By Sagar Shankaran, Founder of CallSphere
Minimum pitch, board size, class, ITAR status: the eleven facts a buyer's assistant needs before your contract assembly shop makes the shortlist in 2026.
Key takeaways
Go and open it. Most contract assembly websites say some version of the same four things: quality is our passion, we are ISO certified, we offer turnkey and consigned, we treat every customer like a partner. Somewhere there is a photo of a reflow oven that is not your building. What is almost never on the page: maximum board size, minimum pitch, whether you run leaded and lead-free on separate lines, whether you are ITAR registered with the State Department, and whether you will take a 25-piece build at all.
For twenty years that was fine, because the reader was a sourcing engineer who would call and ask. In 2026 the first pass through your page is often not a person. It is an assistant working for that engineer, told to come back with five assemblers who can do Class 3 with 0.4 mm pitch BGAs at 500 pieces a year, reading forty capabilities pages in the time it takes you to walk to the SMT floor.
Being findable by another company's AI means publishing the specific, checkable facts about your line — board size, pitch, class, certifications, minimum order, turn times — in plain text on a page a machine can read, so that when a buyer's assistant builds a shortlist, your shop survives the filter instead of being silently dropped.
Watch a hardware startup's operations lead do this. Rev B board, 214 unique parts, a 0.4 mm pitch BGA and two QFNs on the bottom side, 500 boards for the first production run, conformal coating because it goes in an outdoor enclosure. They do not open ThomasNet and start dialling. They describe the job once, and an assistant gathers candidates — directories, assembler websites, IPC member listings, the instant-quote portals — and returns a table.
The columns: minimum pitch, maximum panel size, IPC-A-610 class, J-STD-001 certified operators, AS9100D or ISO 13485, ITAR registered, X-ray inspection in-house, conformal coating types, prototype turn, minimum order value, location. Every shop that did not publish a value gets an empty cell. Empty cells lose — not because anyone decided to exclude you, but because the reader has five shops with full rows and no reason to chase blanks.
The uncomfortable part: you never see it happen. No bounced email, no "thanks, we went another way." The RFQ simply never arrives, and your month looks the same as always.
Hear it before you finish reading
Talk to a live CallSphere AI voice agent in your browser — 60 seconds, no signup.
Here is the list a buyer's assistant is trying to fill in, and every one is a fact you already know: smallest component placed in production (01005 or 0201), smallest BGA pitch (0.4 mm, 0.35 mm), maximum and minimum board dimensions and whether you need a break-away rail, board thickness range, whether you run a dedicated leaded line for legacy defense work, in-circuit test versus flying probe versus functional test only, conformal coating chemistries (acrylic, urethane, silicone, parylene), whether box build and cable assembly happen under your roof, standard and expedite prototype turns, and minimum order value.
Add the certification facts with dates: AS9100D certificate number and expiry, ISO 13485 registration, ITAR status, whether you hold CMMC Level 2 or are working toward it, a certified IPC trainer on staff, and your counterfeit-avoidance program under AS6081. Each is a filter somebody is running. A defense prime's buyer filters on ITAR and CMMC first. A medical OEM filters on ISO 13485 and Class 3. A startup filters on minimum order value, because half the shops they called last time would not touch 500 pieces.
flowchart TD
A["Sourcing engineer: 500 boards, Class 3, 0.4 mm BGA, conformal coat"] --> B["Assistant gathers assembler capability pages and directories"]
B --> C{"Does the shop publish pitch, board size and class?"}
C -->|No| D["Dropped from the table, no email ever sent"]
C -->|Yes| E{"Certifications current and minimum order under 500 pieces?"}
E -->|No| D
E -->|Yes| F["Shop lands on the five-row shortlist"]
F --> G["RFQ packet arrives: Gerbers, BOM, quantity breaks"]
G --> H["Your quoting desk answers a request you never had to chase"]
Two things landed close together. Assistants that take a goal and work across apps and files for hours — Claude Cowork since January, ChatGPT Work since 9 July 2026 — put real research capability in the hands of a non-technical operations person. And agent-to-agent conventions matured enough that a business can publish a machine-readable description of what it does and what it will accept, so the assistant on the other side is not guessing from a marketing page.
What this means for you is dull and entirely within your control: the shortlist is built from published facts, not from who answered the phone. In 2024 an assistant asked to find contract assemblers returned the biggest names and the best search rankings. In 2026 it returns a filtered table, and the filter runs on specifics. If the specifics are not on your site, no amount of relationship-building fixes a filter you were never in.
There is a second effect worth noting: the RFQs that do arrive are better qualified. If your page says you do not run boards over 18 by 24 inches and your minimum order is $1,500, the jobs that would have wasted your estimator's afternoon quietly stop arriving.
This is not a website rebuild. It is one page, written by your quality manager and SMT process engineer over a long lunch, in plain sentences with numbers in them. "We place 01005 passives and 0.35 mm pitch BGAs in production." "Maximum assembly size 20 by 24 inches." "Three SMT lines, dedicated leaded line for legacy defense assemblies." "AS9100D certified, expires March 2028. ISO 13485 registered. ITAR registered with DDTC." "Prototype turn five business days, 48-hour expedite available. Minimum order $1,500."
