Granola, Read AI, Otter, Fireflies: Why You Only Need One Meeting Agent

Your VP of Sales bought Gong. Your VP of Customer Success bought Otter. Your Head of Product bought Granola because the designer said the UI was nice. Your CTO has Read AI running on enterprise Zoom calls. And IT just renewed Fireflies because someone in marketing put it on a credit card eighteen months ago. You are paying four vendors to do the same job — transcribing meetings — and on a stack-level seat-compression model, only one of them gets credit.

The four agents all hit the same number

Pull the SeatCompress catalog and the picture is almost embarrassingly clean. Four meeting-AI agents have approved AgentToolImpact rows against Zoom, and three out of four hit the exact same compression percentage:

Twenty-to-twenty-five percent. These are seeded AgentToolImpact values — most entered as judgment passthroughs before the deriveCompressionPct cap-clamp algorithm shipped, which is why they exceed the strict 0.20 augmentation cap that any new sourced claim would have to clear. Meeting-transcription agents don't replace Zoom; they sit on top of it. They make the same calls more useful. That's why every one of them lands in the augmentation band rather than the 0.65 vertical-replacement band where Decagon and Sierra live for Zendesk.

The CFO question is not "which of these is best." The CFO question is: how many of them is your finance system willing to pay for, given that the seat compression credit is bounded by the MAX of the group, not the sum?

The MAX-overlap rule, in one paragraph

When multiple AI agents target the same SaaS tool, the SeatCompress analysis engine uses Math.max(...impacts.map(i => i.pct)) — not the sum. Two agents that both claim to compress Zoom by 25% don't compress it 50%. They compress it 25%, full stop. The second agent is paying for itself out of zero incremental seat savings.

This rule exists because the underlying physics demand it. The "compressible seats" on a Zoom license are users who are in meetings that an AI summarizer can replace, partially. There is one population of those users. Three different products fighting over the same population doesn't multiply the population. Adding Fireflies on top of Otter on top of Granola doesn't let you cancel 75% of your Zoom seats — you've already accounted for the 25% the first time.

We wrote this up in detail in Why "unused seats" is the wrong metric in 2026, but the short version is: stack-level math is the only math a CFO should care about. Per-vendor ROI decks live in fantasy.

What you're actually paying for the redundancy

Here is the per-seat list on each agent at 50 seats deployed:

• Granola Business: $14/user. Flat at every scale. No enterprise breakpoint published.
• Otter Business: $20/user, dropping to $14/user at the 50-seat enterprise threshold.
• Fireflies Business: $19/user, dropping to $15/user at the 100-seat threshold.
• Read AI: $19.75/user, rising to $29.75/user above 10 seats (different pricing topology — the "enterprise" tier is more expensive, not less, because it bundles compliance features).

At 500 deployed meeting-takers — call it a sales org plus a CS org plus product leadership at a 12,000-employee SaaS company — your annual cost on each is:

• Granola: 500 × $14 × 12 = $84,000/yr
• Otter: 500 × $14 × 12 = $84,000/yr (past the 50-seat enterprise threshold)
• Fireflies: 500 × $15 × 12 = $90,000/yr (past the 100-seat threshold)
• Read AI: 500 × $29.75 × 12 = $178,500/yr

If three of those four are running concurrently in different departments — the realistic state at most large enterprises we audit — you're paying somewhere between $250K and $350K per year for redundant transcripts. The MAX-overlap rule says the seat compression credit caps out at whatever one of those agents would have earned on its own.

The buying decision collapses to two variables

Once you accept that you pick one, the decision tree shrinks to two questions:

1. What's your effective per-seat price after the enterprise breakpoint?
2. Does the agent compress anything besides Zoom?

Per-seat price at your deployment size

Below 50 seats, Granola wins on raw price at $14/user flat — no threshold gating. Otter is $20 list, Fireflies is $19 list, Read AI is $19.75 list. If you're piloting with one team — say, a 25-person sales org — Granola is the obvious default.

Between 50 and 99 seats, Otter Business matches Granola at $14/user because the 50-seat enterprise threshold has kicked in. Fireflies is still at $19/user list (its threshold is 100, not 50), so Fireflies loses this band cleanly.

Between 100 and 499 seats, Otter at $14 still leads but Fireflies at $15 is close enough that the second variable (multi-tool compression) starts mattering more than the dollar difference.

Above 500 seats, all three (Granola, Otter, Fireflies) are within $1/user of each other and the decision is no longer about price. It's about what else the agent does.

Multi-tool reach

Granola's AgentToolImpact row list, in our catalog, contains exactly one entry: Zoom 0.25. That's it. Granola is a one-tool agent. If your only telephony/conferencing surface is Zoom, that's fine — the simplicity is a feature.

Otter compresses Zoom 0.25 plus RingCentral 0.20, Dialpad 0.20, and Loom 0.15. If your sales org runs Dialpad and your support org runs RingCentral, Otter is capturing compression credit on three tools where Granola captures it on one.

Fireflies sits between them: Zoom 0.25 plus RingCentral 0.20 and Dialpad 0.20 — no Loom coverage.

Read AI compresses Zoom 0.20 (not 0.25), but adds Microsoft Teams 0.20 and Slack 0.15. If your enterprise is Teams-centric — which most enterprises with 5,000+ employees actually are, despite what your sales org tells you — Read AI is the only agent in the group that touches your real conferencing surface.

