SkillInvoke for agencies
AI process capture for agencies

Your agency runs on what's in three people's heads. Get it out.

Your best people talk through how they actually do the work, in voice notes, on their own time. An AI interviewer asks the follow-ups, a human checks every line, and you get back a tested AI playbook the whole team can use in ChatGPT, Claude or Copilot.

  • About 20 minutes of voice notes
  • The expert approves every line
  • No workshops, no new software
MA
Maya, Head of AccountsProcess: client onboarding
INTERVIEW
trigger steps exceptions escalation mistakes done-when

Client onboarding

Playbook by Maya · drafting from interview
    Sound familiar?

    The same question, asked for the fortieth time.

    Your processes work. They just live in Slack threads, old decks, and the heads of the people you can least afford to interrupt.

    “Where's the onboarding checklist?” “Ask Sam.”

    Every new starter, week one

    Every account manager's monthly report looks slightly different.

    The client notices

    When a senior leaves, the process leaves with them.

    Then you rebuild it from memory

    Nobody has time to write the SOP. And nobody reads the one that exists.

    Last updated: 2023

    We bought everyone ChatGPT. It writes nice emails.

    And that's about it so far
    How it works

    From voice notes to a playbook in about a week.

    No workshops and nothing to install. Your expert talks and we do the rest, with a real person in the loop at the point that matters.

    01 — YOU

    Pick a process and send a link

    Choose one thing your agency does again and again. Send your expert a link. That takes two minutes.

    2 minutes
    02 — YOUR EXPERT

    They talk. The AI asks the follow-ups.

    They answer in voice notes, whenever suits them. The interviewer chases the “usually”s, the exceptions and the judgement calls. Then they drop in three good examples and one that went wrong.

    ~20 min, async
    03 — A HUMAN

    We review and ask what the AI missed

    A person reads the whole thing, sends a few follow-up questions, and builds the playbook. Every line links back to the moment it was said.

    Human reviewed
    04 — EVERYONE

    Tested, approved, in your tools

    We run the playbook against your real examples and show you what passes. The expert signs it off, and it goes into ChatGPT, Claude or Copilot for the whole team.

    Live in your tools
    playbook · client-onboarding · v3Approved by Maya

    Client onboarding

    Owner Maya Adeyemi, Head of Accounts · Captured from 4 voice notes · 18 min
    Starts when
    A signed proposal lands in the CRM as “Won”▶ 0:04
    Steps
    Read the sales handover for anything non-standard we promised▶ 0:12
    Agree the three numbers the client cares about before any report▶ 0:21
    Kickoff within 5 working days, using the deck in /Clients/_Kickoff▶ 0:39
    Judgement calls
    Retainer over £8k → a director joins kickoff. Ask Maya if unsure.▶ 1:03
    Scope differs from the proposal → stop and flag it to Maya before kickoff.▶ 2:41
    Tested against Maya's examples
    ✓ Acme retainer✓ Northside launch✓ Fern & Co✗ Rushed Q4 project → fixed in v3
    What you get back

    Not a wiki page. A playbook that does the work.

    • i

      The real version, not the official one

      The interviewer asks about the last real case, not the ideal one. That's where the exceptions and workarounds come out.

    • ii

      Every line shows where it came from

      Click a timestamp to hear the moment it was said. Nothing is invented, and the expert can check every line.

    • iii

      Judgement stays with your people

      The playbook handles the routine work and sends the hard calls back to the person who owns them.

    • iv

      Tested before you trust it

      We run it on your real examples and show what passes and what doesn't, so you're not taking it on faith.

    For the people being interviewed

    This isn't about replacing Maya. It's about not interrupting her forty times a week.

    For agency owners

    Scale your best people.

    • New starters up to speed in days, not a quarter
    • The same standard on every account, whoever runs it
    • Knowledge stays in the agency when people move on
    • The ChatGPT licences you're paying for finally get used
    For your experts

    Hand off the boring 40%.

