Audit
We analyse your current AI spend: volume, distribution across providers and models, and usage patterns. Results are delivered within 72 hours of receiving your consumption data.
Audit pricing and termsHow it works
Four stages, from the first measurement to recurring savings. None requires a migration or planned downtime.
Your applications
Your AI gateway With your credentials
Your providers With your contracts
Determines which model serves each task and delivers the configuration to your gateway. Updated every month. No migration, no intermediary in your requests and no planned downtime.
Where we fit
Your AI gateway remains yours. We determine which model serves each task and deliver that configuration to you; your requests still travel from your applications to your providers, with your credentials and your contracts.
The 4 phases of
Savings proposal based on your actual usage
We quantify the achievable saving against your own consumption rather than generic estimates, together with its expected impact on response quality.
Applied in one afternoon
We deliver the configuration that assigns the right model to each task, and your team applies it in its own AI gateway, with its own credentials, in one afternoon. No change of provider, no migration and no planned downtime.
Continuous optimisation and monthly reporting
Savings are recalibrated continuously as your usage patterns and market prices evolve. Each month you receive a report comparing your invoice with and without e-ficient.
Three ways of working
You decide the level of involvement. We never operate inside your systems without an explicit request.
Audit with no access to your systems
We work from the usage records and invoices you provide. No access to your systems.
Configuration applied by your team
Our core service. We deliver the model assignment for each task, backed by quality measurements, and your team applies it in its own AI gateway. Every month we update it for new models and prices.
Delegated management in your infrastructure
If you prefer us to manage model assignment, we install an intermediate component (proxy) inside your infrastructure, with your credentials, which you can switch off at any time. We always start with non-critical tasks.
Why it works
Diagnosis before promise
We first measure your actual spend and its composition. Only then do we quantify the possible saving, using your data rather than industry averages.
No intervention in your systems
We deliver the configuration that assigns the right model to each task, and your team applies it in its own AI gateway, with its own credentials. Your requests never pass through us.
Verifiable savings every month
Each month you receive your invoice compared with and without e-ficient. If the saving is not in the report, it does not exist.
What about response quality
A saving that degrades quality is not a saving; it is a displaced cost. We measure quality before we start and monitor it continuously. If an optimisation would degrade it, we tell you and do not apply it.
What happens to your data
We operate in compliance with the GDPR. The diagnosis relies exclusively on aggregated consumption data. If you engage the service, you provide the sample of requests we use to measure quality, stripped of personal identifiers such as emails, phone numbers, ID numbers or IBANs, and covered by a data processing agreement. We never access your systems.
Audit turnaround
72h
From receipt of your consumption data. Includes achievable savings and their budget impact.
Applying the configuration
Continuous operationNo service interruption
A configuration change in your own AI gateway, with no migration and no planned downtime. Reviewed every month.
Who it is for
Companies already running AI in production and watching the invoice grow without clear control: finance leadership that needs predictability, technology leadership that does not want to slow delivery, and procurement that needs to justify the spend.
Use case examples
Two common scenarios and the measure that resolves each one.
Data extraction from documents
A company processes thousands of invoices and delivery notes every month with the same large model for every task: reading the document, classifying it and extracting four fields.
e-ficient measures the cost per task and finds that most of the spend is concentrated in the simplest, most repetitive one. We assign classification to a lighter model, reserve the large model for complex documents and move work that does not need an immediate response to the cheaper deferred-processing rate. The output is unchanged; the invoice is reduced.
Internal search over documentation
A team queries its internal documentation through an assistant. Over time, the instructions sent to the model have grown: guidance, examples and entire documents sent with every request, whether used or not.
e-ficient identifies which part of that information is resent without being used, ensures the fixed part is reused across requests and paid for only once, and limits each query to the relevant fragment instead of the full document. Cost no longer grows with the volume of information.
Find out how much of your AI spend is avoidable
The initial audit quantifies, within 72 hours, the achievable saving on your actual usage, along with its impact on budget and quality.
Request an auditFrequently asked questions
How much can my company save?
The usual range is 30–60% of spend on AI models; the exact figure depends on your consumption patterns, and the audit confirms it within 72 hours from your own data. If, with the configuration applied, the saving falls short of your monthly fee, you cancel with no lock-in and no penalty.
What does the audit involve?
An analysis of your current AI spend: volume, distribution across providers and models, and the sources of avoidable cost. Pricing and terms are set out on the Pricing page. Audit pricing and terms
Do you have access to my systems? Do I need to change provider?
No. We deliver the configuration and your team applies it in its own AI gateway, with its own credentials. You keep your contracts and service levels. We only operate a component of our own if you request it, and always inside your infrastructure.
What happens to response quality?
It is measured before we start and monitored continuously. If an optimisation would degrade it, we tell you and it is not applied.
How is the saving verified?
Each month you receive a report comparing your invoice with and without e-ficient. If the saving does not appear in the report, it does not exist.
Which companies is it for?
Companies running AI in production whose invoice is growing without clear control. Below a certain spend level the service does not pay for itself, and we will tell you so.