This is the buying decision in one line: single Zoom shop → Granola. Mixed Zoom + telephony → Otter. Teams-first → Read AI. Fireflies is dominated in every scenario by one of those three.

Worked example: a 12,000-employee SaaS company

Imagine a 12,000-headcount enterprise software vendor. Sales is on Salesforce ($100/seat) and Outreach ($100/seat). Support runs Zendesk ($115/seat) and Intercom ($85/seat). Engineering is on Jira ($8.15) and Confluence ($5.16). The conferencing stack: Zoom across 4,800 seats at $18.33/seat/mo and Microsoft Teams across the same 12,000 employees at $12.50/seat/mo because nobody ever got fired for buying both.

The current meeting-agent state, audited: Granola at 320 seats in product and design ($53,760/yr); Otter at 410 seats in sales and CS ($68,880/yr at the $14 enterprise rate); Fireflies at 180 seats in marketing ($32,400/yr at the $15 enterprise rate, past the 100-seat threshold); Read AI on a 30-seat exec pilot at the $29.75 enterprise rate ($10,710/yr).

Combined annual spend on meeting agents: ~$165,000.

Now run the SeatCompress engine. For the Zoom line item — 4,800 seats × $18.33 × 12 = $1,055,808/yr — the MAX-overlap rule says: the compression credit is max(0.25, 0.25, 0.25, 0.20) = 0.25. The engine attributes $263,952/yr of theoretical compression credit to Zoom, then discounts by the year-1 realization factor of 0.4 on deploy_agent actions to get $105,580/yr of realistic year-one savings.

That's the credit. One time. Not four times. Whether you have one meeting agent or four, the Zoom compression line in the analysis is the same number.

So the spend math: $165,000 paid out, $105,580 in realistic year-one Zoom seat compression. You're $60K underwater on your meeting-agent portfolio, before you've even talked about whether the productivity story actually shows up in the calendar.

The fix: consolidate to Otter at the $14 enterprise rate across 600 seats (combining the existing Granola, Otter, and Fireflies populations). Annual cost: 600 × $14 × 12 = $100,800/yr. Same MAX-overlap credit on Zoom ($105,580 realistic). Now you're +$5K and you have one vendor invoice instead of three. The Otter pick also captures secondary compression on Dialpad and RingCentral if those exist in the stack — pure upside the multi-vendor sprawl was burying.

If you want the math on your own stack, the seat compression calculator will run it on whatever tool inventory you paste in.

Where the catalog stops and judgment starts

A few caveats so the math is honest.

The 0.20 augmentation cap is real. The deriveCompressionPct algorithm caps every augmentation-class compression claim at 0.20 — even when a vendor case study shows 40% or 60%. That cap is documented in why we source every compression percentage. The reason the values you see in this post are 0.25 not 0.20 is that several of these were entered as judgment passthroughs in the seed before the cap-clamp algorithm shipped. The next catalog audit will likely revise some of them down. The buying conclusion doesn't change — these agents still all hit roughly the same band.

Read AI's lower Zoom compression (0.20 vs 0.25) is a data choice, not a product judgment. Read AI markets itself as a meeting-intelligence platform with stronger analytics; the lower compression number reflects that it doesn't displace Zoom's recording / playback functionality as cleanly as a pure-transcription agent like Otter. We could defend a 0.25 reading. We could also defend a 0.18 reading. The point of the cap-clamp algorithm is that these distinctions matter at the margin, not at the buying decision.

None of these agents compress Microsoft Teams the way Otter compresses Zoom. Teams has its own first-party Copilot story — Microsoft 365 Copilot at $18/user, currently with zero approved AgentToolImpact rows against Teams in our catalog because the AI-vs-AI cleanup pulled them. If your conferencing is Teams-first, the meeting-AI compression story is materially weaker than the Zoom story until first-party Copilot earns a defensible compression number through customer telemetry. That's a different blog post.

What the CFO does Monday morning

1. Pull the GL line items for every meeting/transcription/note-taker tool. You're looking for Granola, Otter, Otter for Business, Fireflies, Read AI, Krisp (adjacent), Avoma (technically a different class — sales coaching — but often confused), Gong (sales coaching, definitely different), and any line item with "AI notes" or "meeting AI" in the description.
2. Map each one to a deployed seat count and the active conferencing tool it sits on. Zoom seats, Teams seats, Dialpad seats, RingCentral seats. The agent is only earning compression credit against the tools it actually integrates with.
3. Apply the MAX rule per conferencing tool. Whichever agent shows up multiple times against the same conferencing tool — pick one. The others are paying duplicate rent.
4. Use the enterprise breakpoint as your tiebreaker. Otter at 50 seats. Fireflies at 100 seats. Granola has no breakpoint to wait for. Read AI's enterprise tier is more expensive, not less — be careful.
5. Consolidate before the next renewal cycle. If three of these have renewals within 90 days of each other (common — Q1 and Q3 are the procurement peaks), the renegotiation lever is straightforward: pick the survivor, negotiate the enterprise rate on the consolidated seat count, kill the rest at notice period. The renegotiation playbook anatomy post walks through the script.

The point isn't that one of these agents is better than the others. The point is that the seat-compression math only credits you for one of them, and your finance system isn't structurally able to tell you which one. That's what this calculator exists to fix.

If you're paying for more than one meeting agent right now, you're paying twice for credit you can only earn once. Pick one. Cancel the rest at renewal.