Your analytics can already tell you what your market is saying. It cannot tell you what to do about it.
We read language data the way a linguist does — the frames, the metaphors, the anchors people reason with — and return a list of things to change.
The problem with descriptive text analytics
Sentiment scores, keyword associations, topic clusters and LLM summaries are genuinely good at describing a corpus. They are accurate, they are detailed, and they stop exactly where the decision begins.
A specialist insurer came to us with that problem. Their supplemental health and life policies landed well with clients over 45 and missed everyone younger. They had done the sensible thing: collected an omnichannel corpus — support recordings, emails, social chatter, news, their own marketing — and run top-tier tooling over it. Not just sentiment: keyword association, emotional analysis, per-segment LLM summaries. The reports were perfectly accurate and entirely descriptive. The company knew its messaging was not landing. It had no idea which words to change.
They rewrote the marketing and A/B tested the result. It moved the needle slightly. It felt like throwing darts in the dark.
What we do instead
We do not count words. We reconstruct the conceptual frames a group uses to make sense of a thing, and check them against the frames your copy assumes. Where those two disagree, you have your answer — and usually it is not the answer a keyword report would suggest.
Two findings from that insurance corpus, neither of which the analytics platform could have produced:
"Security" is two different concepts. The marketing was built on calm, safety and protecting people you love from the risk of illness. For the over-45 cohort that is exactly right: security is a safety net, and what they are buying is peace of mind in a crisis. For the under-45 cohort, security is a springboard — health optimisation, physical performance, hitting earning targets. They are buying the freedom to keep living actively, without a setback slowing them down. The same word, two incompatible frames, one of which the campaign was actively arguing against.
Price is anchored to different things. Older customers anchor cost to salary, so "a small percentage of your income for peace of mind" reads as reasonable. Younger customers anchor to lifestyle, where a salary percentage feels abstract and heavy. Reframed as everyday habit — the cost of four or five barista coffees a month — the value landed immediately.
A tool can group price and security into clusters. It cannot tell you to move your entire pitch from "a percentage of income for a calm safety net" to "a few coffees for the freedom to live actively."
What it was worth
After rewriting the campaign language around the frames its audience actually used, the insurer recorded a 7.5% increase in positive answers — prospects formally entering the sales pipeline. At the top of a financial-services funnel, a shift of that size carries a long way down.
Why financial services first
Most of our published research runs on financial language, because it is the best available material for studying how people reason in public under uncertainty — millions of decisions, narrated in real time, with a price series attached that records what those people actually did.
- Market Metaphors reads 158,666 financial news headlines from 2009–2020 across five crises. It also settles a question worth knowing: the metaphors do not lead the market. Returns predict metaphor intensity five to six trading days later, and not the other way round. Financial journalism chronicles; it does not forecast.
- The Narrative Engine applies a metaphor-annotation protocol to FOMC governors' speeches, where every word is weighed before delivery. Governors differ consistently in whether the economy appears as a machine to be tuned, an organism, an object under pressure, or a traveller on a path — and different metaphors license different policies.
The team behind that: two decades of NLP much of it in fintech and regtech, including a compliance startup taken to acquisition, alongside a doctorate in the conceptual metaphors of political discourse. Neither of us is a finance professional and we do not pretend otherwise — we are linguists who find financial language unusually rich.
What you get
- A diagnosis. Which frames your audience is using, where your language collides with them, and what that collision is costing you.
- A list of changes, ordered — the specific reframings to make, not a set of themes to consider.
- The evidence. Every claim traced to the passages in your corpus that support it, so your team can check the reasoning rather than take it on trust.
- A working session to walk through the findings with the people who will act on them.
Who it is for
Financial services first — insurers, banks, fintech, asset managers — because that is where most of our published work sits and where the register is hardest to read from the outside.
But the method is not sector-bound. It applies wherever a large body of text carries decisions inside it: customer support archives, sales calls, clinical notes, policy consultations, press coverage. If you have the corpus and a question worth answering, the sector is negotiable.
It is not the right engagement if what you need is a dashboard to monitor, or a model deployed into production. We do the reading, the reasoning, and the recommendation.
Need the analysis built and visualised rather than diagnosed? See Commissioned Analysis.
Begin with a conversation
A short call to understand your situation, and to tell you honestly whether this would help. No charge, and no pitch — if it is not the right fit, we will say so.
Book a 15-min call Send a message
Or write to hello@crowintelligence.org.