Skip to content
CarabAI

Work

See the systems in action

CarabAI is early. Rather than present results we cannot yet stand behind, this page shows a system you can use right now and states plainly what is not here.

Read this first

What you will not find here

Most agency work pages open with logos and percentages. Ours does not, because none of those numbers have been independently verified for CarabAI yet, and inventing them would make everything else on this site worth less.

  • Client logos
  • Testimonials or quotes
  • Revenue or conversion increases
  • Time-saving statistics
  • Named case studies
  • Industry benchmarks

Demonstrations

What we can show you today

Each card below is a demonstration of how a system works, not an account of what it did for somebody else. One of them you can run yourself, right now, against your own answers.

No client results are published yet

Nothing appears here until an implementation has run long enough to compare against the baseline agreed before development, and the client has given written permission. The demonstrations below show how the systems work; none of them is a client result.

Interactive demonstration

AI opportunity audit demonstration

Custom AI applications

The audit on this site is itself a working system, and the one demonstration you can run yourself. It is built the way we build client work: deterministic logic wherever a number must be reliable, a language model only where judgement is genuinely needed, and a fallback that keeps the whole thing working when the model is not available.

What it shows

  • Scores computed by published arithmetic, never by a language model
  • Model output validated against a strict schema before anything is rendered
  • A complete deterministic report when the model times out, refuses or fails
  • A private, non-indexed result link that needs no account and no email address
Run it now
Walkthrough on request

Lead qualification and follow-up demonstration

Lead qualification and follow-up systems

How an enquiry is acknowledged within minutes, asked the qualifying questions that decide whether it is worth a human's time, and either booked or routed with the answers already attached.

What it shows

  • First response sent while the enquiry is still warm, on every channel that reaches you
  • Qualifying questions drawn from your own criteria, not a generic template
  • The handover a salesperson receives, with the answers already captured
  • Where the system stops and asks a person to decide

Shown and explained in a call against your own workflow, rather than published as a scripted demo. It is a demonstration of the system, not a client result.

Walkthrough on request

AI phone-agent demonstration

AI phone and customer-service agents

How a call that would otherwise ring out is answered or acknowledged, what the caller is asked, and what lands in your systems before anyone on your side picks it up.

What it shows

  • An enquiry captured outside working hours instead of lost
  • The escalation rules that hand a call to a person, and when they fire
  • The record written back, so nothing is retyped afterwards
  • What the caller is told about who they are speaking to

Shown and explained in a call against your own workflow, rather than published as a scripted demo. It is a demonstration of the system, not a client result.

Walkthrough on request

Administrative workflow demonstration

Workflow and administrative automation

How a repetitive administrative sequence — extract, validate, move, file — runs end to end, and how the exceptions it cannot settle are routed to a person rather than guessed at.

What it shows

  • Information extracted and validated instead of re-keyed
  • The confidence threshold below which a human reviews the item
  • The audit trail left behind for each processed item
  • What happens on a bad input, rather than only on a clean one

Shown and explained in a call against your own workflow, rather than published as a scripted demo. It is a demonstration of the system, not a client result.

Walkthrough on request

Internal knowledge assistant demonstration

Internal knowledge assistants

How a team asks in plain language and gets an answer drawn from the documents and systems you already have, with the source shown so the answer can be checked.

What it shows

  • An answer with its source document cited, not an unattributed assertion
  • The behaviour when the answer is genuinely not in the material
  • How permissions keep a person from retrieving what they may not see
  • How the material is kept current as documents change

Shown and explained in a call against your own workflow, rather than published as a scripted demo. It is a demonstration of the system, not a client result.

Inside the demonstration

How the AI Opportunity Audit is built

The audit is not a lead-capture form with a score attached. It is a working example of how we build: published rules for anything that must be reliable, evidence quoted back from your own answers, and a result that is identical every time you run it.

Deterministic scoring

Every number comes from published arithmetic on your own answers. The same answers always produce exactly the same result, every time.

Evidence you can check

Every risk signal quotes the answers that triggered it, and names the metric that would make it measurable. Nothing is asserted that you cannot trace back.

No AI in the analysis

The report is generated by our own code. No language model is called, no tokens are spent, and your answers are never sent to a model provider.

How the overall score is composed

Four components, each scored 0–100 from your answers, then blended by fixed weights. The weights are published here because a score you cannot interrogate is not worth much.

  • Business Impact35%
  • Automation Potential30%
  • Measurement Clarity20%
  • Implementation Readiness15%

Operational Exposure is scored separately and is not part of this blend. It describes risk, not opportunity.

The eight questions

One question per step, about three to five minutes in total. No account, and no email address required to see the complete result.

  1. 01What does your company do?
  2. 02How many people work in the business?
  3. 03Which area would you most like to improve?
  4. 04Which problems occur regularly?
  5. 05Approximately how many employee hours are spent on repetitive work each week?
  6. 06How frequently do leads or customer requests receive a late response?
  7. 07Which tools does your business currently use?
  8. 08What would you most like to achieve?
Run the free audit

What comes next

What will appear on this page

As implementations run long enough to compare against the baselines agreed before development, results will be published here. Each one will state the method, not just the outcome.

The metric and its baseline

What was measured, how it was measured, and what it read before anything was built.

The observation period

How long the comparison ran, so you can judge whether the result is meaningful.

What did not work

Projects that were stopped, and why. A results page containing only successes is a marketing page.

Client results will only appear with the client’s written permission, and anything identifying will be removed unless they ask for it to stay.

Would you rather see it applied to your own business?

Run the audit against your own answers, or tell us what you are trying to improve. Either way you will get a specific answer rather than a case study about somebody else.

Tell us about your projectRun a free AI opportunity audit

No prepared brief required. Tell us what you’re trying to improve, and we’ll help you identify the next step.