Selling agents as services with success-based pricing

Restructures how an AI-product company prices and delivers value: instead of charging a per-seat SaaS fee (which fights the customer's natural AI-substitution motion), package agents as services with success-based pricing tied to a customer-meaningful outcome. Two-tier model at one reference deployment: (1) a copilot tier ($5–15K/mo at the reference scale) where the agent connects to client systems, runs analyses, generates rules; the human at the client still owns; (2) an outsourced agent-driven service (high prices, success-fee economics) where the AI company takes responsibility for the outcome and ~90% of the work is done by agents, with ~1 customer-success person per client. The pricing pattern: "Don't charge 3× your old tool — charge $1 per ticket when the customer pays $10/ticket. 10× cheaper for them, you keep margin because agents do it."

Two structural changes have happened simultaneously: (a) AI software directly makes or saves money in ways the customer can measure (ad spend allocated, tickets resolved, deals won, claims processed), so paying per-outcome is now legible; (b) the customer's instinct is to substitute your tool with their own agent the moment your seat price gets noticeable. Success-based pricing converts both into your advantage. You ride the customer's value capture; the customer pays only when they win; the customer can't easily DIY because your moat is the catalog of tested workflows + tuned rules + connected data + ML on top of rules. Same reference deployment reports new business lines (paid AI consulting, paid AI services for clients) opened by this packaging shift.

1. Pick one measurable outcome. Examples by vertical: per-ticket resolved (support), per-incremental-conversion (sales), per-dollar retained (retention), per-shipped-MR (engineering), per-correct-claim-decided (insurance), per-incremental-installed-base (acquisition). Must be (a) traceable to your agent's action, (b) already metered by the customer's existing systems, (c) measurable in a billing cycle.
2. Establish the baseline. With each pilot customer, agree the non-AI baseline in writing for 4–8 weeks: current conversion %, current cost-per-ticket, current churn rate. Without a baseline, "success" is a fight.
3. Two-tier packaging.

• Copilot tier: flat fee ($5–15K/mo at reference scale), unlimited usage, the agent connects to the customer's systems, the customer's team still operates. Sell as "10× speed-up of your existing team." Land it with department heads (CMOs, VPs of growth), not the operators (media buyers, support agents) — operators often resist.
• Outsourced tier: success fee per outcome. Common shapes: $X per resolved ticket (typically 10–20% of customer's current cost-per-ticket), Y% of incremental revenue attributable to AI-generated changes (cap at 25–30% of net new), $Z per closed-won deal. Bake in floors (minimum monthly) and ceilings (cap to control your downside in a great month).

4. Hire from the industry. Performance marketers selling to performance marketers; doctors selling to doctors. Retrain on AI in ~8 weeks; the domain credibility unlocks the conversation that pricing can't.
5. Build the inside structure. Layers = data → product → agents → tools → skills → automations. Each layer is "the same problem the enterprise already pays someone to solve," unified under one runtime. (See token-metabolism #155 and one-prompt operations #156.)
6. Phase the customer. Most won't sign for outsourced on day 1. Standard path: Copilot 3–6 months → metrics prove → Outsourced for one workstream → expansion.
7. Build the attribution machinery. A read-only side-car that records every agent action and links it to the outcome event (resolved ticket, won deal). This is what makes invoices defensible.
8. Open new business lines. Paid AI consulting (inbound is already happening to firms that publish about their transformation), paid AI training for clients, paid AI services for clients. Same skill set, three SKUs.

Sample success-fee contract shape (skeleton — get a lawyer; this is for orientation):

1. Service. {Provider} delivers AI-driven {workstream, e.g., customer support
 tier-1 resolution} to {Client}'s {customer base / inbound queue / channel}.

2. Baseline measurement period (4 weeks). During the Baseline Period, the
 parties record:
 - current cost per {unit} ("Baseline Cost"): __________
 - current resolution / conversion / outcome rate ("Baseline Rate"): __________
 - {1–2 other anchoring metrics}

3. Pricing.
 3a. Service fee: {currency}__/{unit} × number of {units} where Provider's
 AI action was the proximate cause. Floor: {currency}__/month.
 Ceiling: {currency}__/month.
 3b. Success fee: __% of Incremental Value, where Incremental Value =
 (Baseline Cost − Realized Cost) × {units} in the billing cycle,
 net of agent operating costs.

4. Attribution. Provider operates a read-only attribution log capturing every
 agent action and its outcome event. Client retains read access to this log
 for the duration of the agreement plus 18 months.

5. Customer-success staffing. Provider assigns one named Customer Success
 Engineer per Client for the duration of the agreement.

6. Out-clause. Either party may terminate with 60 days' notice if the
 Realized Rate falls below {threshold} for 8 consecutive weeks.

7. Data. Client data remains Client's property. Provider's agent outputs and
 tuned rule sets remain Provider's property. {Client may license a tuned
 ruleset for use after termination at a fee of {currency}__.}

8. Audit. Annual third-party audit of attribution log, paid by Provider,
 shared with Client.

Outbound pitch (for a CMO/VP-of-growth on ad spend optimization):

On $1M/mo ad budget, 20% efficiency improvement = $200K/mo.

We sell two ways:
1) Copilot — your team uses our agent. $10K/mo, unlimited. Typical lift
 we see in pilot: 8–15% inside 90 days.
2) Outcome — we take the workstream. We charge 25% of incremental savings
 versus your trailing-12-week baseline. Cap at $50K/mo.

Either way, we keep one CSM per account. You can move from (1) to (2) any time.

Pilot: 4 weeks copilot + 4 weeks baseline + 8 weeks outcome (optional).
We'll write the baseline and the attribution log together.

Reverse-margin pricing rule of thumb:

Your price per outcome = (customer's current cost per outcome) × 0.10–0.20
Your margin per outcome = your price − (your token cost + CSM allocation)
Target gross margin: ≥ 75% on the outsourced tier.

• Operators resist; sell to the head. Reference deployment finding: media buyers themselves often resist; CMOs/VPs of growth are excited. Tailor the motion accordingly.
• Baselines are the contract's real heart. A success fee on an unclear baseline is unenforceable. If the customer won't agree a baseline in writing, walk.
• Don't price below your token cost in a bad month. Floors protect you from "they had a bad month, we worked twice as hard."
• Watch for AI substitution. As foundation-model costs drop, expect customers to attempt DIY. Your moat is the catalog (workflows + tuned rules + connected data + ML on rules), not the LLM call. Publish enough to be the trusted vendor, withhold enough to be hard to clone.
• Don't enter regulated verticals (healthcare, finance) with success-fee until you have audit-grade attribution. The legal exposure on misattributed savings is real.

• Pure copilot (lighter): Start here. Get to 10 paying customers on flat-fee. Use those to figure out outcome attribution before you sell outcomes.
• Heavier (full outsource): AI agency model — you run the workstream end-to-end, customer never sees the agent. Used by emerging firms in compliance, support, and outbound. ~1 CSM per ~5–10 client accounts at maturity.
• Hybrid with tooling: Charge per-outcome AND license the tuned rule set after termination (Section 7 of the skeleton contract). Lets you give the customer a graceful exit while preserving a stream.