Nine Hours per Custom Quote — Hand the Whole RFQ Packet to Claude Cowork, Not to Your Applications Engineer

Nine hours. That is roughly what a custom quote packet costs an applications engineer at a mid-sized machine builder, spread across four to nine calendar days because it queues behind two live projects and a change order that has to go out first. Not nine hours of thinking. Nine hours of assembling: reading the specification, digging out what the last three similar jobs actually cost, chasing lead times, filling a price build workbook, and writing a scope and exclusions letter that has been written two hundred times before with different nouns in it.

Meanwhile the request itself arrived through a portal with a fourteen-page specification, a piping and instrumentation drawing, a plant utility sheet, a line layout in AutoCAD, and a terms sheet with liquidated damages in it. The plant engineer needs numbers in two weeks because his line comes down in March. You will get him something. Whether it is a real packet or a budgetary number depends on how the week goes.

Count what is actually in the packet before you argue about the hours

Write down everything a finished quote contains at your shop and the number stops being controversial. Machine sized to the throughput and the product. Option selections — drive package, washdown-rated stainless, allergen changeover, spare drive. Cost build against the closest past job. Lead time checked on the items that actually move: gearmotors, servo drives, controllers, stainless plate. General arrangement drawing referenced. Scope and exclusions letter that says who sets anchor bolts, who runs utilities to within five feet of the machine, who pays for travel to the factory acceptance test. Compliance basis stated — the panel built to the industrial machinery electrical standard and listed, guarding to the recognized safeguarding standard, the safety circuit rated at the level the customer's corporate standard demands, sanitary construction if it touches food. Recommended spare parts list. Payment terms and a validity window that is short because steel and tariffs move.

That is not a document. That is eleven small jobs, nine of which do not require an engineer's judgment at all — they require an engineer's access to your files.

Know exactly what your applications engineer does with those nine hours

Two of them are genuine engineering: is our machine the right machine for this product at this rate, and where does their specification conflict with how we build. The other seven are retrieval and formatting. He opens the ERP for job costs on three past machines, opens a folder of old quotes to copy a scope letter, emails a buyer to ask whether the gearmotor is still fourteen weeks, and rebuilds a price sheet because eleven copies of the workbook exist and nobody knows which is current.

The seasonal squeeze makes it worse. October brings the capital flush — customers with use-it-or-lose-it budget who need a quote now and a purchase order before the December cutoff. January brings the budget release wave. And if you exhibited at a fall trade show, the booth leads land as real requests for quote inside the same six weeks. Your quoting capacity is flat and your demand is not.

Know what shipped in 2026, and why the 2024 chatbot could not do this

Two things landed this year that change the shape of the job. Anthropic shipped Claude Cowork on 12 January 2026, and expanded it to mobile and web in July. OpenAI shipped ChatGPT Work on GPT-5.6 on 9 July 2026. Both accept a goal rather than a question: they connect to your apps and files, break the job into steps on their own, work for hours without supervision, and hand back a finished artifact — a completed workbook, a drafted packet, a deck. Both were aimed at people who do not write code, which is the part that matters to you, because it means your inside sales coordinator can drive this and not just your one engineer who likes computers.

The 2024 version could write a paragraph of your scope letter if you pasted the specification into a box. It could not open the folder, read the last three comparable jobs, fill your price build workbook, and put a draft packet in the right place with a list of every assumption it made. That gap — between answering and finishing — is the whole 2026 story.

flowchart TD
  A["Request for quote lands in the shared sales inbox"] --> B["Coordinator states one goal and names the packet"]
  B --> C["Read the spec, the drawing and the utility sheet"]
  B --> D["Pull actual job costs from the three closest past machines"]
  B --> E["Check lead times on gearmotors, drives and stainless"]
  B --> F["Draft scope, exclusions and spare parts list"]
  C --> G["Assembled draft packet in the quotes folder"]
  D --> G
  E --> G
  F --> G
  G --> H["Applications engineer prices it, flags risk, signs"]

Write the goal, not the task list

Here is the difference in practice. A task list says "summarize this spec." A goal says something closer to this:

Build the quote packet for the request for quote in the folder named Midwest Dairy Line 3 Filler. Use our standard scope and exclusions letter. Find the three closest past jobs by rate and product and show me their quoted cost against their actual cost. Flag every line in their specification we do not comply with, and say why. Leave the selling price blank and give me the cost build. Put the draft in the 2026 Quotes folder and list every assumption you made.

