What you need, where to find it, and the costs almost everyone forgets.
Estimates are fine — the app grades how complete your figures are and adjusts how
confidently it answers.
The minimum
Two numbers get you a usable answer. Everything else sharpens it.
You need
Where it comes from
Your yearly IT running cost
Last year's accounts — see the checklist below
Required
A cloud cost, per year
A provider quote, or their pricing calculator
Required
Big spend already coming
Servers due for replacement in your horizon
Strongly advised
Cost to set up
Migration quote, training, parallel running
Optional
Cost to leave later
Data egress charges at your data volume
Optional
Expected gains
Your own judgement — be conservative
Optional
Leave anything you do not know blank. Blanks are counted and reported, never
quietly treated as zero.
Building your current running cost
This is the number people most often get wrong, almost always by
understating it — which makes staying put look cheaper than it is.
Count all of this
Electricity for the machines, and the cooling that removes their heat
Rent or opportunity cost of the space they occupy
Software, operating system and database licences
Maintenance and support contracts
The share of IT salaries spent keeping it running, not improving it
Backup storage, off-site copies, and testing that they restore
Insurance, security audits and compliance reviews
Your own time, if you are the person who fixes it at weekends
Commonly missed on the cloud side
The support plan — the free tier rarely suffices in production
Data transfer out, which is charged and grows with use
Backup and snapshot storage, billed separately from compute
Staff training, and the productivity dip while people learn
Running both systems in parallel during the move
Getting your data back out if you ever leave
If you only have monthly figures, multiply by twelve
before entering them. The app asks for yearly totals and multiplies by your horizon itself —
entering a monthly figure understates the result twelve-fold.
Being honest about gains
This is where business cases go wrong, and it is usually not deliberate.
Enter only gains you would defend if challenged by someone sceptical. Three tests worth
applying:
Extra sales. Would they actually happen, or is this the best case? Halve your
first instinct, then enter that. Reduce it to profit using your margin — revenue is not value.
Time saved. Only count hours that go somewhere productive. Two hours a week
returned to someone who then has two more hours of slack is not a gain.
Losses avoided. Multiply the cost if it happens by how likely it is in a year.
A £50,000 outage you would expect once a decade is £5,000 a year, not £50,000.
The app deliberately provides nowhere to enter agility, resilience, staff skills or
credibility with larger customers. Those are real and often decisive, but putting invented
numbers on them is how a business case stops being evidence and starts being advocacy.
Weigh them yourself, alongside the figure.
Reading the answer
Look at these first
The confidence grade, above the result. If it says the inputs are too thin,
the headline figure means nothing yet.
The downside test. Whether the case survives costs running 20% over and gains
landing 30% short matters more than the headline, because it depends less on your
estimates being right.
The fit gauge. If the numbers and your business profile disagree, the profile
is usually the better guide — it does not depend on estimates.
Then decide
If the shortcut check says the quick sum reached the same answer, trust the quick sum.
If it says they diverged, verify the single largest item it names — the whole
conclusion turns on it.
Print the business case. It carries its own assumptions and limits, so a lender or
board member who was not present can still interrogate it.
Try it without your own figures
The demo loads a deliberately ordinary case — a twenty-person firm with an ageing server,
a moderate cloud quote and modest gains. It is not engineered to produce a flattering result,
because a demo that always says yes would teach you nothing about how the tool behaves on
your own marginal case.