Now onboarding a limited number of teams Request access
This study is not published yet. See what is live on the blog.
RESEARCH

Most purchase enquiries never reach a person

We asked 2,786 companies to quote equipment for federal contracts and tracked what happened to every request. Most did not get turned down. They were never seen.

Over two months we asked 2,786 companies to quote equipment for federal contracts and tracked what happened to every request. Most did not get turned down. They were never seen. This is the anatomy of that, measured message by message, because the failure is a process failure and processes can be fixed.

For every 100 companies asked to quote a federal contract, 4 send a price and 1 sends a price we can use.

And the companies that route your request onward to a dealer never come back with one at all: 0 usable quotes across 196 conversations.

Who is asking. Celestix builds the sourcing system behind Aeris Ops, a United States federal supplier. We read federal solicitations, work out what equipment they call for, and ask manufacturers and distributors to quote them. We measured this because it is our own problem: every request that stops at a ticket number is a bid we cannot complete. Every request counted here was a real request for a real solicitation we were bidding. Nothing was sent to generate data.
2,786companies asked for a price
606companies where a person answered
431companies answering only by machine
1,703companies that said nothing at all

The demand behind these requests is real and not in short supply. The public award record for this kind of equipment shows 598 awarded solicitations worth $1.9 billion, at a median of 3 bidders each. Nothing here is about a shortage of work. It is about the work and the companies who could do it not meeting.

Where a request actually stops

Counted in companies. One base throughout: every company our enquiry was delivered to, and each share is of that same base.

2,786
Companies the request reached
the enquiry was delivered
100%
1,038
Something came back
any reply, human or machine; delivery bounces excluded
37%
606
A person answered
someone read it and responded
22%
100
A price arrived
3.6%
32
A price we could use as it stood
right item, complete terms
1.1%

1,703 of 2,786 companies produced nothing at all. Not a refusal, not a holding reply. Silence.

What "a person answered" counts. Any point in the conversation, whether they replied to the first approach or a later one. In practice it is almost entirely the first: of the 606 companies where a person engaged, 604 had already replied before we sent any reminder, and only 2 answered for the first time because of one. Nothing here depends on chasing.

The companies that route you onward never come back with a price

83 companies answered by naming somebody else to talk to: their distributor, their dealer, the division that handles it. Across every conversation with those companies, on every solicitation, they produced this many quotes we could use:

0

usable quotes across 196 conversations. Everyone else produced 33 across 1,808.

The obvious objection is that this is circular: a company that refers you elsewhere is telling you it will not quote. That objection is right about half the data. The test is what those same companies did on requests where they did not refer us on.

ConversationsCountA person engagedSent a priceUsable quotes
Where the company referred us onward9888%2.0%0
The same companies, other solicitations9836%3.1%0
Companies that never referred us1,80840%6.2%33

On requests where they were not referring anyone anywhere, those companies still produced no usable quote at all. Their price rate was lower too, but on 98 conversations that difference is not large enough to mean anything by itself. So the tautology does not account for it.

One number in that table is an artefact and should be read as one. The 88% engagement on the referral row is guaranteed: a referral is a person writing back, so that group cannot help but look responsive. Strip those conversations out and the same companies engage at 36%, slightly below the 40% of everyone else. Their responsiveness was never the finding. The zero is.
How strong is this. Everyone else produced 33 usable quotes across 1,808 conversations, a rate of 1.8%. At that rate these 196 conversations would have been expected to produce three or four, and they produced none; the chance of that by accident is about 2.8%. The price-rate difference, 3 against an expected 6.1, is inside the noise and proves nothing on its own. The finding rests on the zero, and at this sample size it should be read as a strong signal rather than a settled fact.

Most of what does come back is not a person

This is the part that hides the problem from both sides. A buyer sees replies arriving and assumes the message got through. Separate the automated ones and the picture changes: 344 companies replied only with an automated acknowledgement that was never followed by anything, and a further 87 replied only with marketing, sent in answer to a request for a price.

22%
15%
63%
A person engaged 606
Machine only 431
Silence, bounce or unclassified 1,749
Sent a price
100
Asked us a question
169
Said they were willing
155
Declined
164
Pointed us elsewhere
18
Automated acknowledgement, nothing after
344
Marketing reply only
87
Something else
1
Undeliverable
45
Nothing at all
1,703

Every company, by the furthest its answer got. A company approached on several solicitations is counted once at its best outcome, so these sum to the full base of 2,786.

