A 350-server AWS migration, priced at three levels of detail

Business description

The customer already runs its production systems on AWS. Its development and engineering estate still runs on very old hardware, on operating systems and software past vendor support and without current licences. Every patch cycle costs engineering hours the team would rather spend on product, and each year on that hardware adds risk the leadership team can see but could not yet price.

The question for the leadership team was practical: what does it cost to bring that estate to AWS, which servers should move, and how much detail do we need before we commit? The customer did not connect its production AWS accounts to Sherpa. It shared the facts about that environment, and Sherpa’s agent relied on what it was told.

The challenge

• The facts lived in pieces. Server exports from two virtualisation environments, a discovery tool’s usage data, nine application questionnaires, recorded planning sessions and planning forms each held part of the picture. No two sources agreed on the server count.
• A server list is not a plan. Knowing what runs on a box says nothing about what it is for, who owns it, what depends on it, or whether the business wants it in the cloud at all. Priced without that context, a migration estimate carries every unwanted server for three years.
• Automated sizing needs a referee. AWS Transform, AWS’s migration assessment service, sizes and prices servers quickly, but it works only from the files it is given. It does not know which servers the customer has already decided to retire or keep.
• A one-person cloud team. The engineering lead carried cost, security and infrastructure decisions alone. There was no Cloud Center of Excellence to gather requirements, hold the context and guide the decision.

Sherpa as the Customer’s AI Cloud Center of Excellence

Sherpa is an AI platform for AWS. A Cloud Center of Excellence captures the business context (what each system is for, who owns it, what the business expects from the move), gathers the telemetry (what the servers really do), and turns both into decisions the business can act on. Sherpa’s Migration Assessment does that work as one guided conversation across seven sections, and it turns context and telemetry together into agentic, holistic answers the business can understand.

Context is gathered deliberately. Sherpa worked through nine application questionnaires and more than 130 questions across the seven sections: business goals and sponsors, what each application is for and who owns it, its tier, its dependencies, its recovery needs, and what should stay, retire or move. It read each document, repeated back what it understood, recorded only what the customer confirmed, and asked when a document was silent. Nothing was assumed, and what was still unknown was shown as unknown, never as zero.

Section	What it determines	What it produced for this customer
Introduction & Kickoff	Context: sponsors, goals, the first data request	Goals and sponsors recorded; 2 data requests opened; the customer’s existing AWS support plan noted as a fact
Business Scope & Owners	Context: why now, what is in scope, who owns each system	Ranked business outcomes, change freezes, funding and the owner of the on-premises cost, recorded as decisions and open items
Estate Data	Telemetry: the server inventory	Four exports combined into 429 servers, 216 with measured usage; duplicates across environments listed; scope proposed from the documents and confirmed; readiness checked before any automated sizing
Workload & Domain Discovery	Context joined to telemetry: applications, owners, dependencies	19 documents read; 12 applications with owners and tiers; 8 dependencies and 5 file shares; servers matched to applications only where the evidence supported it
Infrastructure	Decisions: what stays, retires or moves	From three recorded sessions: 67 servers stay, 4 retire, 28 more confirmed to move; the scope settled at 350; the customer’s right-sizing holds recorded
Target State & Pricing	Agentic output: sizing, pricing, the business case	Term, AWS support plan and storage growth recorded as fields; AWS Transform’s results checked against the scope; two prices side by side; the customer’s choice recorded
Review & Compile	Agentic output: the plan	12 waves carrying 41 confirmed servers with environment, cutover window and rollback; 17 risks with owners; 429 = 350 moving + 79 out of scope

Three levels of pricing, one estate

The customer was offered the price at three levels of detail, and could have stopped at any of them. The same servers were run through AWS Transform each time, with more context added at each level.

	Level 1: AWS Transform on the server exports	Level 2: AWS Transform with measured usage	Level 3: AWS Transform with Sherpa
What went in	The virtualisation exports	Plus the discovery tool’s usage data	Plus nine application questionnaires, the recorded sessions, the planning forms and the business goals
Servers AWS Transform priced	419	412	343, from a scope of 350 the business agreed
AWS Transform’s own estimate, a year (compute, licences and network)	$214,351 on a three-year Compute Savings Plan	$232,262, 8% higher: the servers were busier than assumed	$163,240 on a three-year Reserved Instance
What else is priced	Nothing. Databases, block and file storage and platform services are outside the figure	Same	Sherpa’s full priced lines: compute on the basis the customer chose, databases, block and file storage, platform and network: $210,329 a year, $630,987 over three years
What the business learned	A compute number for a server list. Its first plan dropped the powered-off servers and counted shared-name servers twice until Sherpa corrected it	The sizes improved; the questions did not. Still no owners, no scope decisions, and nine same-name server pairs unresolved	Which 79 servers to keep or retire, who owns what moves, the sizes the customer chose to hold, a wave plan with servers in it, the risks with owners, and a price it can trace line by line

The more context Sherpa had, the better its answer. With the business goals and application context beside the telemetry, it cut the scope from 425 servers to the 350 the business meant to move, honoured the customer’s own right-sizing decisions, priced the databases, storage and platform services that a server-only estimate leaves out, and recorded the support plan, term and growth the customer chose. Level 3’s figure leaves out the AWS support plan, which the customer already pays for on its production accounts.

Pricing basis

AWS Transform sized the servers and priced them on its own standards: a three-year Reserved Instance with no upfront payment, licence included where the operating system needs it. Sherpa checked that plan against the agreed scope, priced the same servers on its own basis and showed both figures side by side; the customer chose AWS Transform’s basis, with network priced once on Sherpa’s own line.

Scope guarded at every step. AWS Transform’s first plan kept four servers the customer had excluded, mis-added its own counts and left one server out without a reason. Sherpa caught each difference before anything was approved, drafted the correction, and asked the customer before sending it.

Results and business impact

• The three-year figure. On AWS Transform’s basis, the priced lines come to $210,329 a year, $630,987 over three years, for 350 servers in scope, of which AWS Transform matched and priced 322. The figure leaves out the AWS support plan, disaster recovery, backups, two databases awaiting their details, five desktops and the five servers that were not sized. It is a cost estimate, not a saving: no on-premises cost was supplied.
• Every figure is traceable. We checked every number in the assessment’s outputs (264 of them) against the pricing workbook or a recorded answer: all 264 traced, none was invented. Every fact Sherpa said it had saved was in the record.
• Measured accuracy. We score each assessment on four things: whether what Sherpa says it saved was saved, whether it covered the assessment’s questions, whether every number traces to a source, and whether it ever guessed. This engagement scored 99.5% with exports and usage data, and 98.3% with every input, against our 98% bar.
• A plan with servers in it. The wave plan carries 41 confirmed servers across 12 waves, each with its application, environment, cutover window and rollback, and 17 risks with named owners.

For the leadership team, the result is a migration budget they can defend line by line, a scope the business agreed, and the open questions listed instead of hidden inside a total. The development estate that was costing engineering hours now has a priced path to AWS, and the customer has a Cloud Center of Excellence that already holds its context for the move itself.