Placement Assessment

Where should your AI run? Find out in two to four weeks.

On-prem, colocation, managed cloud or a mix. We take two or three of your real use cases, test them against your own data, and give you a recommendation with the numbers, the risks and an exit path.

Book a discovery call See Technical Financial Models
Drawing: the placement decision A workload card describing a use case, its data class and its volume feeds three paths: local, on-prem or colocation; cloud, on a managed AI platform; and hybrid, split by data class. Each path carries a three-to-five-year cost line. All three converge on a recommendation box with its assumptions, risks and exit path, and a signature line. Summit Logic Placement decision Dwg SL-0014 Sheet 1 of 1 Rev A Illustrative 1 Workload Workload card Use case Data class Volume and pattern 2 Paths Local On-prem or colocation Cloud Managed AI platform Hybrid Split by data class Three-to-five-year TCO on each path 3 Recommendation One placement, stated plainly Assumptions, risks and an exit path Signed
the problem

Every vendor recommends its own answer

A GPU vendor recommends GPUs. A cloud provider recommends its cloud. A colocation provider recommends a cage. Each answer can be right for some workloads and expensive for others, and the decision usually gets made before anyone has run the workload on either side.

what you get

Seven things, written down

Built from your use cases and your data, not from a vendor benchmark.

Workload cards

Two or three use cases described the same way: what they do, how often they run, the data they touch and what good output looks like.

A classified data-flow map

Where each data class comes from, where it goes, and where it is allowed to be processed.

A benchmark on your test set

A local model and a cloud model run against your own examples, scored on quality, speed and cost per successful result.

A threat and control review

The identity, logging, encryption and approval controls each option needs, and who owns them.

Local, cloud and hybrid designs

A reference design for each option, sized to the workloads on the cards.

A three-to-five-year TCO

Each option costed on productive utilization rather than nameplate capacity, including the people to run it.

A recommendation

One placement, with the assumptions behind it, the risks that would change it and an exit path if they do.

how we stay neutral

Independent of what you buy

An assessment is only useful if the answer was not decided in advance.

Fixed fee

Scoped on a 45-minute call and fixed before the work starts. The fee does not change with the answer.

Published method

You see how each option is costed and scored, and every assumption is written down where you can challenge it.

Independent of the outcome

Our fee is the same whatever you buy, whoever you buy it from and whoever finances it. Some assessments recommend the cloud. That is the point of doing one.

The five questions we ask first

They decide most placements before a benchmark runs.

  1. 01

    What does the workload do?

    The task, the volume, and how often it runs: steady every day, or in bursts.

  2. 02

    What data does it touch?

    The data classes involved, and where each one is allowed to be stored and processed.

  3. 03

    How good does the output need to be?

    What a successful result looks like, and what a wrong one costs.

  4. 04

    Who will run it?

    The team that operates it day to day, and what they already know how to run.

  5. 05

    Over what horizon?

    How long the decision has to hold, and what would make you revisit it.

The decision stays yours

The assessment gives you the evidence and a recommendation. Which option you buy, and from whom, is your decision, and nothing we recommend depends on it.

Scope a Placement Assessment

Fixed fee, scoped on a 45-minute call. Bring one workload you are deciding where to run.

Book a discovery call