A runbook for fixing the RFQ question frequency structure before collection, so buyer question counts stay comparable, auditable and usable.

投稿日時: 10月 21, 2026

RFQ Question Frequency: How to Structure the Data Before You Collect It

RFQ Question Frequency: How to Structure the Data Before You Collect It

Procurement teams rarely lack RFQ correspondence. They lack a way to count it. A folder of supplier questions reads like evidence until you try to ask how often a theme appears, across which projects, raised by which discipline, and answered from which source. Every attempt to build an rfq question frequency structure hits the same wall: the questions were captured for action, not for counting, and nobody agreed the unit, the tags or the denominator before the first file was sent.

Direct answer. This runbook fixes the data structure before collection begins. It covers the unit of analysis, the question bank, the tagging axes, the row-level field set, the answer formats, the dry run and the version freeze, then shows how to compute rfq question frequency from the frozen data. Every figure here is your own planning assumption computed from your own records. There are no industry averages in this piece

Why the count is only as good as the schema

Frequency is a property of a defined population. Before you can say a question is asked often, you must be able to name what was counted: which RFQs, which revision, which respondent, which appearance. Two teams can build opposite tables from the same correspondence, because one counted every appearance and the other counted every distinct question per cycle. Neither is wrong. Set side by side, neither is usable. Every commercial claim here has to be checked against the smart door locks.

So the schema comes first. This is the sequence the runbook walks through.

flowchart TD
    A["Agree the unit of analysis"] --> B["Build the question bank from existing records"]
    B --> C["Tag every question on six axes"]
    C --> D["Declare answer formats and the unknown path"]
    D --> E["Dry run on two real RFQ cycles"]
    E --> F["Freeze and version the structure"]
    F --> G["Collect rows against the frozen sheet"]
    G --> H["Compute coverage breadth depth and open rate"]
    H --> I["Feed the worksheet the management view and the supplier conversation"]

Each step below carries an output somebody else can pick up. A step with no output is not finished.

Step 1 — Fix the unit of analysis

Decide what one row of your question frequency data structure represents, and write that decision into a short note you keep with the sheet. Four bases are common, and each answers a different question.

Counting basisOne row meansIt answersIt hides
Appearanceone occurrence in one RFQ filehow much supplier effort a theme createsrepeat asks inside one thread
Distinct question per cycleone question per cycle, counted oncehow many cycles a theme toucheshow often a team asked twice
Distinct question per respondentone question per person per cyclewhich role keeps raising a themevolume within a role
Answer eventone logged responsewhere the effort wentquestions never answered

Pick one primary basis and derive the others from the same rows. Running all four at once and calling the result a frequency is the failure to design out.

Step 2 — Build the question bank from records you already hold

You do not need new research to start. Your existing records already contain the questions; they lack structure. Mine four sources: supplier correspondence and clarification logs from recent cycles, bid levelling notes, internal site-team emails, and the door data that came back from surveys such as the Office Smart Lock Site Survey or the Smart Lock Retrofit Site Survey.

Copy each question into the bank as written, keep the verbatim text in one field, and assign a provisional ID. Do not merge near-duplicates yet. Merging is a judgement call, and judgements belong after the counting framework is agreed.

Step 3 — Tag each question on six axes

The rfq question taxonomy is simply the tag set, and six axes are enough for most programmes. Each axis earns its place by answering a decision you will actually face.

AxisClosed value listDecision it drives
Themedoor fit and prep, credential and access, electronics and power, mechanical hardware, commercial and trade, compliance and documentation, logistics and customs, service and spareswhich themes need a standing answer
Owning disciplinemechanical, access control, commercial, logistics, legal, project managementwho is asked to answer
Programme stagesurvey, specification, quotation, evaluation, order, handoverwhen the question became urgent
Answer ownersupplier named role, or the role you requirewhere the answer must come from
Evidence typeproduct page field, drawing, sample, third-party document, written confirmation, not yet availablewhat closes the question
Blocking statusgate, clarification, information onlywhether it can hold up a decision

Keep the value lists closed. An open tag list is where the structure quietly collapses: two people spell one theme differently, the counts split, and reconciliation takes longer than collection did.

Step 4 — Define the row structure

Each row is one instance in the rfq data schema, and every column has to earn its place. The field set below covers the questions you will ask of the data: how often a theme appeared, who raised it, what closed it, and what stayed open.

erDiagram
    RFQ_CYCLE ||--o{ QUESTION_INSTANCE : contains
    QUESTION_INSTANCE }o--|| QUESTION_BANK_ENTRY : coded_as
    QUESTION_INSTANCE ||--o{ RESPONSE : answered_by
    RESPONSE }o--|| EVIDENCE_SOURCE : cited_from
    RFQ_CYCLE }o--|| PROGRAMME : belongs_to
FieldTypeFilled byPurpose
rfq_idtextyouties the row to a cycle
cycletextyouseparates revisions of one RFQ
project_reftextyoukeeps programmes separate
date_raiseddateyou or respondentshows sequence
respondent_rolelistrespondentseparates breadth from volume
source_recordlistyouwhich document carried it
question_idtextyoustable pointer to the bank
verbatim_textlong textrespondentground truth, never overwritten
theme_taglistyouthe counting dimension
discipline_taglistyouroutes the answer
stage_taglistyoushows when it appeared
answer_ownerlistyouwho must close it
answer_formatlistyouhow the answer must be written
answer_textlong textrespondentthe answer as given
answer_sourcelistrespondentwhere it came from
unknown_reasonlistrespondentwhy it stayed open
captured_bytextyouaudit trail

The procurement question log built on this field set is reproducible by a colleague who was not in the room when the rules were set.

