Tea terraces on a steep hillside in early morning shade

Take AI from pilot to production, one gate at a time.

Terrace runs the enterprise AI program: a portfolio that climbs through stage gates, a registry of every production agent, governance that keeps pace, and realized value by business unit.

Book a walkthrough Open the demo workspace

Software for enterprise AI programs.
Stockholm and New York.

Every AI initiative
on one staircase.

Strategy, Pilot, Production, Operations. Each column is a stage and each move up is a gate review. Drag a card up a step and Terrace books the review, gathers the evidence and asks the right approvers.

Open the portfolio
42initiatives on the staircase at Northmere Group
17live in Production or Operations
27agents in the registry, each with an owner
$16.4Mrealized in FY26 to date, signed off by Finance

AI programs run on Terrace

Four steps.
One program.

  1. 01 · Strategy

    Ideas get an accountable owner, a value hypothesis sized with Finance and a risk tier before anyone builds.

  2. 02 · Pilot

    A time-boxed build against success criteria agreed up front. Evals run on your own cases, not a vendor demo.

  3. 03 · Production

    Live with real users, human oversight and a rollback plan. Every agent enters the registry on day one.

  4. 04 · Operations

    Handed to the business with a runbook and an on-call owner. Value is measured against the baseline every month.

Rice terraces seen from directly above, curved walls forming stepped bands

Pilots stall when nobody owns the next step. Terrace gives every initiative an owner, a gate and a date.

Gate reviews with evidence, not slide decks.

Each gate has a checklist your risk, finance and security teams agree once. Evidence attaches from the systems that hold it: eval runs, the signed value case, the privacy assessment, the pen test.

7 checks
for the production gate, editable per business unit
4 approvers
sign in Terrace, with comments kept in the decision log
1 export
to your audit trail, with every artifact linked

Every production agent,
with an owner and an off switch.

The registry lists each agent with its initiative, model, autonomy level, accuracy against target, runs and monthly cost. When something drifts, the owner hears first and the kill switch is one confirmed click away.

Open the registry
AgentInitiativeOwnerStatusAccuracyRuns a weekCost a monthKill switch
Motor FNOL Voice AgentAG-28 · v1.0Motor first notice by voiceMei TanakaLive96.2%2,100$2,300
Claims Intake TriageAG-01 · v3.2Claims intake triageMei TanakaLive97.8%48,200$14,300
Call Summary WriterAG-08 · v3.4Call summariesOliver GrantLive97.5%71,200$9,600
Policy Q&A AssistantAG-06 · v5.0Policy Q&A for the contact centreOliver GrantLive96.9%39,500$18,700
Fraud Signal ReviewerAG-04 · v2.0Fraud signal reviewMei TanakaDegraded93.1% / 956,900$4,700
Invoice MatcherAG-13 · v2.9Invoice matching and disputesHannah BeckerLive99.1%22,800$5,200
Email RouterAG-07 · v4.1Customer email routingOliver GrantLive98.6%61,300$3,800
27 agents in production 400,220 runs a week $124,400 a month across 8 models
A concrete outdoor staircase climbing a hillside in morning light

Policy, models and risk, next to the work.

Governance lives on the same record as the initiative. Model Risk sees what is about to reach production, Compliance sees which policy each agent falls under, and nobody keeps a second spreadsheet.

AI Acceptable Use Policy v3.242/42
Model Risk Management Standard v2.017/17
Human Oversight Standard v1.426/27
Privacy Impact Assessment for AI v2.127/28
Evaluation and Monitoring Standard v2.225/27

Value you can
show the board.

Every initiative carries a value case signed by Finance. Terrace tracks realized value against it each month, by business unit, net of what the agents cost to run.

$16.4Mrealized in FY26 to date
86%of plan to date ($19.0M)
$14.1Mnet of run and team costs
Realized vs projected by business unitFY26 · $M
Claims
6.0 / 10.9
Customer Operations
4.7 / 7.1
Underwriting
2.1 / 5.0
Finance
2.1 / 3.7
Risk & Compliance
0.9 / 2.5
IT & Security
0.6 / 2.3
Realized to dateProjected FY26
From gross to netFY26 to date · $M
Tea terraces on rolling hills in low afternoon light

“We had sixty pilots and no idea which ones mattered. A year on Terrace, seventeen initiatives are live and the CFO quotes our realized value in board meetings.”

Ingrid HalvorsenChief Data & AI Officer, Northmere Group
11 weeks
median pilot to production, down from 31
17
initiatives live in Production or Operations
0
findings in the 2026 model risk audit

Running in four weeks.

Terrace connects read-only to the systems you already use, so the first month is about your portfolio, not an integration project.

  1. Week 1

    Connect

    Single sign-on, Jira or ServiceNow, your model gateway and data catalog. Read-only to start.

  2. Week 2

    Map the portfolio

    Import the initiative spreadsheet. Terrace drafts stages, owners and value cases for each team to confirm.

  3. Week 3

    Set your gates

    Start from templates aligned to NIST AI RMF, ISO/IEC 42001 and the EU AI Act, then edit them with Risk.

  4. Week 4

    First gate review

    Run a real review in Terrace and send the board pack straight from it.

Connects to
  • Jira
  • ServiceNow
  • Slack
  • Microsoft Teams
  • Snowflake
  • Databricks
  • Okta
  • Entra ID
  • AWS Bedrock
  • Azure OpenAI
  • Datadog
  • Workday

Plans.

Annual contracts. Every plan starts with a 30-day pilot on your own portfolio.

Program

For a central AI team getting its first initiatives through gates.

$60,000a year

  • Up to 30 initiatives
  • Registry for 25 production agents
  • 3 gate templates
  • Value tracking by business unit
  • SSO with Okta or Entra ID
Book a walkthrough

Enterprise

For groups running AI across several business units and regions.

$180,000a year

  • Unlimited initiatives and agents
  • Custom gates per business unit
  • Finance sign-off on value cases
  • Kill switch through your model gateway
  • SCIM, audit export, 99.9% uptime SLA
Book a walkthrough

Sovereign

For regulated firms that need Terrace inside their own perimeter.

Custompriced by deployment

  • Private cloud or on-premises
  • EU or US data residency
  • Model risk workflows for SR 11-7
  • Customer-managed encryption keys
  • Named success team
Talk to sales

Questions.

A stone staircase rising toward a skylight inside a concrete building
Is Terrace a model or an agent builder?

No. Terrace runs the program around the models and agents you build or buy. It holds the portfolio, the gates, the registry, the policies and the value record, and connects to the runtimes where agents actually run.

How does the kill switch work?

Terrace calls your model gateway or agent runtime to revoke the agent's credentials and routes its queue to the fallback you set, usually a human team. It asks for confirmation, logs who pulled it and why, and notifies the owner.

Can we use our own gates and checklists?

Yes. Templates are a starting point. Most customers keep three gates and edit the checks per business unit, for example a stricter production gate for anything customer-facing.

How is realized value measured?

Each value case names a baseline, a metric and a method, signed by Finance at the production gate. Terrace pulls the metric monthly from your warehouse and shows realized value next to the projection.

Where is our data stored?

In the EU (Frankfurt) or the US (Virginia), your choice. Sovereign customers run Terrace in their own cloud account or data centre. Terrace never sends your data to a model provider.

Who uses it day to day?

The AI program office, Model Risk, Compliance, Finance and the business owners of each initiative. Executives mostly read the portfolio and the value page.

Bring your initiative spreadsheet. We will show it back to you as a staircase in the first call.

45 minutes with a product lead. Or open the demo workspace.

Terrace takes enterprise AI from strategy to production.