How we work

Context

A business opening its first United States site against a fixed trading date. Technical ownership sat with an existing team in another region, with no overlap for most of the working day, and nobody in market was accountable for the order things happened in.

What was actually wrong

The plan looked healthy. Nothing was overdue, and the schedule was green. Reading the sequence rather than the dates showed two dependencies scheduled backwards: an insurance controls pack due after the policy was to be bound, and content testing due after the site it tested was to go live. Separately, the internet circuit had not been ordered, and quoted installation lead times across candidate addresses ranged from two weeks to six months. At the longer end, the service would have gone live months after the site was meant to trade.

What we did

1. 1

   Compared every carrier serving every candidate address, on lead time first and price second, because the schedule risk dwarfed the cost difference

2. 2

   Established which providers absorbed construction cost and which recovered it through the contract term, since the quoted monthly figure is a floor rather than a price where it is recovered.

3. 3

   Recommended ordering on lease execution day rather than sequencing with the fit-out, with fixed wireless as a parallel order to protect the date

4. 4

   Rebuilt the dependency register to separate who routes a question from who can answer it

5. 5

   Ran a reversed-dependency pass across the whole programme plan

6. 6

   Specified the support model as delegated actions rather than delegated services, so it could actually be priced and committed to

What changed

The connectivity decision moved from a procurement line to the top of the critical path, where it belonged. Both reversed dependencies were corrected before either became a problem. The support model was agreed with a stated boundary short enough that a new person could read it and know who did what.

A six-figure sequencing error caught before it became a delayed opening.

What this involved

• Dependency register
• Carrier comparison
• Delegated administration

Context

A group operating several legal entities in more than one currency produced its management reporting by hand each month, pulling from an accounting platform, a project system and a time-tracking tool, then reconciling in a spreadsheet.

What was actually wrong

Not all of the historical data was reachable. Some records sat on boards that had been deleted or made inaccessible, which meant any honest pipeline would be incomplete. The commercial structure also did not match the accounting structure: some clients belonging to one part of the group were invoiced from another part's books, so consolidated revenue did not reconcile to the sum of its parts. And the finance team maintained its own annotations in a spreadsheet with no home in any system.

What we did

1. 1

   Connected each source into a warehouse, with the application reading only from defined metric views rather than raw tables

2. 2

   Accepted and stated the real coverage figure rather than quietly excluding what was missing, and documented exactly what the gap consisted of

3. 3

   Built an explicit mapping table for the commercial-to-accounting relationship, confirmed by finance, rather than inferring it

4. 4

   Joined the finance team's existing annotation spreadsheet to the warehouse rather than requiring them to abandon it

5. 5

   Treated entity as a configurable dimension rather than hard-coding the group structure, since groups merge and split

6. 6

   Deferred the AI layer entirely until every source was integrated, because analysis over partial data produces confident answers about a partial picture

What changed

Reporting that previously took days of assembly became current and traceable, with data coverage shown as a property of each figure rather than an unstated assumption. Where a figure could not be stood behind, it was shown as unavailable rather than estimated.

The monthly assembly stopped being a person's job.

What this involved

• Cloud warehouse
• Scheduled pipelines
• Versioned metric definitions

Context

An organisation whose Windows sign-in ran through a credential provider that did its job but offered no way to reason about device state, location or risk before granting access.

What was actually wrong

The obvious case was cost, and the licensing arithmetic did favour the move. But a migration framed only on cost tends to complete as a lift and shift that saves money and changes nothing about security. The existing implementation plan had sound phasing and licensing, and was missing the two things these plans almost always miss: the federated sign-on configuration back to the productivity platform, and detailed conditional access policy specifications rather than a stated intention to implement them.

What we did

1. 1

   Designed the conditional access policies in full, with group scoping and failure behaviour, before touching anything

2. 2

   Configured and tested federated sign-on with a small group while the old path still existed

3. 3

   Established the specific administrative roles across both vendors as a prerequisite to starting, not as a task within it

4. 4

   Completed unenrolment of the old path rather than running both indefinitely, so the new one had to be complete

5. 5

   Enrolled devices in phases with compliance policies applied by group

6. 6

   Wrote removal of elevated access into the project close criteria at the same time as granting it

What changed

Access became conditional on device compliance rather than on credentials alone, and the licensing structure reduced the monthly cost against the previous arrangement. The elevated roles granted to execute the work were removed at close rather than quietly persisting.

A materially better access posture, at a lower monthly cost than what it replaced.

What this involved

• Endpoint management
• Conditional access
• Federated sign-on