How to Calculate SaaS Seat Compression After AI Agent Deployment

A 10,000-employee enterprise with 250 SDRs paying $100/seat for Outreach and $49/seat for Apollo (catalog mid-market rates; list prices are higher) generates per-tool gross savings of $9,950/month or $119,400/year against AiSDR's compression. Subtract AiSDR's year-1 cost of $25,800 (the $10,800 subscription plus the $15K services-overhead default for flat-fee agents) and year-1 net is $93,600. Apply the dashboard's 0.4 realization factor to that year-1 net (not steady-state) and the defendable line item is $37,440/year. Steady-state at year 2+ climbs to $119,400 − $10,800 = $108,600/year as setup amortizes. That $37,440 is the number you should bring to your CFO meeting — not the gross savings the vendor pitch quotes.

Most AI agent vendors will quote you the gross. That's why the math matters.

Try the free calculator — 15 seconds, no signup.

The problem: vendor pitch math is theoretical max

Every AI agent vendor leads with the same slide: "We replace 60% of your tier-1 tickets" or "We compress 55% of SDR seats." Those numbers are real. They're also the ceiling, not the floor.

Four things compress that ceiling fast:

1. Agent cost is a real line item. AiSDR is $900/month flat. Sierra is $6,000/month plus $35K setup. Decagon is $5,000/month plus $25K setup. Glean is $6,000/month platform plus $60/user (enterprise rate $45/user above the 100-seat threshold) plus $50K setup. None of these are free, and below an unlock threshold they cost more than the seats they replace.
2. Year-1 includes setup. Flat-fee enterprise agents always have a services overhead — sometimes explicit (Sierra $35K, Decagon $25K, Glean $50K), sometimes implicit (we apply a $15K default to flat-fee agents that don't disclose). The dashboard's year-1 viability math folds this in. Most vendor decks omit it.
3. Year-1 realization is much lower than the slide implies. Internal tooling rollouts, edge-case handling, change management, and the inevitable "we still need a human for that" exceptions chew through 60% of the theoretical savings in the first 12 months (the dashboard's realisticActionableSum uses exactly 0.4 as the year-1 realization factor for deploy_agent actions). This is industry rule-of-thumb territory — not a number we've back-tested against your specific data — but every CFO we've spoken to confirms it directionally.
4. Pricing model gates whether seats compress at all. Per-seat tools (Salesforce, Outreach, Slack, Zendesk) shrink with utilization. PEPM tools (Workday, BambooHR), per-endpoint tools (CrowdStrike), and usage-billed tools (Datadog, New Relic, Snowflake) don't. If your "seat compression" target is Workday, the math is zero before the agent cost even shows up.

A recent BVP State of the Cloud report put public-cloud SaaS spending growth at double-digit rates even as macro budgets tighten. The gap between "what you bought" and "what gets used" is widening alongside that spend (Productiv and Zylo SaaS Management Index reports document the per-seat utilization side directly), and AI agents make the gap worse — not better — if you don't run the math correctly.

The methodology: a five-step formula

Here's the framework SeatCompress runs on every analysis. It's the same one you should run before any AI agent purchase.

Step 1 — Per-tool gross savings.

gross_savings_monthly_per_tool = compressionPct_per_tool × activeSeats × monthlyCostPerSeat

Critical: per-tool, not blended. AiSDR's catalog compression on Outreach is 30%, on Salesloft is 30%, on Apollo is 20%, on ZoomInfo is 20%, and on Zoho CRM is 25% (AiSDR does not target Salesforce — that's a different agent class). You do not multiply a vendor's headline percentage by the combined stack cost — you multiply each tool's per-seat cost by that tool's compression rate, then sum across tools the agent targets.

Note also activeSeats, not contractedSeats. If you have 100 Outreach seats contracted but only 60 active users, the agent compresses 60 — you can't compress what isn't there. (The other 40 seats are classic shelfware. That's a separate renegotiation lever, covered in our shelfware post.)

Step 2 — All-in agent cost (year-1 vs steady-state).

For per-user agents — note many enterprise agents (Cursor, Lavender, Glean) are actually dual-priced with both a flat platform fee and a per-user line; use the per-user-only path for the genuinely user-priced ones (Sourcegraph Cody, CodeRabbit, Continue, Qodo, Lindy):

agent_cost_monthly = costPerUser × seatsDeployed
agent_cost_year1   = agent_cost_monthly × 12 + explicit_setup_cost

For flat-fee agents (Sierra, Decagon, AiSDR, Intercom Fin's notional-monthly):

agent_cost_monthly = monthlyCost
agent_cost_year1   = monthlyCost × 12 + max(15_000, monthlyCost) for unspecified setup
                   OR monthlyCost × 12 + explicit_setup_cost

Use steady-state for years 2+, year-1 for the first-year defensible number.

