Can't an LLM already do this?
It's a fair question. Large language models are improving fast. They can summarise a transcript, detect hedging language and flag that a speaker "seemed hesitant." We think this is the right question to ask — and we think the answer matters.
The difference between EchoDepth and a general-purpose AI is not capability. It is evidence. A description of tone is not the same as a scored, reproducible, audit-ready output — and in the contexts where EchoDepth is used, that difference is the whole point.
What LLMs can do. What they cannot.
Yes — a large language model can summarise a transcript and say 'the speaker appears confident' or 'hedging language detected'.
EchoDepth produces a 0–100 Trust Score with a per-second Credibility Signal timeline — not a description of what happened, but a scored, timestamped record of when delivery quality dropped and by how much.
Why it matters: A description of tone is not defensible in a board meeting or a regulatory review. A scored, reproducible output is.
General-purpose models are trained primarily on English-language Western content. Their sense of what 'confident delivery' sounds like reflects that bias.
EchoDepth is calibrated across 14 cultural and linguistic cohorts in 6 countries. A Trust Score in a Singaporean investor call reflects Singaporean delivery norms, not a Western proxy.
Why it matters: Cultural miscalibration produces false positives and false negatives. In a regulatory or investment context, those errors have consequences.
An LLM can tell you how a communication compares to general training data. It cannot tell you how it compares to 300 earnings calls from FTSE 250 companies in the same sector.
EchoDepth accumulates benchmark data with every scored communication. What is a typical Trust Score for a CFO earnings call? What does a high-performing change communication look like in financial services? That comparison data is the growing asset.
Why it matters: The value of a score is determined by what you can compare it to. Benchmarks take time and volume to build. They cannot be replicated by an LLM on day one.
LLMs produce conversational outputs — useful for exploration, not for compliance documentation. They are not designed to produce signed, reproducible, consent-documented evidence.
Every EchoDepth analysis is produced under a documented methodology, a signed Data Processing Agreement, explicit consent, and ICO registration ZB915623. The output is structured for audit — timestamped, reproducible, and defensible.
Why it matters: In FCA Consumer Duty, investor relations and legal contexts, the output needs to hold up in an audit. Conversational AI output does not. EchoDepth output does.
Any consultancy could instruct an LLM to score communications. The output would not carry the authority of a named methodology with documented calibration.
When a consultancy delivers an EchoDepth-scored report to a client, they are delivering a named, published methodology with documented accuracy. The Cavefish name on the report carries the same function as a rating agency name on a credit assessment.
Why it matters: As the methodology embeds in consultancy workflows, it becomes part of how those firms do business. The switching cost grows independently of what underlying AI can detect.
Where the defensibility comes from
EchoDepth's position is not built on being the most capable AI. It is built on four things that take time and deployment history to accumulate — and that a general-purpose LLM cannot replicate on day one.
Validated methodology
The scoring framework is calibrated across 6 countries and 14 cultural cohorts, with documented accuracy. It is not a prompt. It is a methodology — which means it can be named, cited, referenced and defended in ways that conversational AI output cannot.
Benchmark data
Every scored communication adds to a growing comparison dataset. What does a high-trust earnings call score? What is a typical Trust Score for a change communication in financial services? This benchmark data is a proprietary asset that cannot be replicated without deployment history.
Governance and audit layer
EchoDepth outputs are consent-documented, reproducible, signed and structured for audit. ICO registered ZB915623. DPA provided as standard. In regulated contexts — FCA, investor relations, legal — this is the difference between a useful tool and a defensible process.
Partner distribution
As consultancies embed EchoDepth into their client workflows and deliverables, the switching cost grows independently of what underlying AI can detect. The methodology becomes part of how those firms do business — a distribution moat that compounds over time.
Why Moody's wasn't replaced by a spreadsheet
Anyone could build a spreadsheet that estimated credit risk. The reason Moody's wasn't replaced by a spreadsheet is that their value was never the formula — it was the named methodology, the accumulated benchmark data, the institutional trust, and the fact that their rating appeared on the cover of a bond prospectus and carried legal weight.
EchoDepth is building toward the same structural position in communication assurance. An LLM can tell you a CFO sounded uncertain. An EchoDepth Trust Score of 67, produced under a documented methodology and presented in a board-ready PDF, tells an investor that the preparation was rigorous — or wasn't — in a way that is defensible, repeatable and citable.
The risk is not that AI gets better at emotion detection. The risk is building a business that depends only on the detection capability rather than the methodology, governance and distribution built around it. EchoDepth is designed to sit in the second category.
Common questions
Can't a large language model already do what EchoDepth does?
LLMs can summarise content and flag surface-level tone signals. EchoDepth applies a validated, culturally-calibrated scoring methodology and returns a reproducible, auditable output. The difference is not capability — it is evidence. A signed, timestamped Trust Score of 67 produced under a documented methodology is not the same as an LLM saying a speaker 'seemed hesitant.'
What makes EchoDepth defensible as LLMs improve?
Four things: benchmark data (comparison context an LLM cannot replicate without deployment history), validated methodology (calibrated across 6 countries, 14 cultural cohorts), governance layer (consent-documented, reproducible, structured for audit), and partner distribution (as the methodology embeds in consultancy workflows, switching cost grows independently of detection capability).
Why does the governance layer matter?
In regulated contexts — FCA Consumer Duty, investor relations, legal — the output needs to be defensible, not just useful. A FCA audit does not accept 'AI said the customer seemed distressed.' It requires a documented methodology, a reproducible process and a signed audit trail. EchoDepth is built to produce that. A general-purpose LLM is not.
Is EchoDepth a SaaS platform or a service?
Currently both. EchoDepth is a service-led product company — most engagements begin as high-touch pilot programmes with scored PDF outputs delivered within 5 working days. API access and platform integration are available for partners and enterprise deployments. The roadmap moves toward product as benchmark data and partner volume accumulate.
See the methodology in practice
Submit a piece of content. Receive a scored EchoDepth report — methodology documentation, Trust Score, Credibility Signal timeline and coaching notes. Free, no commitment.