Meet Your AI Director
Sixteen specialists work overnight across three desks: Marketing AI, Recruitment AI, and Sales AI. Each morning, their best ideas to move your goals forward, with the proof and the work already done. One click to approve.
Real screens from Mizuhiki Search, a demo agency. Every name and figure on them is a sample.
How it works
From a Night's Reading to a Morning's Decisions
01 · Every morning
The brief opens your day
The AI Director reads every specialist's proposals, ranks them against your goals, and keeps the ones worth your morning. The rest wait in the Inbox.
Admin · Home
Open it live02 · Inbox
Each card carries its proof
What the agent found, why it matters, the numbers behind it (filled in by code from your own data, never typed by the model) and the change it would make.
Admin · AI team
Open it live03 · Approve
Approve, and the change is made
A job fix applies with a version you can undo. A page becomes a draft in the visual editor. Bigger work becomes a task.
Admin · Jobs
Open it live04 · Board
The work, for people and agents
Approved proposals and your own tasks on one board, assigned to a consultant or an agent.
Admin · AI team · Board
Open it live05 · Goals
Aimed at what you're aiming for
Set one to three goals, such as applications a month or a search to rank for. Every agent ranks its ideas by them, and the brief reports against them.
Admin · AI team · Goals
Open it live06 · Agents
You decide how far each one goes
Switch any agent off, choose when it runs, which data it can read, whether it only suggests or also drafts, and a monthly run limit.
Admin · AI team · Agents
Open it live
Open it liveEvery number on a card comes from your own records.
Open the demo and check oneSee Each Tool at Work
Walk-throughs on real screens
Ask
Ask About Your Own Numbers
A question in plain words, answered only from your site, jobs and pipeline. Every figure the answer used sits under it and opens the screen it came from.
- If a number can't be matched to your data, the answer isn't shown
- Reads your goals and the team's open cards
- Says when the data can't answer, and what would
Open it liveBrand book
One Voice for Every Draft
AI drafts your brand book from your current site. Once you approve it, every post, page and reply the team writes follows it, and anything that breaks it is stopped before it's public.
- Positioning, proof points, voice, words and boilerplate
- Japanese and English side by side
- Every approved version kept
Open it liveAI credits
Every Credit Accounted For
Every month includes 300 credits. Settings shows where each one went, by area, by person and by day, and each agent has a monthly limit you set.
- Each AI action uses only what it needs
- Top up 500 for ¥15,000 when you need more
- Any period's activity as a CSV
Open it liveThe team
One Director, Three Desks
Each desk is built for the person who owns its result. The AI Director runs all three, and one price covers the whole team.
AI Director
Every morning, the team's best ideas ranked against your goals, in a brief you can read in a minute.
- Today's decisions from the whole team's proposals (3 by default, or as many as you choose)
- Progress against your goals
- 01Each agent reads your site, jobs and market, and writes its proposals.
- 02The AI Director ranks them against your goals and keeps the best: three by default, or as many as you choose.
- 03Each card shows the numbers behind it and the change, ready to preview.
- 04You approve, change or reject. A reason you give becomes a rule every agent reads.
You Set How Far Each Agent Goes
- On or off, one by one
- When it runs
- Which data it can read
- Suggest only, or draft the change too
- A monthly run limit
- How many ideas reach your brief
- Rules it must never break
Marketing AI
Get found in search, AI answers, and social, and know which channel led to placements.
Brand Keeper
Every month, the live pages that stray from your brand book. Every draft the team writes is checked against it first.
- Checks the words the AI team writes first; a proposal whose wording breaks a rule never reaches your Inbox
- A short note when a draft's tone is off
- Live pages that break the book, and pages that contradict each other
SEO and Content
Every week, the searches you're just off page one for and the ones you don't show up for yet, with the page or post to draft for each.
- Searches just off page one
- Searches in your field your site doesn't show up for yet, and who ranks instead
- Sector page drafts, and a weekly check of each (jobs, record, logos, FAQ)
- Posts to draft
AI Search
Every week, what ChatGPT and Google's AI answers say when candidates and HR teams ask about your field: who they name instead of you, which pages they cite, and the page worth making.
- The agencies AI names instead of you, for the questions candidates and HR teams ask
- The kinds of pages AI cites, and the one worth making
- When AI gets a fact about you wrong: the sentence, and where to fix it
Link Builder
Every week, the sites that started or stopped linking to you, and the ones that link to rival agencies but not to you, each with what to ask for. A consultant sends the request.
- Sites that started or stopped linking to you
- Sites that link to rival agencies but not to you
- A person always sends the request
Social Editor
Every week, up to three posts drafted from your new jobs, blog posts and salary guide, each on a free day in your calendar, and one in each consultant's own voice. A person always posts.
- Drafts for your new jobs, posts and salary guide, each on a free day in the calendar (up to 3 a week)
- One a week in each consultant's own voice, for those who set it up
- Each month, a read of which of your posts went on to applications. A person always posts
PR and Events
Every week, a sourced press release for what you publish, entries for the awards you add a month before each deadline, and event ideas.