Step 3 — Net savings (steady-state and year-1).

net_savings_steady = sum(gross_savings_monthly_per_tool) × 12 − (monthlyCost × 12)
net_savings_year1  = sum(gross_savings_monthly_per_tool) × 12 − (monthlyCost × 12 + setupCostUsd)

If steady-state is positive but year-1 is negative, you're in the messy middle — viable as a 24-month commitment, harder to defend on a 12-month payback gate. (setupCostUsd is explicit when known — Sierra $35K, Decagon $25K, Glean $50K; for flat-fee agents without a disclosed setup we use max($15K, monthlyCost) as the conservative default. Per-user agents like Cursor and Lavender default to $0.)

Step 4 — Realization discount.

year_1_realistic = max(0, net_savings_year1) × 0.4

This is the formula the dashboard's realisticActionableSum runs. The 0.4 factor is applied to year-1 net (which already includes setup), not to steady-state. That distinction matters: a sub-unlock deployment with a negative year-1 net produces $0 defendable, even though steady-state would still pencil. We use 0.4 as the default because half the bias should be on the side of "this won't be as good as the vendor says" and the remaining 60% shows up over years 2–3 as deployment matures, edge cases get automated, and the team rebuilds workflows around the agent. The renegotiate sibling factor (Layer 1 contract drops, no agent involved) is 0.5.

Step 5 — Pricing-model gate. Before anything else, gate the math on the SaaS tool's pricingModel. Only per_seat rows produce compression. Skip PEPM, per_endpoint, and usage rows entirely — those compress via rate negotiation, not seat drops.

Try the free calculator — 15 seconds, no signup.

The MAX-overlap rule: when two agents target the same tool

Real-world catalogs have multiple agents that compress the same tool. Decagon and Sierra both target Zendesk. Artisan's Ava and Regie.ai both target Outreach. Cursor Business and GitHub Copilot Enterprise are head-to-head substitutes (not "compressors" of GitHub Enterprise — that's a conceptual confusion worth naming).

The rule: never sum the compression percentages.

If Decagon compresses Zendesk by 65% and Sierra compresses Zendesk by 60%, the combined compression is 65%, not 125%. Both agents are doing roughly the same job; deploying both doesn't double the compression — it duplicates the cost. The right move is to pick the higher-compression agent (Decagon at 65%, which sits at the vertical_replacement cap) and treat the second as redundant.

This is the MAX-overlap rule, and it's the single most common mistake we see in vendor-built ROI calculators. They sum, because summing makes the savings number bigger. SeatCompress takes the max, because that's what actually happens.

There's a contrarian read here: deploying a second agent on the same tool is almost never worth it. The first agent caught the easy 55–65%. The second agent's marginal contribution is the 5–10% the first one missed — usually not enough to clear that agent's own monthly fee plus its own setup cost. CFOs who approve "AI portfolio" purchases of three or four agents per category are usually paying for two redundant tools.

A real example: 10,000-employee enterprise, sales stack

Let's run the math on a concrete scenario. 10,000-employee enterprise SaaS company. 250 SDRs in the sales org. We use catalog mid-market per-seat prices throughout. (Smaller deployments — under ~54 SDRs at these per-seat rates — won't clear AiSDR's year-1 unlock threshold once you include the $15K services-overhead default; for those, the right move is to wait or use a per-user agent like Lavender that scales down cleanly.)

Stack inputs:

• Outreach: 250 SDR seats × $100/seat = $25,000/month
• Apollo: 250 SDR seats × $49/seat = $12,250/month
• Total compressible SDR-tool spend: $37,250/month, or $447,000/year.

(Salesforce is in the broader sales stack but AiSDR doesn't compress it — that compression target belongs to a different agent class like Salesforce Agentforce at 30% or 11x Alice at 30%. Don't blend it into the AiSDR math.)

Agent: AiSDR.

• Pricing: $900/month flat fee, $10,800/year subscription, year-1 effective $25,800 (with the $15K services-overhead default for flat-fee agents)
• Compression targets: Outreach 30%, Apollo 20% (per-tool catalog rates)

Step 1 — Per-tool gross savings:

Outreach:    250 × $100 × 0.30 = $7,500/month
Apollo:      250 × $49  × 0.20 = $2,450/month
TOTAL gross savings:           = $9,950/month  →  $119,400/year

Step 2 — Agent cost: $10,800/year steady-state, $25,800/year in year 1.

