Skip to content

All practicesRun

Managed Services — Growth

Everything in Managed Services Essential, plus a team that keeps your AI systems getting cheaper and better: evaluation run every month, cost and performance tuning, moves to new model versions once tested, and a quarterly review of what to do next.

Book a Free SessionSee how it runs

Sound familiar?

It's live and it mostly works. But the model bill goes up every month, a new model came out and nobody's had time to check whether we should switch, and we only find out the answers got worse when a customer complains.

What you leave with

  • Monthly report

    Format: Document

    How we know it’s acceptableDelivered by the fifth working day; covers evaluation trend, cost per unit and changes made

  • Evaluation results

    Format: In your repo

    How we know it’s acceptableRe-runnable by you; trend tracked month to month

  • Change log

    Format: In your repo

    How we know it’s acceptableEvery change has its measured effect recorded

  • Model upgrade report

    Format: Document

    How we know it’s acceptableBefore and after evaluation scores; client sign-off before release

  • Quarterly review pack

    Format: Presentation

    How we know it’s acceptableReviewed with your owner; recommendations agreed or declined

  • Runbook updates

    Format: In your repo

    How we know it’s acceptableReflects the current system after every change

Also included

  • Everything in Essential: monitoring, alerting, incident response, monthly health report

How it runs

Monthly

Onboarding

  1. 2 to 4 weeks

    Onboarding

    Setup, alerting and monitoring system assessment, access, monitoring (through Essential), baseline evaluation and baseline cost.

    Out

    Onboarding report, agreed service levels and evaluation measures.

Then, on a repeating cadence

  1. Every month

    Week 1, evaluation run and cost review; weeks 2 and 3, tuning or model upgrade work; week 4, monthly report.

  2. Every quarter

    The quarterly review, with recommendations for the next quarter.

Is this right for you?

This is for you if

  • Buyerthe person who owns the AI system's results and running cost after launch: a CTO, head of product or head of platform.
  • Also has to say yesFinance, because it is a recurring cost; procurement, for the contract and service levels; security, because we keep ongoing access to production.

When this isn't the right fit

  • Your system is not live or stable yet. Finish the build first.
  • Your system is small and model spend is low. If the savings we can find are smaller than the fee, Essential is the honest recommendation.
  • You have a platform team that wants to own this. The Eval Harness Sprint lets them run it themselves.
  • You expect new features under the plan. New features are an AI Engineering Pod or a sprint, not managed services.
  • Another vendor built it, with no documentation and no evaluation set. An onboarding assessment comes first.

What’s included, and what isn’t

In scope

  • the listed systems
  • monitoring and incident response to the agreed service level
  • monthly evaluation
  • optimisation
  • model upgrades
  • monthly report
  • quarterly review

Out of scope

  • new features or workflows (a pod or a sprint)
  • fixing defects in another vendor's code beyond an agreed allowance
  • your infrastructure outside the AI system
  • model and hosting costs
  • 24/7 cover unless priced
  • building an evaluation set from scratch (onboarding)

Before we start

Before day one

  • The system is in production, with documentation or a handover from the build team.
  • Production access and deployment rights agreed.
  • An evaluation set exists, or the onboarding work to build one is agreed.
  • Your named owner.
  • Incident contacts.
  • The Essential runbook in place.

What we’ll need from your team

  • Your owner: two hours a month to review the report, and two hours a quarter for the review.
  • An engineer: one to two hours a week for deployment approvals.
  • A business stakeholder: to sign off whenever a model upgrade changes behaviour.

Pricing

What we need to give you a quote

  • A list of systems and their architecture.
  • Volume and model spend.
  • How access works.
  • Any existing evaluation sets.
  • Service-level expectations.
  • Incident history for the last three months.

After we hand over

Who owns it after we leave

This is an ongoing service, so ownership is shared. You still name an owner who reviews the reports and approves changes. On exit, we hand everything over under a written plan, so the system doesn't become nobody's job.

How success is measured

  • Cost per request or per task over time
  • Evaluation scores over time
  • Incidents and time to restore
  • Service-level compliance

What comes next

Questions buyers ask us

  • For uptime, yes. What usually doesn't get done is the evaluation and the model upgrades, because nobody schedules them.

  • Fair. If your spend is small and the system is stable, buy Essential. Growth pays off when the savings or the quality risk are bigger than the fee.

  • Evaluation and tuning run every month, and the monthly report shows exactly what was done.

  • Everything lives in your repository, and there's a notice period.

  • That's why we evaluate before we switch, and why upgrades are part of the service.

Start with a free session.

We’ll tell you if this is the right programme.

Book a Free Session