One agent per pillar. Start with the one matching the pillar you scored weakest, and build the second only once the first is running.
Every one of these is draft only. It produces something a person reads and decides on. None of them send anything, change a record, or contact a customer. That is deliberate: rung 1 of the agent constitution, and it is where every agent starts.
Each was picked on four tests. One person can build it in under a week. It needs only data the team already has. Somebody outside the team notices the output. And it is appropriate for a team at stage 1 or 2, which is where most teams actually are.
What it does. Reads open-text feedback and groups it into themes, so a pile of comments becomes a short list of recurring problems.
Data it needs. Support tickets, and any survey verbatims and public reviews you have. Tickets alone are enough to start.
What it produces. A ranked list of themes with the volume behind each and example quotes.
What a human still decides. Which theme is worth acting on this quarter, who owns it, and what the customer gets told once it is fixed.
Success metric. Days from feedback arriving to a named owner accepting it.
What it does. Scores your customer base for advocacy readiness, so you stop asking whoever the account manager happens to remember.
Data it needs. Renewal history, support history, and any survey scores or community activity you have.
What it produces. A ranked candidate list with the signal behind each name.
What a human still decides. Who actually gets asked, what they get asked for, and who does the asking. The agent never contacts anybody.
Success metric. Time to fill a reference request, measured in hours.
What it does. Watches your forum or Slack group and flags threads that are gaining traction or going unanswered.
Data it needs. Post and reply activity from whatever community space you already run.
What it produces. A daily list of threads needing attention, and which customers are answering other customers.
What a human still decides. Which thread gets a reply, what the reply says, and which of those helpful customers is worth recognising.
Success metric. Share of questions answered by another customer rather than by staff.
What it does. Compares which features accounts actually use against what your help content covers, and shows where the two do not line up.
Data it needs. Product usage data and a list of your existing content. Help-centre articles count; you do not need a learning platform.
What it produces. A gap report ranking unused features by how many accounts they affect.
What a human still decides. Whether each gap is a training problem or a product problem. Some features go unused because nobody explained them, and some because they are bad.
Success metric. Adoption of the target feature among accounts that saw the content, against those that did not.
What it does. Tracks each new account against your onboarding milestones and flags the ones that have stopped moving.
Data it needs. Signup dates, login activity, and whatever onboarding steps you already track, even in a spreadsheet.
What it produces. A list of stalled accounts, how long each has been stuck, and at which step.
What a human still decides. Whether a stalled account needs a call, a nudge, or nothing at all. Some accounts go quiet because they are fine.
Success metric. Share of accounts reaching first value within your target window.
What it does. Collects everything scheduled to go to customers from every team into one calendar, which is usually the first time anybody has seen the total.
Data it needs. Send logs and scheduled campaigns from marketing, product, support, and customer success.
What it produces. One calendar view, plus a count of messages per account per week and a flag wherever two teams hit the same customer on the same day.
What a human still decides. What gets cut, what gets moved, and who has to be told. This is the agent most likely to start an argument, which is the point.
Success metric. Messages per account per month, and the trend after you start cutting.
What it does. Maps senior contacts at your strategic accounts against your own executives, and shows which accounts have no relationship at all.
Data it needs. Your CRM contact records and a list of your own executives.
What it produces. A coverage map, and a gap list of strategic accounts with nobody assigned.
What a human still decides. Which executive takes which account, and whether they have the capacity to mean it. An assigned sponsor who never shows up is worse than no sponsor.
Success metric. Share of strategic accounts with a named sponsor who made contact last quarter.
Fill in the adoption block in the agent constitution first. It takes ten minutes and it is what lets you answer the question your RevOps or legal team will ask, which is who is accountable when this gets something wrong.
Kevin Lau, The Customer Continuum. These seven are drawn from a larger set of 135 agents across the seven pillars. Start with one.