can’t supply, a price below cost locked in for four years, a pack size that isn’t registered, a bid nobody approved, and a short-dated batch shipped to hit a date.
A few days faster is not worth that. Automate what you can undo. Sign what you can’t.
A day in the life
One tender at Aurelis Pharma
A purchasing group of seven hospitals publishes a four-year tender for injectables: seven lots, five of them molecules Aurelis makes. The bid window is five weeks.
The notice goes live on TenderNed. Nobody sees it.
The tender watch agent matches the notice to five Aurelis molecules and opens the opportunity with the 150-page tender attached.
The notice sits on the portal.
The tender analysis agent turns 150 pages into a checklist of 212 requirements, flags the penalty clause for shortages, and checks four years of volume against the production plan. Lots 1 to 4 fit. Lot 5 needs a second shift from year two.
Still unseen.
Go / no-go, signed by the head of tenders. At the morning stand-up she reads the one-page brief: go on lots 1 to 4, no-go on lot 5. Six minutes, and the reasoning is on record.
Still unseen.
The specification agent maps lots 1 to 4 to registered products. Lot 3 asks for a pack of 20 vials; Aurelis is registered for 10 in the Netherlands. It drafts a clarification question for the hospitals.
Still unseen.
Regulatory sign-off, by the head of regulatory affairs. She confirms the pack-size gap and sends the question, well before the clarification deadline.
The weekly portal check. The tender desk finds the notice and forwards it to supply planning: “Can we do this?”
The pricing agent proposes a price per lot within the legal maximum, with three years of public award notices for these molecules and the margin at each price point.
Supply planning has a full inbox.
Final price and submission, signed by the commercial director and the managing director. The commercial director raises lot 2 by 3%: Aurelis is the only bidder with stable supply there. The reason is recorded with the price. The bid goes out four weeks early.
Supply planning answers: probably, but they need volumes per lot. Someone starts reading the 150 pages. The go/no-go meeting is set for Thursday, the first slot everyone can make. The decision: go on all five lots. Nobody notices that lot 5 needs a second shift.
Bid submitted. The tender desk is on the next one.
Regulatory checks each lot by hand and finds the pack-size gap on lot 3 on Wednesday. The deadline for clarification questions passed on Monday.
Prices are built from last year’s spreadsheet. Finance reviews the margin on Thursday.
Compliance, legal and the managing director approve by email, one after the other. The bid goes out two days before the deadline, lot 5 included, lot 3 with the wrong pack size.
Automate what you can undo. Sign what you can’t.
Every step gets three questions: can the decision be undone, does it commit the company, and does it need judgment? Where a person decides, the decision needs a named owner, the evidence behind it, and a record you can audit.
How we build it
Project Atlas
The rule in this case is the rule Atlas is built on. Atlas is the agent-first software we are building at truck8.ai, and it treats every commitment the way this case says it should.
More about Project AtlasAgents draft
Agents read, check and draft. Fast, and never final on their own.
People sign
A named person signs every commitment. No agent can sign on anyone’s behalf.
Evidence stays
Every decision keeps what it was based on, in a record you can audit months later.
Lean first, then agents
- 01
See the waste
Map your current Lead to Order value stream and put lead time next to work time. In most companies, that gap is the whole story.
- 02
Remove it
Cut approval layers that add nothing, agree on one source of truth for the customer’s request, and stop re-typing between systems.
- 03
Add agents, with decision points
Let agents do the reading, drafting and checking. Use the three questions to decide, step by step, where a person signs.
About this case
Aurelis Pharma and Werf Noorderhaven are fictional companies. All figures are illustrative: they show the pattern, not a prediction of your results.