How Mandate Compression Works
What compression changes, what it never touches, and why shorter mandates help SI and Phona.
What It Is
When you create a mandate, Zyntro turns your guidance into Machine Instructions — the structured rules SI follows. Those rules are written in full sentences, because that is how they are first generated.
Nobody ever reads them as prose, though. Your contacts never see a mandate. The only readers are Zyntro's own AI: the part that writes your emails and texts, the review that checks each draft before it goes out, Phona on a call, and live chat.
Compression rewrites the descriptive text in a mandate so it says the same thing in fewer words. Every rule, condition and limit stays. The padding goes — "Make sure that you", "It is important to", sentences that restate their own heading.
It covers the Machine Instructions and the mandate's notes — the Assumptions and Potential Conflicts or Risks shown on the middle card of the mandate page. Your original guidance and the mandate summary are left as they are.
Why It Matters
A mandate is not read once. Every active mandate that applies to a contact is read in full each time SI writes to that contact, each time a draft is reviewed, and at the start of every Phona call. A brand with eight mandates carries all eight, word for word, into every one of those moments.
Phona is where this matters most. A call starts with a fixed amount of room for instructions, shared between the call flow, the mandates attached to it, and what Zyntro knows about the person being called. Room taken by wordy mandates is room the call flow and the contact's history cannot use. Compressed mandates leave more of it for the call itself.
The same holds for everything SI writes: tighter instructions keep the rules that matter in sharper focus, with less filler between them.
Compression is free, and it changes nothing until you have read the result and chosen to apply it.
How It Works
1. Only descriptive text is offered for rewriting
Zyntro picks out the fields that hold descriptive text — anything five words or longer. Everything else is never offered to the AI at all, so it cannot change:
- The structure: every field name, every list, and how many items each list holds
- Numbers, yes/no settings and short labels such as
consultativeorbrand - Lists of exact wording — the words and phrases to use or avoid, example phrasings, phrasing patterns and the signals used to spot an objection
- Names: competitors, audience segments, the subject of a comparison
Those lists are protected even when an entry is long, because SI is meant to use or avoid that wording exactly as you wrote it.
2. Each field is rewritten on its own
The AI rewrites each field in fewer words. It must keep every instruction and condition, keep the strength of each rule — never stays never, prefer stays prefer — keep the situation each rule applies to, and add nothing new. A field that is already tight is left alone.
3. Zyntro checks every rewrite
Before you see anything, each rewritten field is compared with the original. If the shorter version lost a number, a name, or any wording the original had in quotes, the field goes back to its original text. These are listed in the review window as kept as written, with the reason.
4. You review, then apply
The review window shows every changed field side by side, and how much smaller the mandate becomes. Nothing changes until you click Apply.
Applying updates the same mandate — it is not a new version, and its version number does not change. Every call flow, stream, broadcast, live chat widget and recipe that uses the mandate goes on using it. Zyntro keeps the original wording, so you can restore it later.
What compression changes
| Part of the mandate | Compressed? | Why |
|---|---|---|
| Descriptive fields in Machine Instructions (tone, opening and closing patterns, escalation rules, scenarios, adjustments, narratives) | Yes | Written for an AI reader, so full sentences add nothing |
| Assumptions and Potential Conflicts or Risks | Yes | SI reads these alongside the instructions |
| Words and phrases to use or avoid, example phrasings, phrasing patterns, objection signals | No | Used word for word |
| Competitor names, segment labels, other names | No | A changed name is a different name |
| Numbers, yes/no settings, short labels | No | Never offered for rewriting |
| Field names and the shape of the instructions | No | Parts of Zyntro look for fields by name |
| Your guidance, the mandate summary, Applies to, priority, status | No | Outside the machine instructions |
Examples
An escalation rule loses its padding, not its trigger
Before: "Escalate to a human agent if the client expresses high anxiety about timing (buying vs. selling first) or mentions complex financial conditions that need manual review."
After: "Escalate to human agent if client shows high anxiety about timing (buying vs. selling first) or mentions complex financial conditions needing manual review."
Both triggers survive, and so does the example in brackets that tells SI what timing means here.
A condition is shortened because its heading already says when
Under Ask Clarifying Questions When, the entry "The prospect's budget is unknown and a 'total cost' comparison is needed to give useful advice" becomes "Budget unknown; 'total cost' comparison needed for useful advice."
The heading already carries the when, so the entry only needs the condition. The quoted term 'total cost' is kept exactly — quoted wording always is.
A rewrite drops a competitor's name
Suppose a comparison rule says "Do not criticise Acmeworks directly; focus on what the contact gains", and the rewrite comes back as "Don't criticise competitors directly; focus on contact's gains". The check sees that a name went missing, so that field keeps its original wording and appears under kept as written with the reason lost the name Acmeworks. Every other field still compresses.
Info: Compression changes how a mandate is worded, never what it applies to. Its scope, priority and status stay exactly as they were.