Then have your web person publish the same facts in machine-readable form — an afternoon's work — so an assistant reading the page gets clean answers rather than parsing prose. And add one thing most shops lack: a clear statement of what an RFQ package must include — Gerbers, ODB++ or IPC-2581, BOM with manufacturer part numbers, centroid file, fab and assembly drawings, quantity breaks, class and test requirements — and where to send it. An assistant that can see what you need assembles the package correctly, saving your estimator the two-email round trip that costs a day.
Rather than guessing at RFQs you cannot see, measure the ones you can. Illustrative assumptions: 26 RFQs a month, 9 of them a poor fit — board too large for your rails, quantity below your minimum, a Class 3 requirement you would subcontract, or a leaded process you no longer run. Your estimator averages 2.5 hours before finding the mismatch, at a loaded $86 an hour.
Still reading? Stop comparing — try CallSphere live.
CallSphere ships complete AI voice agents per industry — 14 tools for healthcare, 10 agents for real estate, 4 specialists for salons. See how it actually handles a call before you book a demo.
| Line | Before publishing specs | After |
|---|---|---|
| RFQs per month | 26 | 21 |
| Poor-fit RFQs per month | 9 | 3 |
| Hours burned on poor-fit work | 22.5 | 7.5 |
| Cost of that work per month | $1,935 | $645 |
| Reclaimed estimator hours per year | — | 180 |
| Cost of the work to publish specs | — | about 6 hours, once |
The total RFQ count goes down in this example, and that is the point — the five that stopped arriving were never going to be built here. The 180 reclaimed hours go to quotes you can win. If publishing your specs also brings in one well-matched program a year, the arithmetic stops being close.
Three real cautions. First, do not publish pricing. A price on a page becomes a ceiling, and material is 60 to 80% of the number anyway and moves weekly. Publish capabilities, minimums and turn times, never a per-placement rate.
Second, be careful what you claim. If your page says you hold CMMC Level 2 and you are mid-assessment, you have created a flow-down problem for a prime contractor and a bad conversation later. Certifications get verified. Put the certificate number and expiry on the page and set a reminder to update it — an expired date is worse than no date.
Third, none of this replaces the relationship. An assistant can shortlist you; it cannot decide to trust you with a program carrying FDA exposure. What gets you from shortlist to purchase order is your program manager on a call, a plant tour, a clean first article on AS9102 Form 3, and a customer who has watched you handle an engineering change without drama. The page gets you into the room. It does not close.
It overlaps, but the target differs. Search optimisation is about ranking for phrases people type. This is about surviving a filter: an assistant checking six hard requirements. You can rank first for "PCB assembly" in your state and still be dropped from every shortlist because nobody can tell whether you run 0.4 mm pitch.
Your competitors already know what a shop your size can do — they walked the same aisles at IPC APEX and quote against you every month. The people who do not know are the buyers. In a trade where the win comes down to turn time, class and whether you will take the quantity, hiding the specifics only costs you.
Treat them as a channel, not a threat. They serve the low-mix prototype buyer well, and that buyer often grows into a production program you want. This post is about the jobs those portals cannot price — Class 3, conformal coated, ITAR, box build — the jobs where a shortlist gets built by research rather than by uploading Gerbers to a web form.
Ask every new inbound RFQ one question: how did you find us? Log the answer. Within two quarters you will see whether "our assistant found your capabilities page" shows up, and how those RFQs convert against referrals. One field in your CRM, and the only honest measurement available.
The moment your shop clears the filter, somebody makes contact — often outside your hours, because a startup's operations lead works nights and a West Coast defense buyer calls at 4:30 your time. CallSphere builds AI voice and chat agents for business phone lines and website chat that answer around the clock, ask which class and quantity the caller needs, capture the details for your program manager, and book the follow-up. Getting on the shortlist is the hard part; being unreachable afterwards is the avoidable one.
Written by
Sagar Shankaran· Founder, CallSphere
LinkedInSagar Shankaran is the founder of CallSphere, where he builds production AI voice and chat agents deployed across healthcare, hospitality, real estate, and home services. He writes about agentic AI, LLM engineering, and shipping voice agents that handle real calls in production.
See how AI voice agents work for your industry. Live demo available -- no signup required.
How a buyer's AI assistant screens consulting firms in 2026, why gated PDFs make you invisible, and the nine facts to publish in plain text this month.
A 640-line BOM scrub eats three days of your buyer's week. Here is what splitting the RFQ across several agents does to quote throughput at an EMS shop.
Seasonal spend ceilings, per-person limits, 75% alerts and cost per quote packet: budgeting AI at a contract assembly shop without stalling the quoting desk.
How a plant engineer builds the first list of machine builders in 2026, and the five specs an equipment OEM must publish as text to survive that pass.
Buyer-side software makes the first cut on injection molders. The press list, cleanroom class and certificate dates that decide if the RFQ ever reaches you.
In 2026 the first thirty minutes of an SBA referral happens without you. Publish the credit box facts an assistant needs, or get dropped from the shortlist.
© 2026 CallSphere Inc. All rights reserved.
Made within San Francisco
Watch how CallSphere handles real customer calls, schedules appointments, and processes payments — live.
Try Live DemoBook a DemoCalculate Your ROI