Step 5 — Set the answer format and the path for an unknown

Every question needs a declared answer format, because free text cannot be counted. Use a small fixed set: yes or no, a number with its unit, a document reference, a per-SKU list, a drawing dimension with tolerance, or a named route to the answer. Add one allowed answer to every question: not yet available. That cell is the most valuable one in the sheet, because a question that was asked and could not be answered is a finding about your own documentation, not about the supplier.

Ask for specifications against a named configuration and read them from the dated product page for the specific SKU rather than from a portfolio average. Write credential questions narrowly, as which media a specific model reads, and keep ecosystem names out of the record until they have been verified for that model.

CONFIRM-PER-SUPPLIER: any lock specification, credential capability or compatibility value is confirmed on the dated product page for the specific SKU, not taken from this article.

CONFIRM-WITH-LOCAL-LEGAL: if the log records named individuals, retention and lawful basis for that record are your local legal question.

Step 6 — Dry-run the sheet on two real cycles

Before the structure goes live, run it against two recent cycles you already hold. Two responders work independently and cold, with no coaching. You are testing the instrument, not the people.

CheckWhat to watchWhat it tells you
Ambiguityone question read two waysthe wording needs narrowing
Double countinga multi-part ask split across rowsone question belongs in one row
Unanswerableno answer owner in reachthe question is out of scope
Loadminutes per respondent observedhours each discipline will spend
Tagging driftone theme, two tagsthe value list needs a boundary rule
Unknown overusemost rows close as not yet availablethe evidence path is missing

Cost the load as your own assumption: questions per respondent multiplied by the minutes you observed on the dry run equals the hours you are asking each discipline to spend. Publish that number to the responding disciplines before the sheet lands, not after the first complaint.

Step 7 — Freeze the structure, then change it only by amendment

Freeze means the field set, the tag lists and the counting rules are versioned and dated on day one. Later changes arrive as numbered amendments with a date and a one-line reason. Two rules follow.

Do not recode history silently. If a tag changes meaning, re-tag the affected rows and record that you did. If a theme splits in two, apply the split to the frozen set rather than to rows collected afterwards.

Keep the raw field intact. The verbatim question text is the ground truth and your codes are a view over it. When a colleague disputes a count, the resolution is to read the raw field, not to argue about the tag.

Step 8 — Compute frequency from the frozen data

Counting starts only now, and the numbers are your own, computed from your own records.

MetricComputed asWhat it does not tell you
Coveragecycles containing the theme ÷ cycles in scopehow hard the theme was to close
Breadthrespondents raising it ÷ respondents in scopehow often each respondent asked
Depthmean appearances per cycle across all cycles in scopewhether the asks came from one role
Open rateinstances closed as not yet available ÷ instances in scopewhether the evidence will arrive
Effortinstances logged ÷ cycles in scopethe value of closing them sooner

Write each formula in the same note that holds the unit of analysis, because every metric can be recomputed from the same rows under a different rule. A reviewer who disputes a figure should be able to reproduce it from the frozen file without asking you anything.

Six failure modes to design

  • The answer counted as the question. A response from the supplier is not a question raised, and mixing them inflates every rate.
  • Synonyms merged after counting. Combining two tags once totals exist changes the totals without changing the data.
  • Unknown read as no. A missing answer is a documentation gap pattern, not a negative finding about the supplier.
  • One multi-part question counted twice. Split questions ask for splitting at the source, so a row stays a single request.
  • Weighting applied after the fact. Reweighting a frozen set by project size turns the table into a different table.
  • Supplier clarifications counted as buyer demand. The question that raised a clarification and the clarification itself are two records.

Where the finished data goes next

The dataset is an input to four other jobs, each with its own format: the recording sheet in RFQ Question Frequency: Worksheet for Recording It Properly, the management view in RFQ Question Frequency: Reporting Format for Management, the boundaries of the data in RFQ Question Frequency: What It Can and Cannot Tell a Buyer, and the handover in RFQ Question Frequency: Turning It Into a Supplier Conversation.

One warning travels with the file. Frequency is not priority. A theme in every RFQ can be a routine fact, and a theme in two RFQs from two disciplines can be a blocker. Read the interpretation and supplier-conversation pieces before you take any count into a meeting.

Closeout checklist

  • The unit of analysis is written down, dated, and attached to the sheet.
  • The question bank is populated from existing records with verbatim text preserved.
  • Every question carries a theme, a discipline, a stage, an answer owner, an evidence type and a blocking status.
  • Every question declares an answer format and allows not yet available.
  • The dry run is complete on two real cycles and the load estimate has been published to the responding disciplines.
  • The structure is versioned and frozen, with later changes numbered as amendments.
  • The counting rules sit in the same note as the unit of analysis.
  • The finished file has a named owner and a review date.

Handing the structure

If you want this applied to a live programme, send the RFQ template, the question log you already hold, and the door schedule for the sites in scope. We map your questions onto the field set above, mark what your own records already answer and what has to go to the supplier, and return the frozen sheet with its amendment log attached. Start with a quote request and send the documents with it.

カート
このサイトでの体験を向上させるには、Cookieを許可してください Cookieポリシー
Chat on WhatsApp