is a bill you pay. How much money are you wasting on AI re-reasoning? Most LLM‑orchestrated workflows have a flaw: they improvise every step — on your dime, at scale, without guarantees. Pega locks logic upfront during design time, ensuring controlled outcomes at a fraction of the cost. Calculate your exposure See how it works The hidden cost of re-reasoning LLM workflows don't just call agents. They rethink everything, every time. To understand how costs compound, here’s a look at how most agentic workflows operate today—and where they fall short. 5-20× LLM-orchestrated workflows cost 5 - 20× more per run than Pega's targeted execution. And the gap compounds with every additional workflow step, as context windows grow and token usage accelerates. Untethered agents Where costs compound Most AI orchestration relies on LLMs to manage workflows at runtime. Each step requires inference to interpret context and decide what’s next—even when logic hasn’t changed. As workflows scale, repeated inference drives up token use and cost. Pega’s predictable approach Efficient and predictable by design On Pega platform, deterministic orchestration executes the workflow, calling AI agents on any platform to perform specific tasks, like processing documents, performing research, or synthesizing content. Logic is defined once at design time. Workflows run efficiently, every time.
Reimagine your workflow now with Pega Blueprint. Try it Your AI Cost Reality Check How much are you
spending today? Configure your workflow scenario below for a quick estimate, or switch to Advanced for granular token and pricing controls. Simple estimate Advanced options Expert Start with a scenario — or set your own below Workflows per month workflows Total steps per workflow steps AI agent steps agent steps Show my annual savings Workflow configuration AI agent steps Total workflow steps Monthly volume (cases) Per agent call — Pega targeted step Input tokens (Pega step) Output tokens (Pega step) Per step — LLM orchestration Context growth / step Output tokens / step Token pricing ($ per million tokens) Input token price Output token price Pega platform cost Price per case ($) Calculate Savings Your estimate Here's what Pega
saves you. Without Pega — annual spend All steps sent to Claude Sonnet ($3/$15 per 1M) With Pega — annual spend Includes $0.60/case Pega platform cost Estimated annual savings Based on 10,000 workflows/month · 40 total steps · 28 agent steps % saved Talk to an expert See the data Estimates use the same underlying model in both Simple and Advanced modes: Pega cost = (agent steps × scoped token cost) + platform fee; AI-native cost = Σ(step i) [2,000 + i × context growth per step] input tokens + output tokens at every step. Default pricing: Claude Sonnet $3/M input · $15/M output. Pega platform fee: $0.60/case. All assumptions adjustable in Advanced mode. Estimates are illustrative. Re-reasoning Over Time Why AI-native orchestration
costs grow quadratically. Orchestration thinking costs grow at every step – and accelerates as context grows. Pega’s deterministic approach to workflow development and orchestration has no such overhead, for dramatic savings potential over time. Cumulative cost per workflow run — step by step 1,000 monthly cases · Claude Sonnet pricing · costs compound as context grows Pega (deterministic + targeted agents) AI-native (re-reasons every step) Step Pega (targeted agents) LLM-orchestrated How we calculated this — full assumptions & methodology Full model — assumptions, token math & formula methodology All values reflect your current calculator inputs. Updates dynamically as you adjust settings above. Download CSV Print / Save PDF Input Assumptions Calculated Results Per-run comparison. Multiply by monthly volume for total cost. Formula methodology Complete the calculator above and click "Show my annual savings" to see your estimate. Token pricing sourced from provider API documentation, April 2026. Default pricing reflects Claude Sonnet ($3/M input · $15/M output). AI-native orchestration cost models context accumulation as an arithmetic series — the cost driver is quadratic with workflow length, not linear. Pega cost reflects targeted LLM calls only at designated agent steps, plus optional platform fee. All assumptions are adjustable. Estimates are illustrative.
This ROI calculator provides estimates only and is intended to help you explore potential outcomes based sample information . The calculations rely on assumptions and averages that may differ significantly from your actual experience. Results are not a substitute for professional analysis, and Pega makes no representations or warranties, express or implied, regarding the accuracy, completeness, or reliability of the output. Past or estimated costs/performance is not a reliable indicator of future results. What counts as a workflow?
One end-to-end process your AI handles — a customer service case, a loan application, an onboarding request, a claims review.
Quick estimate: If your team handles ~200 cases a day, that's roughly 4,000/month. Total steps end-to-end
Every action in the workflow counts — data lookups, decisions, status updates, async waits, and AI agent calls. In an AI-native system, the orchestrator re-reads the full history at every one of these.
Typical range: Simple process ≈ 10–20 · Enterprise case ≈ 30–60 Steps that need AI judgment
Of your total workflow steps, how many actually require an LLM — classification, document analysis, drafting, decision-making? The rest are handled deterministically by Pega at near-zero token cost.
Quick guide: Typically 20–40% of total steps · A 40-step workflow might have 10–15 genuine AI agent steps What are AI agent steps? The number of steps inside your workflow where an LLM is actually called — decisions, classifications, drafting. Non-AI steps like database lookups or rule checks don't count.
