A Recognition Program That Doesn't Die in Week Three

They launch a peer-to-peer recognition program. Week one — everyone's thrilled, everyone writes, the digest is full, the portal home page gets 30 thank-yous a day. Week two — the flow weakens, only the activists are still writing. Week three — dead silence. By the one-month mark the program is buried. I've seen this scenario at dozens of companies. And I know where it breaks.

Most recognition programs don't die of indifference. They die of a barrier. The barrier at the moment a person wants to say thanks, sits down to write — and runs into a blank field. "Thanks for the work" sounds formal. "Thanks for yesterday when you..." — and then the words don't come. They close the tab and move on. No thank-you happened.

That's the main mechanism of a program's death. Not a lack of motivation — a lack of words. Programs that ignore this barrier die. Programs that remove it live for years and become part of everyday culture.

Let me share what works in recognition at our clients, on three levels: the AI assistant, the manager dashboard, and the digest ritual.

The Main Barrier Isn't Motivation, It's Wording

In one Workhuman study, 73% of employees said they wanted to thank a colleague in the past week but didn't. The most common reason among the answers: "I didn't know how to phrase it," "I was afraid it would sound formal," "I didn't want to look like a suck-up."

This is a technically solvable problem. Not "we need to motivate the team better," not "let's run a private-feedback training." Simply: a blank text field is too high a starting barrier for a routine action.

Compare it with other everyday actions: sending a chat message, dropping a like, sending an emoji reaction. They all have a minimal starting threshold. That's why people use them. A thank-you in traditional UX is a blank field + a "send" button. That's too much work for an everyday act.

The solution isn't to remove the blank field — a substantive thank-you doesn't fit into an emoji. The solution is to take the first step for the person, leaving them the editing and final approval.

The AI Assistant — a Gray Suggestion in 10 Seconds

The technical approach that works best for us: an AI assistant in ghost-text mode — a gray suggestion in the input field. The employee starts writing a thank-you ("Thanks, Anna, for..."), and at that moment gray text appears in the field — a continuation you can accept with Tab, edit, or ignore.

What the AI assistant relies on. Not a "generator of nice words in general," but context: the recipient's profile (role, recent projects, what they're strong at), the sender's and recipient's recent joint actions (shared tasks, discussions), the company's shared values. When you start writing a thank-you to Anna after she helped with a code review two days ago, the AI knows it and offers a relevant phrasing.

What's important to understand: the AI doesn't write the thank-you for you. It offers a draft the person accepts, edits, or ignores. I consider this a fundamental boundary. If the AI sends on your behalf without your edit, that's no longer recognition, it's autospam.

In our implementation it works like this:

• No more than 1 sentence in the suggestion.
• The gray text is clearly distinct from your black text.
• The feature can be turned off entirely in profile settings (some employees prefer to write from scratch — fine).
• Logging: no thank-you sent without editing is marked as "genuinely written" — the system sees whether the draft was edited.

What we see at clients after rollout: the number of thank-yous grows 3–5x in the first two weeks. After a month it stabilizes at 2–2.5x above baseline. The barrier is removed, and the program keeps living.

The Manager Dashboard — a Blind-Spot List and a Fairness Map

The second layer is for managers. Most team leads are sure they recognize the team evenly. The data usually shows otherwise.

The blind-spot list. A simple table: report names + how many days ago you last thanked them. Red — more than 30 days. Yellow — 15–30. Green — fewer than 15. No emotions, just facts.

What we see when managers first open this panel: 60–70% find at least one person in the red zone, and 80% are surprised. "I was sure I thanked Anna last week. Turns out the last time was two months ago." That's the very recognition gap that isn't visible from the manager's seat.

The fairness map. The distribution of thank-yous you sent to the team over the quarter. If you have five people and a healthy distribution, each gets 15–25% of your thank-yous. If one person gets 42% and two others get under 5%, you have a skew. This doesn't mean you should "fine" yourself — it means it's worth paying attention. You may have a "star" who gets recognition disproportionately, and quiet employees who do quality work but are visually invisible.

A fundamental point on privacy: this panel is visible only to the manager, not to HR. HR sees company-level aggregates, not individual team-lead dashboards. Otherwise it turns into surveillance, and managers start "gaming" the metric.

Digests That Become a Ritual

The third layer is what turns the program into a repeating event rather than a one-off sprint.

The weekly digest comes out every Friday (or Monday — different companies find different things work). In it:

• 5–8 of the week's best thank-yous from the feed
• Brief statistics: how many thank-yous, how many unique senders, how many new people said "thanks" for the first time
• Optionally: a thematic cut — which values were mentioned most

What's critical: the digest is not fully automatic. The AI assembles a draft, the HR team (or a communicator) goes through it in 5 minutes, edits if needed, and publishes. It's a small but important step — it guarantees the digest stays meaningful and doesn't turn into noise.

Why the digest works. When a program has a ritual — a weekly moment where it "happens publicly" — it becomes part of the calendar. Without a ritual, the program exists only in the feed's flow and gets forgotten. With a ritual, it has a permanent place in the week, and the team subconsciously keeps it in mind.

The monthly digest is broader: trends, new "recognition givers" (people who started actively praising for the first time), recognition gaps by department. That's more for the HR team and for the narrative at the all-hands.

An illustration from practice: a mid-sized company launches a weekly digest in the fall, and by winter the number of unique weekly senders grows several times and stabilizes at a new level. The program didn't "spike and crash" — it embedded itself in the weekly rhythm. By industry benchmarks, a healthy peer-to-peer recognition program holds 60–100% monthly active participation (HeyTaco — Recognition Participation Rate). Before the weekly ritual is in place, most programs balance below that line; after, they reach it.

What DOESN'T Change When You Add AI

The most dangerous trap companies fall into when adding AI to a recognition program: "now AI does it for us." That destroys the program faster than its absence.

Recognition still has to be sincere. AI helps phrase it, doesn't do it for you. If a person gets used to AI sending thank-yous without their involvement, it turns into background corporate automation nobody trusts.

Recognition still has to be specific. AI is good at phrasing, not at the substance. Specific actions, specific results — that's your part. AI can suggest a sentence structure, but "Anna did a great code review" is something you know, not the AI.

The manager still has to pay attention. The blind-spot list shows who you haven't recognized. After that it's up to you: recall what good this person did in the past month and thank them honestly. If you can't recall anything, that's not "the AI being bad," it's you not having been engaged with the team for a while.

And one more important point: AI and the manager dashboard are tools for sustaining a recognition culture. If nobody at the company thanks anyone, and the culture isn't at the stage where it counts as the norm, no AI will "launch" the program. Behavior forms the base; AI amplifies it.

The Bottom Line

A recognition program dies not from indifference but from the wording barrier. Remove the barrier and the program keeps living. Three layers that work together:

• The AI assistant removes the blank-field barrier for the sender.
• The manager dashboard shows the blind spots the manager can't see themselves.
• The digest turns the program into a weekly ritual that doesn't get forgotten.

Each layer alone gives a lift; all three together move recognition from "a program" launched for a quarter into part of the culture that lives for years.

And in case you missed it: a recognition program only works when AI helps write rather than writes instead. That difference has to be kept out loud — otherwise it's forgotten, and the program turns into yet another piece of background automation.

Next, we'll dig into SSO for the corporate portal. How to onboard a thousand employees in a week without breaking existing processes. The topic is technical, but decisive — a portal you log into with a separate password is doomed.

If you want the "Launch a recognition program in 30 days" handbook with a concrete day-by-day plan, it's linked in the card. No signup.