Don't book this if
- 01 You haven't scoped the use cases yet: do a Discovery Sprint.
- 02 You want a vendor demo rather than a model: that's someone else.
- 03 You want a single point estimate with two-decimal confidence: ranges are the honest shape.
ServiceModel
The business case, before the purchase order.
A full cost, ROI, and option-comparison model for 3–5 shortlisted use cases. Cloud, self-hosted, hybrid, and platform-native priced side-by-side, as ranges rather than single-point forecasts, with environmental impact modelled.
Terms Fixed fee per shortlisted use case. Bundle pricing applies beyond three. Confirmed at scope.
01 · Fit
Who it's for. Who it isn't.
Book this when
- 01 Your AI programme has to survive a CFO review.
- 02 You're choosing between SaaS, a fine-tune, self-hosting, or sticking on a cloud API.
- 03 You need Scope 2–aligned numbers, not vendor averages.
02 · Deliverable
What lands on your desk.
-
01
The Model. A live spreadsheet / report pairing each shortlisted use case with TCO, ROI range, break-even curve, and option comparison.
-
02
Option comparison. Cloud API, self-hosted, hybrid, and platform-native priced side-by-side with documented assumptions.
-
03
Risk register. Vendor lock-in, data residency, regulatory exposure, annotated per option.
-
04
Environmental impact. kWh and CO₂e per workflow per option, using your grid factor (or a cited default).
-
05
Board-ready summary. One page with the recommendation a finance team can defend.
03 · Phases
How it runs.
Phase 01
Intake
Agree the 3–5 use cases in scope, the options to price, and where the numbers live.
Phase 02
Assumption review
Volume estimates, rate cards, depreciation schedules, residency constraints. Surface what is guessed vs measured.
Phase 03
Model build
TCO and ROI per use case per option. Ranges, not points. Sensitivity tables for the inputs that move the answer.
Phase 04
Sensitivity + risk
Which assumptions break the case if they're wrong? Which risks are absorbable? Annotate the model.
Phase 05
Readout
A session with the person who has to approve the spend. Written report + live model handed over.
04 · Sample
What the output looks like.
Anonymised preview of a Cost & ROI Model deliverable. Three deployment options priced side by side, with assumption ranges and a defensible recommendation.
ALT HORIZON · COST + ROI MODEL
Three options · one workload
Option£ / month£ / yearkg CO₂e / moVerdict
Cloud · frontier API
Claude Sonnet 3.5 · monthly £3,520 £42,240 14.2 Capable · costly
Claude Sonnet 3.5 · monthly £3,520 £42,240 14.2 Capable · costly
★ Self-hosted · 14B
RTX 5070 Ti · UK-hosted £410 £4,920 2.8 What we'd run
RTX 5070 Ti · UK-hosted £410 £4,920 2.8 What we'd run
Hybrid · 60 / 40
Self-hosted 8B + cloud Flash £980 £11,760 6.1 Fall-back · burst capacity
Self-hosted 8B + cloud Flash £980 £11,760 6.1 Fall-back · burst capacity
Assumption ranges
Volume 480–840 cases/wk · UK industrial energy at 12–18p/kWh · GPU depreciation 36 months · 0.6 FTE operator allocation · model refresh quarterly.
Summary
"A model with ranges and assumption-testing survives finance review. A single-point forecast doesn't. This is the difference between "we think it'll pay back in 18 months" and "here is the defensible envelope.""
05 · Alternatives
Not the right fit? Try one of these.
06 · Next step
Bring a workload. Get the numbers back.
A real workload, two sentences. We price it across cloud, self-hosted, and hybrid, then tell you what we'd run — or tell you to wait.
No deckNo sales pressureWe'll tell you to wait if you should
Or email a workload to hello@althorizon.co.uk — one-page model back in 48h.