A ticket number is not an answer. For 344 companies that was the only thing that ever arrived, which from the buyer’s side is indistinguishable from silence, and from the supplier’s side looks like a request that was handled.

Find out what happens to a request sent to you

When we next have a requirement that matches what you sell, we will send you a real request for a real contract, and then send you back exactly what this study measures: whether it reached a person, how long it took, and what the reply contained.

Ask us to measure yoursYour company name and one email address, nothing else. Results go to you and are never published with your company named. We do not send fabricated enquiries, so this waits for a genuine requirement, which in practice has meant around ten working days. A person handles it, not a system.

The request often arrives at the wrong part of the business

When someone does write back to decline, the reason is frequently that the request should never have reached that desk. 52 companies told us plainly that they do not sell the category at all. Another 102 pointed us somewhere else entirely.

Declined for a reason specific to the job
170
Sent us to a distributor or dealer
102
Do not sell this category at all
52

Counted in companies. A company can appear in more than one row across different solicitations, so these do not sum to a total.

83 of those companies went further and named the specific company we should be talking to. That is the most useful reply in the entire dataset, and it costs the sender one line. Those 83 are the same group the dealer section above is about. Of them, 18 did nothing else on any solicitation, which is why the chart shows 18 whose furthest answer was a referral while 102 appear here.

Three days is the whole window

When a person does answer, they answer fast. Of the 231 human replies we could time exactly, 189 arrived inside three days. A buyer collecting quotes against a closing date has usually finished before a slower supplier has begun.

67
under 1h
42
1 to 6h
52
6 to 24h
28
1 to 3d
17
3 to 7d
25
over 7d

Counted in replies rather than companies: 231 human replies with a measurable interval.

8.7hmedian time to reply
70%of replies inside 24 hours
82%of replies inside 3 days
11%of replies take over a week

Conversations that start and then stop

The failure is not only at the front door. 317 conversations, across 265 companies, began properly and ended without a conclusion: 51 after the supplier had already sent a price, 149 after the supplier asked us a question and never returned for the answer. The median one has been quiet for 23.0 days.

Stopped after asking us a question
149
Stopped after saying they were willing
116
Stopped after sending a price
51

Counted in conversations.

Nobody decided to walk away from these. A question was asked, the thread moved to someone else, and it stopped. It is the same routing problem as the front door, one step further in.

What comes back when it works, and what is wrong with the rest

123 conversations produced a price. 35 could go into a bid exactly as they arrived. The rest needed another round with the supplier, and against a closing date another round often means the quote does not make it.

Arrived after the deadline
23
Missing terms the solicitation required
21
Quoted a different item
17
Model could not be verified
14
Offered used or refurbished stock
8
Other
2
Substituted another brand
1
Withdrew
1

87 reasons recorded across the 88 quotes not usable as they stood. A further 13 were never assessed, which is why the appendix reports 75 judged unusable rather than 88. Counted in conversations; in companies it is 100 that sent a price and 32 that sent one usable as it stood.

What a usable quote contains. The exact model named in the request, or a stated equivalent. Factory new unless otherwise agreed. A unit price and a delivered price. Lead time in days. Country of manufacture. A validity date. That is the entire list, and a quote carrying all of it goes straight into a bid without a second exchange.

Some categories answer and some do not

What was being boughtConversationsA person answeredMachine onlySent a price
Medical equipment92240%32%37
Refrigeration and food service22849%32%30
Vehicles and grounds21833%56%12
Communications and security15355%29%7
Facilities and construction12743%36%13
Furniture and fixtures8556%39%3
Material handling8427%32%3
Laboratory2662%31%1

The spread is wide and it does not track how technical the product is. It tracks how the industry handles incoming mail. The rows with the lowest human share and the highest machine share are the sectors that sell through dealer networks, which is the same mechanism as the section above.

The rest of the numbers

Everything else we measured, including the parts that are unflattering to us. Published in full because a study that only reports what suits the author is worth nothing.

The four bases in this study. Companies (2,786): distinct businesses our enquiry was delivered to; every headline figure uses this base. Conversations (2,004): one company on one solicitation. Replies (231): individual human messages with a measurable interval. Solicitations (207): the federal contracts we were bidding. A company appearing on several solicitations is counted once at its best outcome, and figures counted by company are never added to figures counted by conversation.