Tip: If 28 of your 40 workflow steps involve AI reasoning, set this to 28. Total steps vs. agent steps The full length of your workflow end-to-end, including non-AI steps like data retrieval, rule evaluation, and system calls.
Example: A claims workflow might have 40 steps total, but only 28 of those call an LLM. How many cases per month? The total number of times this workflow runs in a month across all users or customers.
Quick estimate: 200 cases/day × 22 working days ≈ 4,400/month. Input tokens per Pega agent call The number of tokens sent to the model for each targeted Pega agent step. Because Pega locks in context before the run, this stays small and fixed — typically just the task prompt and relevant data.
Typical range: 500–3,000 tokens per call. Output tokens per Pega agent call The number of tokens the model returns for each targeted agent step. Pega's structured outputs keep this concise — usually a classification, a short decision, or a structured JSON blob.
Typical range: 100–800 tokens per call. Why does context grow? In LLM-orchestrated workflows, the model's conversation history grows with every step — each prior action, tool result, and response gets appended. This is what causes costs to compound.
Example: If each step adds ~2,000 tokens of history, by step 20 you're sending 40,000 tokens just in context. Output tokens per LLM step How many tokens the LLM generates in response at each orchestration step. This stays relatively constant per step, but you pay it every step — unlike Pega where only agent steps incur output cost.
Typical range: 200–1,500 tokens/step. What's an input token price? What you pay per million tokens sent to the model (your prompts, context, data). Input tokens are always cheaper than output tokens.
Reference prices (Apr 2026): GPT-4o ~$2.50 · Claude Sonnet ~$3.00 · Claude Opus ~$15.00 · Haiku ~$0.25 — all per 1M tokens. What's an output token price? What you pay per million tokens the model generates in response. Output tokens cost 3–5× more than input tokens because generating text is computationally heavier.
Reference prices (Apr 2026): GPT-4o ~$10 · Claude Sonnet ~$15 · Claude Opus ~$75 · Haiku ~$1.25 — all per 1M tokens. Pega platform cost per case The Pega platform fee charged per workflow execution, on top of raw LLM token costs. This covers orchestration, routing logic, audit trails, and compliance tooling.
Set to $0 if you want to compare pure LLM token costs only, without factoring in platform licensing. Departmental operations 10,000 cases/mo · complex case management · 40 total steps · 28 agent steps Enterprise 100,000 cases/mo · multi-system orchestration · 50 total steps · 30 agent steps Scaled volumes 300,000 cases/mo · large-scale automated processing · 60 total steps · 35 agent steps PARAMETER YOUR VALUE NOTES WORKFLOW TOKEN PRICING PER AGENT CALL (PEGA TARGETED STEPS) LLM ORCHESTRATION (AI-NATIVE, PER STEP) AI agent steps Total workflow steps Monthly volume Input token price Output token price Pega platform cost Input tokens / call Output tokens / call Context growth / step Output tokens / step Steps where an LLM is actually invoked (Pega only) All steps end-to-end. AI-native re-reasons at every one Scales cost linearly for both architectures Claude Sonnet ~$3 · GPT-4o ~$2.50 · Opus ~$15 Typically 3-5× input price Metered licensing — deterministic orchestration + audit trail System prompt + case data scoped to that step only Structured result, classification, or short decision Key lever — appended history grows quadratically. Conservative at 2K; real frameworks often 4K-8K Chain-of-thought + next-step decision per master agent call METRIC PEGA
(DETERMINISTIC + AGENTS) AI-NATIVE
(RE-REASONS EVERY STEP) SAVINGS W/ PEGA Input tokens / run Output tokens / run Total tokens / run Token cost / run Platform cost / run (Pega) All-in cost / run Monthly cost Annual cost Cost multiplier lower cost Pega cost per run = (AgentSteps × InputTokens / 1M × InputPrice)
+ (AgentSteps × OutputTokens / 1M × OutputPrice)
+ PlatformCostPerCase Only designated agent steps invoke an LLM. All other steps run deterministically at near-zero AI cost. Context is scoped to each individual call. AI-native orchestration input tokens per run = Σ (step i = 0 to N−1) [ 2,000 + i × ContextGrowthPerStep ]
= N × 2,000 + ContextGrowth × N × (N−1) / 2 This is an arithmetic series. The N×(N−1)/2 term makes cost grow quadratically — not linearly — with workflow length. A workflow twice as long costs roughly four times as much to orchestrate. At step 20, the master agent is re-reading everything from steps 1–19 just to decide what step 20 should be. AI-native total cost per run = (OrchestratorInputTokens / 1M × InputPrice)
+ (TotalSteps × OutputTokensPerStep / 1M × OutputPrice)
+ same targeted-agent costs as Pega The orchestration layer is pure overhead on top of the agent calls both architectures share. The gap widens with every step added to the workflow. steps tokens cases case