A multi-AI workflow for consultants

One consultant can move from analysis to recommendation to a client-ready draft across their own assistants while keeping client boundaries and private notes intact.

3 min read

On Monday morning, one assistant helps you compare three operating models for a client. By afternoon, you want another to turn the recommendation into a concise executive draft. Your first chat also contains fee notes, candid observations about the sponsor, and fragments from a different engagement. None of those belong in the second assistant’s input.

Make one client, one matter, and one approved briefing the boundary.

Separate analysis from the client-ready draft

A useful consultant workflow has three outputs:

  • Analysis: checked facts, calculations, constraints, uncertainties, and source references.
  • Recommendation: the option you support, alternatives considered, tradeoffs, and conditions that would change your view.
  • Client-ready draft: language for the agreed audience, stripped of private working notes and claims the evidence cannot support.

Different assistants can challenge different stages. The second assistant should not infer which parts of a long analytical chat survived your review. Tell it.

Build a single-client handoff

Name the client or use the approved internal matter label, then include only what that engagement permits:

  • the audience and decision they face;
  • the scope and explicit exclusions;
  • verified inputs with document titles, dates, sections, or approved URLs;
  • calculations and assumptions, clearly separated;
  • the recommendation and rejected alternatives;
  • sensitivities, dependencies, and open questions;
  • tone and terminology required for the client-ready version;
  • the exact section or deliverable to produce next.

Keep commercial terms for your own services, internal risk notes, impressions of individual employees, and unrelated client material out. Private notes belong on your side of the briefing. If an observation is necessary, rewrite it as a relevant, supportable fact rather than carrying the aside.

RelayLink carries text, not contracts, workbooks, slide decks, recordings, or document permissions. A path or artifact URL helps the next assistant locate an item; the assistant still needs its own authorized access.

Hand work between your assistants

Connect both assistants to your RelayLink account and address the package to your own email. Draft, review, and confirm every handoff. The model never sends autonomously. Your review should catch copied client secrets, cross-client fragments, unsupported claims, and wording that sounds as though the client said something they did not.

A self-addressed package creates no contact request, email, magic-link page, or second account. It stays unread until one of your assistants pulls it, with the portal inbox as a fallback. RelayLink does not synchronize the two sessions; it gives the destination only your approved briefing. Sending a briefing to yourself covers the full loop.

Your RelayLink account is your identity. The package is your work record to yourself. It does not impersonate the client, turn the model into their representative, or grant permission to contact anyone.

Review before anything becomes client-facing

Ask the drafting assistant to preserve the recommendation’s conditions and uncertainty. Polished prose often drops the caveat that made a recommendation honest. Compare the draft with the evidence and rerun any calculation that matters.

Check names, dates, quotations, totals, and contractual language at the source. A second model is another opinion and another drafting pass, not independent validation. Provenance can show which wording you wrote or approved; it cannot make an unsupported recommendation sound.

When not to use this workflow

Do not send client material to a provider the engagement, employer, contract, or law does not permit. Your approval in RelayLink cannot widen those permissions. Sanitize the briefing only if the remaining facts are still sufficient for the work.

Keep a task in its controlled system when it depends on a complete data room, privileged advice, live financial models, or fine-grained document access the second assistant lacks. If another consultant or the client must make the next decision, this is no longer a self-handoff. Address the person through the appropriate approved channel and let them own their response.

Frequently asked questions

Can I combine notes from several clients in one self-handoff?
Do not. Use one client and one matter per briefing. Mixing contexts makes disclosure harder to detect and gives the destination assistant material it does not need.
Does my RelayLink account represent the client?
No. It represents you. A self-handoff moves your work between your assistants; it does not make either assistant the client or authorize you to speak as the client.
Does RelayLink carry client files to the destination assistant?
No. It carries approved text. The other assistant needs separate authorized access to any contract, workbook, transcript, or document required for the task.