    • Start with the parts of the job they're tired of explaining
    • Their name is on the playbook, and they own it
    • The difficult decisions still come to them
    • Talk it through on a walk instead of writing documents
    Named ownerEvery playbook says whose expertise it is. They approve every change.
    Private transcriptOnly the expert and the reviewer see the raw recording. You see the approved playbook.
    Nothing ships unsignedThe expert can edit or delete any line before it goes live.
    Never used for trainingYour recordings and examples aren't used to train any model, and are deleted on request.
    Where agencies start

    Pick the one that causes the most interruptions.

    Client onboarding most common Brief → scope → quote margin Monthly client reporting consistency Pre-launch QA fewer mistakes Pitch & proposal prep speed New-starter handover ramp time Scope-creep conversations judgement Campaign retros learning Tone of voice per client quality
    Pricing

    Try one process free. Then decide.

    Fixed prices, no day rates. Typical UK process-documentation and AI-build projects start in the low thousands. The founding pilot is priced to get a handful of agencies in early.

    First process
    £0

    See what comes out of your own expert's head before spending anything.

    • One process, one expert
    • AI voice-note interview
    • Human-reviewed playbook
    • Yours to keep, whatever you decide
    Start free
    Founding pilot
    £2,400 one-off

    Up to five processes, made usable across the team. Standard price after the founding round: £3,500.

    • Up to 5 processes and their experts
    • Follow-up questions from a human reviewer
    • Tested against your real examples
    • Set up in ChatGPT, Claude or Copilot
    • A 45-minute team rollout call
    • 30 days of fixes as you use them
    Apply for a founding place
    Keep it current
    £750 / month

    Processes change and models change. Keep your playbooks in step with both.

    • Up to 3 new or refreshed playbooks a month
    • Re-tested whenever a model update lands
    • Changes suggested from team feedback
    • Cancel any time
    Talk about it

    Prices exclude VAT. The founding pilot is limited to 10 agencies because a person reviews every playbook.

    Consultants and agencies offering AI services

    Want to run this for your own clients?

    We're opening the interviewer, the review workflow and the method to a small group of ops consultants, fractional COOs and agencies that want to offer process capture under their own brand.

    Join the practitioner list
    Questions

    Fair questions.

    Couldn't we just ask ChatGPT to write our SOPs?

    You could, and you'd get a tidy generic SOP. What's valuable is what's never been written down: the exceptions, the “it depends”, and what new starters always get wrong. The interview exists to draw that out, and the testing shows whether the playbook actually handles it.

    Will my team think we're replacing them?

    They might, if it's introduced badly. So the expert's name is on the playbook, they approve every line, the raw transcript stays private to them, and the hard decisions are sent back to them. We suggest starting with the work they're most tired of explaining.

    How much of my expert's time does this take?

    About 20 minutes of voice notes, spread over whenever suits them, plus 10 minutes to review and approve. No meetings, unless they'd rather talk it through live.

    Where does our data go?

    Recordings and examples are used only to build your playbook. They're never used to train models, and they're deleted when you ask. Only the expert and the reviewer see the raw transcript.

    Which AI tools does it work with?

    ChatGPT (custom GPTs and Projects), Claude (Projects and Skills) and Microsoft Copilot. We deliver the playbook in whichever one your team already uses. If you use more than one, it works in each.

    Is this a software product or a service?

    Right now, both. The AI runs the interviews and a person reviews everything. We're doing it this way on purpose while we learn what makes a great playbook. Founding agencies help shape the product and keep their founding price.

    What if the free playbook isn't useful?

    Then you've spent 20 minutes and told us something we need to know. No invoice, no follow-up sequence, and you keep whatever we made.

    Start with one process

    Twenty minutes of talking. A playbook you keep.

    Tell us the process that causes the most interruptions. I'll set up the interview link and review the playbook myself.

    No sales call needed for the free process.

    SF
    SteveFounder · reviews every playbook personally

    Capture your first process

    We only use this to reply to you. No newsletter.