Call topic mining
Use this skill once a quarter, when the content plan needs to come from what buyers and clients actually ask rather than from a brainstorm. Produces a ranked list of recurring topics, each with the count of independent calls behind it, where in the buying cycle it comes up, and the one asset that should answer it. Also use for "what should we write about next quarter", "what do prospects keep asking", "content plan from our calls", "which objections keep killing deals".
Christian BeckCOO and Managing Partner, Five FourApplies once a quarter, to a full quarter of sales and client calls at once. Produces a ranked topic list with the evidence behind each one, and the asset each should become.
The play
- Take a full quarter of calls, both kinds. Sales calls say what stops people buying. Client and delivery calls say what confuses people who already bought, which is where the authority content comes from. A quarter of one kind is half a plan. Forty calls is a workable floor; under twenty, wait another quarter rather than ranking noise.
- Extract questions and objections, not themes. The unit is one question asked or one objection raised, tagged with the company, the role and where in the cycle it happened. What qualifies and how to tag it is in references/extraction-rules.md. Theming as you read is the mistake that makes everything come out as "they want more clarity on pricing".
- Cluster the units into topics only after extraction is finished. A topic is a cluster of units asking the same underlying question in different words. The buyer's own phrasing is kept and carried through to the asset.
- Score each topic on four things: how many independent companies raised it, where in the cycle it lands, whether it has cost a deal, and whether existing content already answers it. Thresholds and the ranking matrix are in references/scoring.md. Counting companies rather than mentions matters: one talkative prospect raising a thing five times is one data point.
- Branch by where the topic lives:
- Raised before the first call, in discovery: this is top-of-funnel and belongs in public content.
- Raised at proposal or pricing: this is sales enablement first and content second. A one-pager the team can send beats a blog post nobody reads in time.
- Raised after signing, during delivery: onboarding material, and usually the highest-authority public content the firm can write, because almost nobody publishes it.
- Raised by the same client repeatedly across a project: a process problem wearing a content costume. Fix the process; writing about it will not help.
- Cut to the topics that clear the bar, then map each to exactly one asset and one internal placement. The mapping is in references/topic-to-asset.md. A topic with no owner and no date is a list item, not a plan.
- Compare against last quarter before publishing the plan. Topics that disappeared are content that worked. Topics that persist after being answered are a findability or a sales-process problem, not a content gap, and writing the piece again will not fix it.
What good looks like
- First thing I check that most people skip: the questions asked in the first five minutes of discovery calls, before anyone has framed the conversation. That is the only place the buyer's unprompted vocabulary shows up, and it is the vocabulary the content needs.
- The mediocre version: skimming a dozen recent calls, writing down the four topics that came up in the last two weeks, and calling it a content plan. It is recency, not pattern, and it produces the same four posts every agency in the category is publishing.
- I know the output is right when at least one topic is uncomfortable, meaning the honest answer costs the firm something to publish.
- Second test: the sales team recognises the list immediately and can name the deal each topic came from.
- Red flag: every topic maps to a blog post. Topics that surface at proposal stage almost never want a blog post.
Rules
- MUST extract before clustering. Never theme while reading.
- MUST count independent companies, not mentions.
- MUST include both sales and delivery calls.
- MUST keep the buyer's own words from the transcript through to the asset brief.
- MUST strip client and company names from anything published, keeping role, size and situation.
- MUST compare against last quarter's list before planning.
- NEVER rank a topic raised by fewer than three companies, whatever it felt like on the call.
- NEVER answer a process problem with content.
references/extraction-rules.md
Extraction
One pass through the quarter, pulling units. No clustering, no theming, no deciding what matters. That comes after, and doing it during is what collapses a quarter of nuance into four generic topics.
The unit
One question asked, or one objection raised, by someone who is not on your side of the table.
Q: "How do you handle it when our in-house designer disagrees with the direction?"
Company: 40-person logistics SaaS, Series A
Role: Head of Marketing
Call type: sales, second call
Cycle point: evaluation
Cost a deal: no
Six fields. The question verbatim, because the wording is the deliverable.
What qualifies
- A direct question, whatever the answer was
- An objection, stated or hedged ("I'm not sure how our board would see that")
- A comparison to another vendor or to doing it in-house
- A request for proof, examples or references
- A misunderstanding of what you do, corrected on the call
- A question asked after signing, during delivery
- A question the buyer asks another person on their own side while you listen
That last one is the highest-signal unit there is and almost nobody logs it.
What does not qualify
- Logistics: scheduling, invoicing, who is joining the next call
- Anything your side raised and the buyer merely agreed with
- Small talk, however charming
- A question you prompted by asking "do you have questions about X"
- Your own clever reframing of their question
The prompted-question rule removes a lot. Keep it anyway: a question the buyer would not have asked unprompted does not belong in a content plan built on what buyers actually want to know.
