How we engage

Discovery

We start with discovery because most of the value of an engagement comes from picking the right operational job to work on. The wrong job is worse than no job. Discovery sits in front of every build, and if you decide to proceed, the discovery fee comes off the build.

Two tiers depending on how much groundwork is already done.

Sense Check - from £2,800 + VAT

Half-day session. "Is there a thing here worth doing?"

For teams who've been talking about AI but haven't worked out where it fits. We spend half a day with the people who'd actually use it, look at the work that's eating their week, and tell you honestly whether there's a Loop here or whether your problem is somewhere else entirely.

What you walk away with: a one-page write-up of what we saw and what we'd recommend. Including "don't" if that's the honest answer.

Sense Map - from £6,400 + VAT

One-week strategic mapping. "We know there's something here; what should we actually build?"

For teams who already know the operational job is real and want a structured discovery before committing to a build. We sit in with the people doing the work today, walk through the data, sketch the review surface, draft the DPIA + trustee briefing in outline, and define the eval criteria the system would need to meet.

What you walk away with: a scope document we can both sign, a draft DPIA, a draft trustee briefing, and a build estimate.

Build

Build engagements vary widely. We don't put a fixed-fee price on the page because pricing a £7k focused agentic flow and a £110k multi-month product the same way would be dishonest in both directions.

Three reference points from real recent projects:

• A focused agentic flow for an existing client - around £7,000. Single Loop, integrating with one existing system, light governance footprint, short timeline.
• The initial AIDA build at Breast Cancer Now - around £28,000. Production system, multi-stage extraction across 20+ NHS form structures, governance review, eval calibrated against research-team gold-standard. Has grown since with the work.
• Goose for the National Lottery Heritage Fund - around £110,000 over twelve months. Multi-Loop product, twenty-one specialist personas, vector search, multi-model architecture, beta-to-production rollout across 40 organisations.

Most engagements sit somewhere between. What drives the number:

• Integration complexity. Talking to one existing system is cheap. Talking to a CRM + a finance system + a MEAL stack + a survey platform takes longer.
• Governance bar. A fundraising Loop needs a competent DPIA. A safeguarding Loop needs a competent DPIA + trustee-board briefing + a conservative eval threshold + a quarterly review schedule. The high-stakes work takes longer because it should.
• Scope. One Loop is one Loop. Three Loops sharing the same data is three Loops plus the orchestration. A multi-Loop product (Goose-shaped) is a bigger commitment.
• Pace. Same scope on a tighter timeline costs more because we can't sequence the work the same way.

If you've done a Sense Check or a Sense Map with us, we'll give you a realistic number for your specific build before you commit to anything.

Ongoing care

Care is ongoing, monthly, transparent, cancellable. We keep the system running and honest about its performance - monthly reports of what the eval numbers are doing, what the reviewer's edit rate is, where the model is finding new edge cases. We retune prompts against your team's correction patterns. We update the system when a model upgrade is available and worth taking.

Care doesn't have a fixed price either, mainly because two things drive cost and both vary by orders of magnitude:

• API token usage. A Loop that processes ten cases a week costs differently from one that processes ten thousand. We pass model costs through transparently rather than marking them up.
• System footprint. A single-Loop deployment with a stable integration footprint is light-touch. A multi-Loop product with a quarterly governance review and an evolving feature set is meaningful ongoing work.

We'd rather quote care honestly once we know what your deployment looks like than promise a number we'd need to walk back.

What's yours

Ownership is structural, not negotiated.

• The code is yours. The Next.js app, the Loop definitions, the schemas, the configuration - yours to take in full. If we don't work out, another agency can pick it up.
• The deployment is dedicated to you. A Firebase and Vercel deployment we run for you alone, isolated from every other charity, with UK data residency for stored data. Not a shared, multi-tenant tool.
• The data is yours. Test data is synthetic; production data stays in your isolated deployment, and you get it back in full whenever you ask.
• The audit trail is yours. Every decision the system made, every change your reviewer accepted or rejected. Queryable, exportable, and yours to keep regardless of whether you keep working with us.

What we'll tell you, even if you don't want to hear it

• If we don't think a Loop is the right fit for your charity, we'll tell you. Not every operational job benefits from this; the wrong job is worse than no job.
• If the synthetic eval numbers on a pattern don't translate to your real data, we'll show you the gap honestly. The framework transfers; the calibration is always per-charity.
• If a chat tool would solve what you're asking for, we'll say so. Plenty of charity AI work is well-served by a ChatGPT subscription; this isn't that.
• If we're the wrong agency for your scale, sector, or governance posture, we'll point you somewhere better.