The last sentence is the one that makes this usable. A packet with a numbered assumptions list at the front can be checked by an engineer in twenty minutes. A packet without one is a black box he will rebuild by hand, and then you have paid twice.

Change how the job gets assigned, or the machine has nothing to work with

This is the part owners skip. Three things have to change on your side.

First, write down what "finished" means. If a completed packet is defined only by what a fifteen-year employee habitually includes, nobody — human or otherwise — can be held to it. A one-page checklist of the eleven components solves this permanently.

Second, collapse the workbook problem. One price build file, one scope letter, one folder where closed jobs with actual costs live. If three years of job cost history exist only as printouts in a filing cabinet, that is the real project, and no software purchase substitutes for it.

Third, change what you measure. Stop tracking hours spent quoting and start tracking two numbers: the share of requests returned inside five business days, and gross margin at booking. Those are the numbers this actually moves.

Do the arithmetic on quote coverage, not on hours saved

Hours saved is the wrong frame, because you will not lay off an applications engineer and you would be foolish to. The right frame is how many opportunities you can properly contest. Assumptions, illustrative: 180 requests for quote a year, nine engineering hours per full packet, and 1,080 quoting hours available across the group.

Treat every number in that table as an illustration and substitute your own; the sensitivity is the point, not the total. And note the honest counterweight: quoting everything means winning more, and winning more means engineering and floor capacity you may not have in Q1. If your constraint is build capacity rather than quote capacity, this buys you selectivity instead of volume — which is worth just as much, because you can finally chase the jobs with margin instead of the ones that arrived first.

Keep the human on price, risk and the exclusions letter

Selling price is a judgment about this customer, this competitor and how badly you want a machine in that plant. Nothing about 2026 changes that, and nothing changes who reads the commercial terms: liquidated damages, consequential damages, and a performance guarantee committing you to a rate on a product you have never run are the three clauses that turn a profitable machine into a rebuild at your own cost.

Any statement about compliance gets signed by an engineer. If the packet asserts a safety circuit rating or a sanitary design basis, a person with a name owns that sentence. And the phone call to the plant engineer — the one where he mentions the real driver is a March shutdown and a corporate standard nobody put in writing — is worth more than the packet. Faster quoting is only useful if it buys your engineer time to make that call.

Frequently asked questions

Does this have to connect to our ERP on day one?

No, and starting there will stall you for a quarter. Start with a folder: your last thirty quote packets, a job cost export, and one clean price build workbook. Get a usable draft packet out of that first, then argue about connections.

What stops a salesperson pasting a customer's specification somewhere it shouldn't go?

Use business accounts with the controls turned on, not personal ones. Anthropic's 2 July 2026 governance update added org-level and user-level spend limits, alerts, and control over which models people may use — meaning you can see and cap what happens. Write the rule down: customer drawings stay in company accounts, and breaking that is a disciplinary matter, same as emailing a print to your Gmail.

My applications engineer says it will get the machine sizing wrong.

He is right, and you should not ask it to size machines. Ask it to assemble, to retrieve past job costs, and to flag spec lines that conflict with how you build. Sizing is the two hours of the nine that were always his.

How long until the first useful packet?

If your last thirty quotes and one job cost export sit in one place, a week. If they are scattered across five desktops, a month of housekeeping first — and that month is the actual investment.

Many of these requests still start as a phone call at 4:50 on a Friday from a plant engineer asking whether you build a machine for a product you have never heard of. CallSphere builds voice and chat agents that answer that line day or night, capture the details worth capturing — product, rate, container, plant location, timing — and book the discovery call with your applications engineer.