- A press release, every number sourced, when you publish a salary guide or a poll result
- For awards you add, an entry from your case studies and records a month before the deadline
- Event topics, and the thank-you afterwards. A person always sends
Site Analyst
Every morning, the jobs that get views but no applications, the buttons that underperform and slow first replies, each with a proposed change.
- Jobs with views and no applications
- Calls to action that underperform
- How fast applicants get a first response, and what happens next
Recruitment AI
Every job right and listed, every applicant answered, and the people you already know matched to new roles.
It drafts and suggests. A consultant sends every message, and no one is rejected by an agent.
Jobs Editor
Every night, a check of every job, and each fix ready to apply, plus a rolling check that your jobs show in Google for Jobs.
- Duplicates, broken titles, jobs filed under the wrong sector
- A confidential client named in the posting, and titles that could identify one
- Each with a fix you can apply
- A rolling check, a few jobs each night, that your jobs show in Google for Jobs
Compliance
On every publish, a check for placeholder salaries, expired jobs and missing details. Rules in code, no AI.
- Placeholder salaries (like ¥1), expired jobs, missing location or employment type
- Rules checked in code; no AI involved
Market Analyst
Every week, what the market's postings list that yours leave out, and your salaries next to published guides.
- Duties and requirements most market postings list that yours leave out
- Your salaries next to published salary guides
Candidate Assistant
Reply drafts in your words for every applicant waiting, interview packs with a map and calendar links, and a reminder the day before. A consultant sends each one.
- A reply draft for each applicant waiting to hear back (a consultant sends it)
- Answers to applicants' questions from the job itself; anything it doesn't say becomes "I'll check and come back to you"
- Interview prep points, for the consultant to check before they go in the pack
- Dropped balls: tomorrow's interview with no details sent, a decline nobody was told about
Researcher
Every morning, the people in your pool who match a live role at 75 or above, ready to shortlist in one step.
- New matches of 75 or more for your live roles, ready to shortlist in one step
- Your most placeable people: those who fit several live roles
- Roles at companies you don't work with yet that fit your candidates (a way into BD)
- What the people your team shortlists and dismisses say about the weights
Lifecycle Marketer
Every week, new roles for people who applied in that field before, and check-ins for people you haven't heard from in six months.
- New roles, for people who applied in that field before
- Check-ins after six quiet months, and a note one year after a placement
- Only people who agreed. The AI sees only how many people, never a name or an address
Sales AI
Find the companies hiring for what you recruit, win the meeting, keep the client, and turn results into proof.
Business Developer
Every week, foreign companies that have just started hiring in Japan, each with an intro note drafted.
- Newly arrived companies that have started hiring, with an intro note
- This week's new arrivals that match your saved views
- Follow-ups when something changes, and a market-entry deck for each meeting you book
Client Marketer
Every week, quiet clients who are hiring again with a check-in drafted, and a review deck for each client every quarter.
- Quiet clients who are hiring on AirTA, with a check-in note drafted
- New departments hiring at existing clients, with an intro note
- A review deck each quarter, with each client's results from the last two years
Reputation
Every week, a read of what candidates and clients said, a follow-up note for each low score, and quotes ready to share.
- A follow-up note for each low score (the consultant gets in touch)
- Quotes people agreed to share, ready to show on your site
- A weekly read of the answers, and your rating, with count and period, for the brand book
- Reply drafts for your Google reviews, once your Google profile is confirmed. A person posts each reply
It Reads Only What You Allow
Each agent works on a data level you choose. The levels are database permissions, so no agent can read past its own.
Level 0
Public content
Your pages, jobs and team profiles: only what's already public
Level 1
Analytics
Cookieless analytics and Search Console. No individual can be identified
Level 2
Pipeline counts
Application counts, stages and times. Never names, emails or notes
Level 3
Applicant data (no names or contact details)
Career history, skills, languages and the applicant's own message, with names and contact details removed first. Only the Researcher and the Candidate Assistant, and only once the owner allows it. Never names, emails, phones or LinkedIn
Level 3
Applicants' details
Work history and applications, for the Researcher and the Candidate Assistant only. Off until your owner switches it on and confirms your privacy notice covers it. Names, emails and phone numbers never reach the AI.
It Goes Only as Far as You Say
Set it per agent, alongside when it runs and a monthly run limit.
Suggest only
Cards only: what it found, the numbers behind it and what it would change.
Prepare changes
It also builds the change as a draft in your editor, ready to preview. A person still publishes it.
What No Agent Can Do
Built into every agent from day one. No setting can turn them off.
- Email a candidate or a client
- Decide anything about a person: a match is a suggestion, and no one is rejected by an agent
- Touch money, sign-in or your team's settings
- Delete anything
It Works to Your Goals
Set one to three goals and every agent ranks its ideas by them.
Goals
Forty applications a month, a search you want to rank for, a role you need to fill. The morning brief reports against them.
Your rules
Reject a card with a reason and it becomes a rule the whole team reads before every run, such as never publishing salaries for executive search.
Ask
Ask why applications dipped last week. The answer cites your own numbers and links to the screen that shows them.