Step 3 — Net savings:

Steady-state: $119,400 − $10,800 = $108,600/year
Year-1:       $119,400 − $25,800 = $93,600/year

Step 4 — Year-1 realistic (apply 0.4 discount to year-1 net):

year_1_realistic = $93,600 × 0.4 = $37,440/year

That's the honest year-1 defendable number on a 250-SDR deployment with the Outreach + Apollo stack: $37,440. Not the $119,400 a vendor slide might show on gross. Steady-state at year 2+ climbs to $108,600/year. Defend year-1 internally; budget the longer-horizon win against the 24-month plan.

Now scale up to 600 SDRs at the same per-seat costs (a Fortune-100 outbound org with a $1.5M+ sales-stack line item) and re-run: per-tool gross becomes $9,950 × (600/250) = $23,880/month or $286,560/year. Year-1 net: $286,560 − $25,800 = $260,760. Year-1 realistic: $104,304/year. Steady-state: $275,760/year. The agent's economics improve nonlinearly with scale — that's the unlock threshold playing out in numbers, and it's why this product is built for enterprise SDR teams, not 25-rep mid-market sales orgs.

This particular example tracks closely with our AiSDR + Salesforce deep dive, which walks through the same math at three different headcount levels.

Try the free calculator — 15 seconds, no signup.

How to apply this to your stack this quarter

Five steps. None of them require a platform.

1. Build a per-tool spreadsheet of your top 10 SaaS contracts. Columns: tool name, contracted seats, active seats (logged in last 30 days), per-seat cost, pricing model (per_seat / PEPM / per_endpoint / usage), annual contract value, renewal date. The pricing model column is the gate — only per_seat rows are candidates for AI agent compression.

2. For each per-seat row, identify the candidate agent. Use catalog defaults to start (we keep ours public on the AI agents that replace SaaS seats post). Don't trust a vendor's compression number that's higher than the catalog default — they're optimizing for their pitch, not your reality.

3. Run the five-step formula for each (tool, agent) pair, using per-tool compression rates. Be honest about whether the agent actually targets the tool at all, and whether the compression target is the full per-seat cost or only a subset. AiSDR doesn't compress Salesforce — that's a different agent class (Salesforce Agentforce at 30%, 11x Alice at 30%). AiSDR compresses Outreach at 30% and Apollo at 20% specifically (the prospecting subset of the SDR workflow). Multiply each tool by its own rate, then sum — and don't fabricate a row for a tool the agent doesn't appear against in the catalog.

4. Apply the MAX-overlap rule when two agents target the same tool. Pick the higher-compression option. Treat the second as redundant unless its marginal compression on the gap is large enough to clear its own all-in year-1 cost — usually it isn't.

5. Sum the year-1 realistic numbers across your stack. That's the dollar amount you can credibly defend to your CEO and CFO. If a vendor shows you a bigger number, ask them which step of the formula they skipped. (Usually it's step 2 — they skipped year-1 setup. Sometimes it's step 1, where they used contractedSeats instead of activeSeats to inflate the base, or blended a single compression rate across the whole stack instead of per-tool.)

This won't catch everything. It won't model second-order effects like "Cursor makes engineering 20% faster which means we hire one fewer engineer next year." It won't capture the operational overhead of running three different AI agents in parallel, each with its own admin console and audit log. And the 0.4 realization factor is industry rule-of-thumb — not a number we've calibrated against your specific deployment data, because that data doesn't exist yet for most companies. Treat it as a starting point, not a contract term.

Gartner's $1.5T global AI spending forecast for 2025 puts enterprise AI agent budgets on a steep upward curve. The CFOs who win this cycle aren't the ones who block agent purchases — they're the ones who run the math correctly and approve the agents that clear the five-step gate including the year-1 setup line.

The bottom line

Per-tool gross savings (not blended) minus all-in year-1 agent cost (including setup) minus realization discount, with a MAX-overlap rule on multi-agent overlaps and a hard pricing-model gate on which tools are even eligible. That's the formula. It's not complicated, but most vendor calculators skip at least three of the five steps because doing the math correctly makes their product look smaller.

Run the formula yourself before any agent purchase. If you want the five-step gate automated against your stack — including per-tool compression, year-1 setup amortization, the MAX-overlap rule, pricing-model branching, and the year-1 realization discount — SeatCompress does it in 30 seconds.

Try the free calculator — 15 seconds, no signup.