Skip to main content
TechGuidr

Fixed-price engagement

Your AI might work. Do the economics stack up?

AI is one of the fastest-growing cost lines in business, and much of it is spent paying frontier-model prices for tasks a far cheaper model would do reliably. We review what your AI costs, what it produces, and what to do about the gap.

Use the cheapest model that reliably does the job. Escalate only when the task earns it.

Spend less where intelligence is commodity. Spend more where intelligence matters. That includes the consulting: if a £100k implementation is solving a £20k problem, we will say so.

What we assess

Spend

Model and API spend, licences, infrastructure, vector and search costs, agent and tooling costs, external consulting and duplicated subscriptions.

Token economics

Input and output volumes, prompt size, context-window usage, repeated context, cacheability, embedding volume, unnecessary agent turns and output verbosity.

Model choice

Where frontier models are earning their price and where smaller, specialist or open-weight models would reliably do the job. Routing, fallbacks and quality thresholds.

Architecture economics

Caching, batching, asynchronous processing, retrieval strategy, context trimming, deterministic code versus model reasoning, and hosted versus self-hosted trade-offs.

Business value

Per workload: what it costs, what it produces, cost per task, value per task, payback and the confidence behind the evidence. Savings and revenue, not just savings.

Every workload ends up in one of four columns: scale, optimise, redesign or stop.

What you actually get

Illustrative example, not client data. The real thing carries your spend, your workloads and your evidence.

TechGuidr

Sample extract

AI cost & value review: summary

Current monthly AI spend
£31,200
Frontier-model traffic
84% of requests, most of it routine extraction and classification
Cache hit rate
Effectively zero; identical questions regenerated in full every time
Routable to smaller models
Estimated 60% of traffic at equal reliability
Workload verdicts
One to scale, two to optimise, one to redesign, one to stop
Target architecture
£11,800/month for the same delivered value

Illustrative example, not client data. The real document carries your evidence and your numbers.

£31,200−£9,400−£4,600−£2,900−£2,500£11,800Current spendper monthRoute toright-size modelsCachingand reuseContext andoutput trimmingOne workloadstoppedTargetarchitecture
The same review, drawn: where the £19,400 a month comes out. Illustrative figures; your numbers will differ, the shape rarely does.

Typical investment

Fixed scope. Fixed price. All figures ex VAT; the final fixed fee is confirmed after scoping.

Focused review
from £5,000
Larger or multi-system environment
typically £10,000–£20,000+

Our AI bills are getting stupid.

We hear that a lot. Two or three sentences on your estate is enough to start.

Tell us about your AI spend