AI activity without meetings is cost inflation

More labeled activity is not more pipeline.

TechCrunch reported insurers' claim that hospital AI coding drove about $942 million in extra spend over two years. The Blue Cross Blue Shield Association argument, as covered there: more "complex" coding without matching changes in care. I am not here to litigate healthcare. I am here because the pattern maps clean onto GTM.

AI that inflates activities, tasks, "research complete," and sequence steps without meetings or SQL is the same shape. New labels. Higher cost. Same or worse outcomes.

What is the GTM version of coding inflation?

Activity inflation.

Agents mark research done. Enrichment rows climb. Sequences enroll. Slack fills with "completed" checkmarks. The calendar does not move. Finance still pays for seats, tokens, and the humans babysitting the queue.

Vanity metrics kill GTM teams for a reason. Views, MQLs, and now agent task counts all fail the same grunt test: did a meeting happen? Salesforce still puts about 27% of reps at quota. A dashboard full of green agent tasks does not change that number.

Why does this happen so fast with AI?

Because models are good at producing artifacts that look like work.

A summary. A score. A draft. A completed step in a workflow UI. Each one is easy to log and easy to celebrate. Meetings are hard. SQL is harder. So teams optimize the loggable surface.

Meetings, not theater is the scoreboard. Theater got a new costume. The costume is "AI activity."

Ehrenberg-Bass / LinkedIn B2B Institute: about 95% of buyers are out of market. Agents that research the 95% at machine speed produce beautiful activity and empty calendars. That is cost inflation with a product tour.

What should you measure instead?

Operator metrics. Not vibes.

1. Meetings held from agent-assisted or signal-sourced sends
2. SQL behind those meetings
3. Complaint / rage-reply rate on the wave
4. Human hours spent babysitting the agent per meeting booked

If activity rises and those four do not, you bought inflation. Reply rate vs meeting rate already kills the "our replies are up" story when meetings stay flat. Apply the same knife to agent task counts.

How do you stop the inflation without killing AI?

Rails and scoreboards. Not a ban.

Buy the agent, build the rails. The seat is cheap relative to a burned domain and a confused team. The rails decide whether the seat produces meetings or produces labels.

Write one rule on the wall: no agent KPI ships to leadership unless it sits next to meetings and SQL for the same window. Research complete is a means. It is not a win.

Now some of you are thinking: we need leading indicators. Fine. Leading indicators that never correlate to meetings are decoration. Keep them internal. Do not fund them like pipeline.

What should I do this week?

Pull last month's agent activity report. Put meetings held and SQL in the next two columns. If you cannot join them, your attribution write-back is broken. Fix joinability before you renew a seat.

Then pick the loudest activity metric on the team Slack. Kill it as a success metric for thirty days. Replace it with meetings per signal source. Watch how the conversation changes.

Adapt or fail. Labels are not proof. Meetings are.

Start Signals, Convert, Grow

FAQ

Is the $942M figure a GTM number?

No. It is a parallel from TechCrunch's coverage of insurer claims about hospital AI coding. Use it as a pattern warning, not as a GTM dollar claim.

Are all AI activity metrics bad?

No. Activity that predicts meetings can stay. Activity that rises while meetings stay flat is inflation. Prove the link or demote the metric.

What if leadership loves activity dashboards?

Show the same dashboard with meetings and SQL beside each activity line. Make the gap visible. Argue with the join, not with slogans.

How does this fit Grow?

Grow scales what works. Scaling agent activity without a meeting receipt scales cost. Grow owns the refusal to fund inflation.

Should I pause AI SDR experiments?

Pause success metrics that ignore meetings. Keep experiments that log signal, send, meeting, SQL. Rails first. Labels second.