Chasing a silent supplier barely works

We sent 146 follow-up messages across 118 conversations that had gone quiet. They produced a reply 6 times, which is 5.1%. Whatever recovers a stalled conversation, a reminder is not it.

Conversations we followed up
118
Replied after the nudge
6
Still silent after the nudge
112

What competing quotes do to a price

29 solicitations drew a price from more than one supplier, but most of those prices are not comparable: one supplier quotes a single line and another the whole lot. Restricting to complete quotes for the right item leaves 8 solicitations where a like-for-like comparison holds. Across those the gap between cheapest and dearest had a median of 27.8%.

This is the least robust figure here and it deserves the warning: 8 requirements is a small sample. The same calculation across every priced reply returns 101.2%, which is not a measure of anything except the fact that quotes differ in scope.

A hypothesis we tested and rejected

Two findings look like they might be one: 344 companies answered only with an automated ticket, and companies that route requests to dealers produce no usable quotes. The tempting chain is dealer sales model, therefore automated intake, therefore a ticket, therefore no price. One sentence explaining most of the dataset.

It does not hold. Testing it against the referral signal would be circular, because a company that never wrote back substantively cannot have named a dealer, so we tested it against our own manufacturer and distributor labels instead, assigned when we research a company and independent of how it replied. They cover 1,328 of the 2,004 conversations.

Sales model, labelled independentlyConversationsA person answeredTicket onlyReferred onwardSent a price
Manufacturer31149%21%12%5.1%
Distributor86941%21%4%6.3%

The ticket-only rate is the same for both, 21% against 21%. Whatever produces an automated acknowledgement and nothing after it, the dealer model is not it. What manufacturers do differently is refer, 12% against 4%. The tidier story was available and the data did not support it.

Where the data itself is imperfect

IssueCountWhat it means
Conversations with no stored thread282The exchange happened but the message log is incomplete, so timing cannot be measured for it.
No measurable first-contact time388We know the outcome but not the interval, so these appear only in the outcome counts.
Priced but flagged unusable75A price arrived that could not have gone into a bid without another exchange.
Status and message label disagree5Recorded as quoted while the message itself reads as a refusal. Counted as quoted.
Companies whose only inbound was a bounce44Excluded from every reply figure, because a bounce is not a reply. One further company bounced on one solicitation and stayed silent on another, giving the 45 in the outcome chart.

Method

Every request counted here was a real request for a real solicitation we were bidding. Nothing was sent to generate data, and no company was contacted for the purposes of this study.

The response figures come from our own stored message record for 1 June to 1 August 2026, covering 2,786 companies and 2,004 conversations across 207 federal solicitations. The award figures come from the public record of awarded federal solicitations: 5,175 line items across 598 solicitations, fiscal years 2019 to 2026. The two are never combined in a single figure.

  • A company counts as reached when the enquiry was delivered to it. Companies where the approach never got through are excluded entirely rather than counted as silent.
  • A reply counts as human when a person wrote about the request. Ticket receipts, autoresponders, newsletters and delivery bounces are counted separately and never as engagement.
  • Reply times are measured from our message to the next message from them, so each is an exact interval. That limits the timing figures to the 231 human replies where such an interval exists.
  • Silence is only called silence after three days.
  • Both datasets lean heavily toward the Department of Veterans Affairs, which is most of the award record and about nine in ten of our own requests. Read this as VA-centred medical and facilities procurement rather than as all federal buying.
  • This is one buyer over two months. A supplier with an existing account would treat a familiar customer differently, and these numbers would look better.

Message classifications were produced twice, independently, by different methods and compared: agreement on 698 of 740 messages (94.3%). Disagreements resolve to the more conservative reading. Solicitation categories were reconciled against the federal supply code in the title, which agreed with the independent assignment on 89.9% of the 119 that carry a usable code.

If you sell equipment, we would rather reach the right person

Tell us what you sell and who should receive a request for a price. That single line is the fix for most of what is described above.

Tell us what you sell

This study reports real outcomes from real purchase enquiries sent while bidding live federal solicitations. No supplier is named anywhere in it, and no company is identified as having failed to respond. Underlying figures and method questions: contact@celestix.ai.