What you are paying for
Three line items on a Synarsi engagement. We separate them on purpose, so you can see what changes with workflow complexity and what does not, and so the invoice never has a surprise line.
1. Running costs (tools and tokens)
What the live agent consumes every month: model API calls, vector store hosting, third-party integration tooling, the orchestration platform. It scales with workflow complexity and integration count, not with your headcount. For a beginner / simple workflow with one or two integrations, this lands around $240 per month. Think: a fintech in Manchester with one Snowflake integration running a reconciliation agent. Push the same workflow to a national insurer's FNOL queue with Guidewire, DocuSign, the underwriting platform, and an external sanctions API, and you are at $1,200 to $2,500 per month. We tell you that number before the build, not after the first invoice.
2. Build and integration fee
The one-time cost of designing, building, instrumenting, and shadow-running the agent through cut-over. Quoted per workflow against the scope from phase one of the methodology. Price tracks three things: integration surface area, regulatory constraint, escalation complexity. Not your industry, your headcount, or whether "AI" appears in your job titles. A structured-document workflow with one integration (an HR team parsing standardised offer letters into the ATS) costs a fraction of an FNOL intake build at a regulated insurer with Guidewire, claims handler routing, and a written audit trail for every triage decision.
3. Engagement fee (optional, ongoing)
After cut-over, most teams want us on a low-touch retainer for the first 60 to 90 days, the window when trust calibration is observational and escalation thresholds are still drifting. We run weekly review calls, monthly threshold recalibration, and on-call cover if a severity-flag pages overnight. After 90 days teams usually move to a monthly cadence or hand it back to their internal operator. We do not lock anyone into multi-year retainers. The agent is the deliverable, not the relationship.
Where most teams start
The most common shape on a first engagement: one workflow, one or two integrations, beginner / simple complexity. Running cost ~$240 per month in tools and tokens. Build and integration fee scoped per workflow. Concretely: an insurance ops team in Birmingham running an FNOL triage agent against Guidewire and an email inbox; a property manager in Leeds running a lease abstraction agent against PDFs and the property management system; a recruitment ops team running a top-of-funnel CV screen against the ATS. The second workflow comes once the first is paying back.
How we quote
No calculator on this page, and we will not hand you three pricing tiers in a coloured table. Pricing for the build phase depends on three things:
- The workflow itself. What it is, how many cases per week, what systems are involved (Salesforce, Guidewire, Epic, ServiceNow, SAP, the homegrown system the last engineer who understood it left in 2022).
- The constraint that bounds it. The regulator, the playbook, the audit requirement, the customer expectation. "Nothing weird ever happens" is not a constraint; "the FCA looks at every escalated case" is.
- The integration surface area. How many systems the agent reads and writes, whether they have real APIs, what the auth and rate-limit story is, and whether anyone on your side can grant a service account inside two weeks.
A 20-minute conversation is enough to give you a quote range you can take back to procurement. If a vendor needs four weeks of scoping to give you a price, they are sizing the contract, not the workflow.
What you will not be charged for
No discovery report: the output of scoping is a one-page scope, included in the build fee, not a 40-page slide deck with a maturity model. No per-seat licensing: the agent is yours, one operator or ten or the whole department, the price does not change. No pilots that go nowhere: we do not sell pilots without a written go/no-go gate, the reasoning is here. No slide decks as a deliverable, the deliverable is a running agent. No travel and expenses on top: if we need to be on site, it is in the build fee.
What changes prices most
Two factors move prices more than anything else, and both are knowable on a first call.
Integration access. If every system has a real API with a service-account auth model, the build is roughly half the cost of one where two systems require screen-scraping. We will tell you which group your workflow is in before quoting, and if the answer is screen-scraping for a critical system, we usually recommend not building.
Regulatory load. A workflow that touches PHI, financial transactions over reporting thresholds, FCA-supervised activity, FDA-supervised activity, or anything an external auditor reviews carries a higher audit-log specification, escalation complexity, and review cadence. Automating a banking reconciliation an FCA review will sample is not the same job as automating an internal marketing report.
Frequently asked
Is there a fixed package?
No. Fixed packages mean either you overpay for what you do not need or we underbuild what you do. Procurement would rather see a quote that matches the work than a SKU that matches a catalogue.
Do you take equity or rev-share?
No. Cash only. Equity and rev-share misalign incentives. We would be working for scope creep instead of for shipping, and that is the wrong incentive in this category.
What is the smallest engagement you take?
A single agent on a single workflow with one or two integrations. Roughly $240 a month in running cost plus a quoted build fee. This is the right starting size for almost every team's first build. The boring-workflows-pay-back-fastest argument covers why.