Cycle points
Tag each unit with one. The tag decides what the topic becomes later.
| Tag | When |
|---|---|
pre | Before or at the first call, often in the inbound message itself |
discovery | First call, before anything is proposed |
evaluation | Comparing, bringing in colleagues, asking for proof |
proposal | Pricing, scope, terms, procurement |
delivery | After signing |
renewal | Expansion, renewal, or a second project |
Volume
A quarter at a small agency is typically 40 to 80 calls, yielding 150 to 400 units. If you are getting under two units per call, the filter is too tight, most often because prompted questions are being dropped along with genuine ones asked in response to an open invitation.
Keep the raw file
The unit list is the asset, not the topic list that comes out of it. Next quarter's comparison needs this quarter's units, not its conclusions, and a question that looked like a one-off in March is the third occurrence of something by September.
references/scoring.md
Scoring and ranking
The gate
A cluster is not a topic until three different companies have raised it. Not three mentions, three companies. One prospect who asks the same thing five times is one data point and a sign they were not given a clear answer.
Below the gate, clusters go in a holding file and are re-checked next quarter. A two-company cluster that reaches four next quarter is a trend; the same cluster still at two is a coincidence.
Four factors
| Factor | Scale | Notes |
|---|---|---|
| Reach | companies raising it, 3+ | Raw count, not normalised |
| Cycle point | pre / discovery = 3, evaluation = 2, proposal = 2, delivery = 3, renewal = 1 | Both ends of the cycle score high, the middle less |
| Cost | 3 if it is on record as a reason a deal was lost or stalled, 1 if raised and resolved, 0 otherwise | Needs a lost-deal note to support it |
| Coverage gap | 3 if nothing exists, 2 if something exists but is thin or buried, 0 if well covered | Check the site before scoring |
Score is reach plus the three weights. Rank on it, then read the top of the list with judgment rather than taking the order literally. The scoring is there to stop the loudest recent call from leading the plan, not to replace a decision.
Why delivery scores as high as discovery
Discovery questions are what every competitor also hears and writes about. Delivery questions are what only a firm that has actually done the work hears, which makes them the hardest content for a competitor to copy and the most convincing proof that the firm has done this before.
Most content plans ignore them entirely, because the person writing content is not on delivery calls.
Red flags in the scored list
- Everything scores within two points. The gate is too loose or the clusters are too broad. Split them.
- Top five are all
proposal. This is a sales-enablement backlog, not a content plan. Deliver it as such. - Top topic has reach 3 and nothing else. Barely cleared the gate. Rank it below a reach-8 topic even if the formula disagrees.
- A topic scores high on coverage gap and has been on the list before. It was planned and never written. That is a capacity problem, and adding it again changes nothing.
Quarter-over-quarter
Three movements worth naming when the plan is presented:
| Movement | Reading |
|---|---|
| Topic drops off after an asset shipped | The asset worked. Say so; it is the only evidence content marketing ever produces. |
| Topic persists after an asset shipped | Nobody can find it, or sales is not sending it. Distribution problem, not a content one. |
| New topic enters at high reach | Something changed in the market or in the firm's own positioning. Worth understanding before writing. |
references/topic-to-asset.md
Topic to asset
Each surviving topic becomes exactly one asset and gets one internal placement. One, because a topic split across a post, a video and a newsletter gets none of them finished.
By cycle point
| Cycle point | Asset | Internal placement |
|---|---|---|
pre | Public page or post aimed at the search or the scroll | Nothing. This one is for strangers. |
discovery | Explainer with the firm's actual position in it | Pre-call send |
evaluation | Comparison, worked example or case study with numbers | Follow-up email after call two |
proposal | One-pager the team sends during a live deal | Proposal appendix, and the sales folder |
delivery | How-to or process piece, published | Onboarding pack |
renewal | Short written piece or a slide | Quarterly review deck |
The internal placement is the half everyone skips. An asset that exists but is not in the hands of the person on the call gets found by nobody, and next quarter the topic is still on the list.
The brief
Four lines per topic. More is a content calendar; this is a plan.
TOPIC: Who owns the direction when the client has an in-house designer
Reach: 7 companies | Cycle: evaluation | Score: 14
Their words: "What happens when our designer disagrees with you?"
Asset: Worked example, how a shared-ownership engagement ran, with the
two decisions that went the client's way
Placement: Follow-up after call two
Owner: [name] | Due: [date]
Their words is the mandatory line. It carries the buyer's phrasing into the brief so the finished piece uses the language they actually use rather than the firm's internal vocabulary.
Formats that match
| The topic is | Write |
|---|---|
| A how-does-it-work question | A process piece with the real steps and the real timeline |
| A will-this-work-for-us question | A worked example with a comparable situation and numbers |
| A why-are-you-expensive objection | A piece on what the work involves, or a pricing page change. Not a values post. |
| A comparison to doing it in-house | An honest account of when in-house is the right answer |
| A misunderstanding of what the firm does | A positioning fix first. Content cannot outrun unclear positioning. |
| A delivery question asked repeatedly | A how-to, and a change to the onboarding pack |
The quarterly cap
Four to six assets per quarter for a small agency doing client work. Plan more and the back half of the list does not ship, which costs more credibility than a shorter plan does.
Rank the full list, commit to the top five, and leave the rest visible as the reason the next quarter's plan already exists.