Service lines

Five ways we
put models to work.

Each engagement starts with the smallest system that could possibly prove the business case — then earns the right to grow. Pick a line below to see what it includes, how long it takes and what you actually receive at the end.

01 / Strategy

Applied AI Strategy

A short, intense diagnosis of where AI actually pays off inside your business — and where it will quietly waste a year of budget.

Timeline
2–3 weeks
Team
2 specialists
Starting at
$14,000

We interview the people who touch the process, audit the data you already collect, and score candidate use cases on value, feasibility and blast radius. You leave with a ranked portfolio and a build plan — not a slide deck that restates your own goals.

What's included

  • Stakeholder and workflow interviews
  • Data readiness audit against six dimensions
  • Use-case backlog scored for value and risk
  • Build/buy/fine-tune recommendation per case
  • Cost model with per-request unit economics

You receive

  • Opportunity map with decision rationale
  • 90-day build roadmap
  • Risk register and governance notes
  • Executive readout session

Best fit: leadership teams who have run one or two AI proofs of concept and need to decide what to commit to next.

02 / Knowledge

Knowledge & Retrieval

Turn scattered documents, tickets and wikis into an answerable corpus your teams and your models can trust.

Timeline
4–8 weeks
Team
3 engineers
Starting at
$42,000

Retrieval is where most AI products quietly fail — the demo answers beautifully, then production serves a stale PDF from 2019. We build ingestion, chunking, permission-aware indexing and an evaluation harness so quality is measurable rather than anecdotal.

Capabilities

  • Connector + sync pipeline for your source systems
  • Permission-aware index and access enforcement
  • Hybrid search with reranking and citation tracing
  • Golden-set evaluation suite with regression gates
  • Answer UI or API integration

Technical detail

  • Vector store provisioned in your cloud, not ours
  • Incremental re-index with change detection
  • Instrumented for latency and cost per query
  • Handover runbook and indexing docs

Best fit: support, legal, compliance or ops teams sitting on thousands of documents nobody can find quickly.

03 / Automation

Agentic Workflows

Multi-step automations that read, decide and act across your real systems — with human checkpoints where they matter.

Timeline
5–10 weeks
Team
2–4 engineers
Starting at
$55,000

Triaging inbound requests, reconciling records, drafting responses, escalating edge cases. We build these as explicit state machines with typed tool calls, retries and a full audit trail — so when something goes wrong you can see the exact step that drifted.

What we build

  • Workflow graph with typed tool boundaries
  • Human-in-the-loop review queues
  • Idempotent side effects and safe retries
  • Full run history, replay and diffing
  • Guardrails, rate limits and kill switches

Engineering notes

  • Failures are surfaced, never swallowed
  • Every action carries provenance and an actor
  • Runs in your infrastructure or ours
  • Alerting wired to your existing on-call

Best fit: operations teams processing high volumes of structured-but-messy work by hand.

04 / Decisions

Decision Models

Forecasting, scoring and ranking systems where the math is inspectable and the model's confidence is honest about itself.

Timeline
6–12 weeks
Team
2 data scientists + 1 engineer
Starting at
$68,000

Not every decision needs a large language model — and plenty of them need more than a spreadsheet. We pick the simplest method that beats your current baseline, validate it honestly on held-out data, and ship it with monitoring so decay is caught early.

Deliverables

  • Baseline benchmark and success criteria
  • Trained model with reproducible training pipeline
  • Feature store and scoring API
  • Drift, calibration and fairness monitoring
  • Explainability layer for reviewers

Discipline

  • No metric reported without a held-out split
  • Confidence intervals alongside point estimates
  • Documented failure modes and manual fallbacks
  • Regular retraining schedule agreed upfront

Best fit: pricing, risk, churn, demand or prioritisation decisions currently made on gut feel and spreadsheet tabs.

05 / Operations

Managed AI Operations

Ongoing ownership of the systems we built — or the ones you already have and can't quite keep healthy.

Timeline
Monthly retainer
Team
Shared pod, named lead
Starting at
$9,500 / mo

Models drift, vendors deprecate endpoints, token bills creep. This is the boring maintenance layer that keeps a shipped AI system dependable: evaluations on a schedule, cost reviews, incident response and a standing roadmap for improvements.

In the retainer

  • Scheduled evaluation runs and quality reports
  • Prompt, model and dependency upgrades
  • Cost and latency tuning each quarter
  • Incident response with defined response times
  • Monthly review call with your named engineer

Coverage tiers

  • Core — one system, business-hours support
  • Extended — multiple systems, 4h response
  • Embedded — fractional AI team, on-call participation

Best fit: teams running AI in production without a dedicated ML engineer to keep it alive.

Compare at a glance

Same discipline on every engagement — the difference is what we're producing and how long we stay.

Comparison of Aether service lines by timeline, typical starting price and primary output
Service line Typical timeline Starting price Primary output
Applied AI Strategy 2–3 weeks $14,000 Ranked opportunity map and build roadmap
Knowledge & Retrieval 4–8 weeks $42,000 Permission-aware index and answer interface
Agentic Workflows 5–10 weeks $55,000 Production automation with audit trail
Decision Models 6–12 weeks $68,000 Trained, monitored scoring system
Managed AI Operations Monthly $9,500 / mo Ongoing maintenance, evaluation and upgrades

Ranges reflect typical mid-market engagements. We quote fixed scope after the strategy phase.

Commercial terms, plainly

No open-ended hourly billing, no surprise change orders in month two.

Fixed scope, fixed price

Every build is quoted as a defined deliverable with a stated done condition. If we underestimated, that is our problem, not a change order.

You own everything

Code, models, prompts, infrastructure and documentation. Delivered in your repositories, running in your cloud accounts.

Weekly, not quarterly

You see working software every week. Nothing is hidden behind a "reveal" at the end of a phase.

Two-week exit

Build engagements can be stopped at any phase boundary with two weeks' notice. You keep everything produced to that point.

Questions we get asked

Can you work with our existing stack and vendors?

Yes. Most engagements run against infrastructure you already own — your cloud account, your model provider, your data warehouse. We add the missing layer rather than replacing what works.

Do you use our data to train models?

No. Your data stays in your systems. We do not train third-party models on client data and we configure vendors with retention disabled wherever the option exists.

What if we only have a rough idea?

That is the normal starting point. The strategy engagement exists precisely to turn a rough idea into a scoped, costed build plan before anyone commits to the larger number.

How do you handle security and access?

Least-privilege credentials, secrets kept server-side in your secret manager, and access reviews at each phase. We will complete your vendor security review before work begins.

Can you take over an abandoned AI project?

Often, yes. We start with a short audit — usually a week — to establish what is salvageable, then recommend either a repair path or a rebuild with a clear estimate of both.

Not sure which line fits?

Tell us what's breaking and we'll tell you the shortest path to fixing it — even if that path doesn't involve hiring us.

See the work