Layers 6–7 · Measurement and automation
Attribution from source to closed customer, reporting a founder can read, and automation of the repeated steps, with failure alerts and a manual path.
Automation makes an existing process faster. If the process is wrong, it can make the wrong thing happen more often, with fewer humans noticing. The process therefore gets tested before its steps are automated.
Ariadne starts from the repeated work rather than from a preferred tool: what happens, how often, what it costs in time and errors, what the process should be, and only then which parts should stop involving a person.
Measurement comes first in this pairing for a reason. You cannot sensibly decide what to automate until you can see where the time and the money are going.
Fit
Symptoms
Scope
The channel, campaign, and page that produced an inquiry, carried through the CRM to the signed deal. Built at a scale a ten-person business can maintain, not an enterprise stack that needs a full-time owner.
Qualified inquiries, response time, booking rate, close rate, and channel contribution. One view, refreshed automatically, with the definitions written down so the numbers mean the same thing in six months.
The repeated work mapped and ranked by frequency, time cost, and error cost. Some of it should be automated. Some of it should be deleted. Some of it should stay manual because judgement is the point. That ranking is a deliverable in its own right.
Document and quote generation, scheduling and reminders, intake-to-delivery handoffs, data sync between the tools you already pay for. Built with the right technology for the job, which is sometimes a no-code platform and sometimes a small piece of code.
Summarising inbound context, drafting first-pass responses, classifying and routing unstructured requests, extracting structure from documents. Applied by someone whose research background is in exactly this, which mostly means knowing where a deterministic script is the correct and cheaper answer.
Every automated path reports success and failure. When something stops firing, you find out the same day rather than from a customer three weeks later. Every path also keeps a manual route, because systems fail and businesses cannot.
Process
Every repeated task ranked by frequency multiplied by time multiplied by the cost of getting it wrong. That ranking is a deliverable in its own right.
Some of it should be automated. Some should be deleted. Some should stay manual because judgement is the point. That decision comes before any tool is chosen.
Sometimes a no-code platform, sometimes a small piece of code. Chosen on volume, complexity, failure cost, and who maintains it afterwards.
Every automated path reports success and failure, keeps a manual route, and is documented well enough for another engineer to maintain.
Outcomes
These are the kinds of change this work produces. No numbers appear here, because no client result has been measured and published yet. When one has, it will appear in Work with its baseline, method, and time window.
Tooling
Categories, not endorsements. Where a deterministic script is cheaper and more testable than an AI call, that is what gets built.
The connected view
A growth system is only as strong as its weakest handoff. Fixing one checkpoint in isolation helps only if the adjacent checkpoints can carry the result.
See the whole systemCheckpoint 9 feeds evidence back into checkpoint 1. That loop is the point.
Where this sits in the system
Before this: sales-process automation belongs to the CRM system, and it should be right before the rest is built on top.
CRM and sales systemsAlongside this: the work with no fixed shape — prose, speech, documents — where a model earns its place and a script cannot.
AI automationAfter this: measurement feeds the next improvement, which is what turns effort into compounding.
The whole growth systemStraight answers
If yours is more specific, put it in the audit form.
support@ariadne.fyiIt depends on volume, complexity, what it costs when it fails, and who maintains it afterwards. No-code platforms are genuinely the right answer for low-volume, low-stakes connections, and they get expensive and fragile at scale. I will tell you which side of that line you are on rather than defaulting to whichever one I prefer to build.
Rank repeated work by frequency multiplied by time multiplied by the cost of getting it wrong, then inspect the highest-value candidate. Customer-facing judgement calls carry greater trust risk, so they need a stronger case and a human fallback than deterministic administrative steps.
In a growing service business it usually removes the worst parts of several jobs rather than a whole role: the copy-and-paste, the chasing, the re-keying of the same information into three systems. That is the honest answer, and it is worth being clear about it with your team before the project starts rather than after.
Only where probabilistic output is appropriate. Language models can help summarise, classify, extract, and draft from unstructured input. When an output must be exact and repeatable, a deterministic rule or script is easier to test and control. The design records which kind of component is being used and where a human reviews the result.
Your practical starting point
The audit maps the repeated work in your business, ranks it by what it actually costs you, and tells you which parts are worth automating and which are a process problem in disguise.