AI consulting company in London
We build AI that moves a number you choose.
Dilr.ai runs AI consulting from London, for a two-person office or a regulated bank. We find where AI actually pays, then build and operate it, including AI voice agents that answer your calls. We agree the measure first, and you pay once it moves.
No commitment, no sales deck.
- London, UKCompanies House 16842656
- Regulated sectorshealthcare and financial services
- On-prem and air-gappeddeployments supported
- Four production productsnot prototypes
Open weightsMira-Q2 on Hugging Face, Apache-2.0- Per-country call compliance8 countries
Which DILR product do I need?
Four AI products, each with a published starting price
Pick by bottleneck: calls, shared work, private clinical data or teaching. Three are self-serve. Academy is licensed to institutions and scoped on a call.
From $0.14 / minDilr Voice
Best for AI receptionists and phone calls
Answers and places calls 24/7. Multi-agent handoff, RAG knowledge bases, outbound campaigns, per-country compliance.
- Call comes in
- AI agent answers
- Follow-up runs
Free, then $5 / userCognibl
Best for managing work done by people and AI agents
Work management where people and AI agents share one board. A task cannot reach done without attached proof.
- People and agents
- One shared board
- Proof to close
- Open weights, Apache-2.0
Dilr Mira
Best for private clinical document extraction
Private clinical models, about 3B parameters, that read documents into schema-valid JSON on your own hardware.
- Clinical document
- Your hardware
- Schema-valid JSON
- Licensed per institution
Dilr Academy
Best for universities, colleges and training providers
Builds interactive multilingual courses on demand and teaches them Socratically, with mastery tracking. Licensed to universities, colleges and training providers.
- Asks before it tells
- Interactive lesson
- Mastery tracked
How do you manage work done by AI agents?
Cognibl puts people and AI agents on one board, and nothing closes without proof
A task cannot reach a done status without an attached proof version. The database refuses the move, for a person and for an agent alike.
- Person or agent picks up the task
- Works it on the shared board
- Attaches a proof version: run CSV, artefacts, screenshots, hashes
- Done
- No proof: the move to done is refused
- Proof gateNo task reaches done without an attached proof version.
- Process templatesHarness, Graph, Loop or Custom.
- Skills library over MCPSKILL.md, immutably versioned, deny-by-default gateway.
- Metrics that matterCycle time split by spec, build, verify and settle.
- A real trackerBacklog, sprints, roadmap, P1 to P5, Fibonacci estimates.
- PricingThe free tier is the whole product, for up to 10 members.
Cognibl pricing
- Free$0 up to 10 members
- Business$5 per user a month
- AI$7 per user a month
Why do most AI projects fail to deliver value?
Only 5% of companies get value from AI at scale
Adoption is not the hard part. Capture is. Most programmes never agreed what good would look like, so nobody can show what changed.
88%use AI in at least one function
39%see any EBIT impact from it
5%reach AI value at scale, of 1,250+ companies studied
| Stage | Share of organisations | Source |
|---|---|---|
| use AI in at least one function | 88% | McKinsey, The state of AI in 2025 |
| see any EBIT impact from it | 39% | McKinsey, The state of AI in 2025 |
| reach AI value at scale, of 1,250+ companies studied | 5% | BCG, The Widening AI Value Gap (Sep 2025) |
- 49 points lost between adoption and EBIT impact
- 34 points lost between EBIT impact and value at scale
How is DILR different from other AI consultancies?
We agree the number before we build anything
Pilots rarely fail on the model. They fail because nobody decided which number should move, or owned it after launch.
Use case picked for visibility
It demos well to the steering committee and moves nothing.
Success defined afterwards
The metric is chosen once the result is already known.
Governance retrofitted
Compliance arrives as a scramble, and use case two costs as much as one.
No owner at handover
The team leaves and the capability quietly decays.
Start at the outcome, not the tool
Money, hours, error rates, response times. Whatever you would actually count as a win, we work backwards from that.
Measurement designed before the build
Baseline, metric and review date fixed while it is still cheap to disagree.
Governance in stage 03
Audit-ready by design, so the second placement is cheaper than the first.
Graduated to a named owner
With the runbook and the evals. A dependency on us is a failure mode.
How does outcome-based AI consulting pricing work?
You pay the value invoice only after the number moves
Consulting normally bills for effort. We bill for verified outcomes, agreed in writing before the build.
- 1. Agree
The measure, first
Whichever outcome you picked, its baseline and review date are fixed in writing before the build starts. No moving goalposts, in either direction.
- 2. Build
Shipped instrumented
The placement carries its own measurement, so proving it worked is a query either of us can run, not a slide.
- 3. Verify
Then, and only then, bill
Checked against the agreed baseline. If it did not move, the value invoice does not go out. That risk is ours, deliberately.
What is the DATS AI consulting method?
Five stages take AI from diagnosis to production
DATS is DILR's five-stage AI consulting system. Every engagement is a slice of the same stages, and you can enter at any one.
- 01
Discover and diagnose
4 to 6 weeksMap systems, find leverage, score risk. Placement Diagnostic
- 02
Prioritise and place
2 to 3 weeksThree placements, ranked roadmap. Placement Diagnostic
- 03
Operating model
4 weeksGovernance, RACI, lifecycle. Operating Model
- 04
Pilot to production
8 to 12 weeksReal users, eval harness, cutover. Execution Office
- 05
Scale and run
OngoingDrift managed, next placement. Execution Office
Which AI consulting engagement do I need?
Three consulting engagements, each a slice of DATS
Start with the diagnostic if you are unsure where AI fits. Most clients enter there.
- Stages 01 and 02
Placement Diagnostic
Ranked roadmap and placement map. No tool-shopping.
4 to 6 weeksPlacement Diagnostic details
- Stage 03
Operating Model
Governance, RACI, lifecycle and org design. Your AI, owned from day one.
6 to 10 weeksOperating Model details
- Stages 04 and 05
Execution Office
Embedded delivery, pilot to production, graduated to your team.
12+ monthsExecution Office details
Does DILR work with small businesses?
Yes. We work with a two-person office or a regulated bank.
The method stays the same at any headcount. Only the size of the placement changes.
Small business
Start self-serve, today
Estate agents, clinics and practices that lose business to unanswered calls. An AI receptionist from Dilr Voice starts at $0.14 a minute with $10 of trial credits and no card.
AI receptionist for estate agentsAI receptionist for clinics
Scaling company
One placement, proven
A process that is visibly costing you and no capacity to research it. The Placement Diagnostic picks the one thing worth doing, and the do-not-do list saves you the rest.
Enterprise and regulated
Governed, audit-ready, on your hardware
Healthcare and financial services that cannot send data to a hosted API. Governance designed against the EU AI Act and UK guidance, with Dilr Mira on-premise or air-gapped.
How much does an AI receptionist cost?
An AI receptionist costs $0.14 a minute. Human answering services publish $1.95 to $5.00.
Effective rate per included minute on each provider's own published rate card. Plain division, not modelling.
AI receptionist for UK estate agents
A quarter of UK sellers say 60 minutes is the longest acceptable wait for an agent to reply.
Estate agent receptionistAI receptionist for clinics and GP practices
202167.6%202349.8%Patients who found it easy to get through by phone. It books and routes. It does not triage.
Clinic receptionist
Seller expectations: Zoopla and YouGov via The Negotiator, November 2025. Phone access: NHS England, citing the GP Patient Survey 2023.
Is Dilr Voice better than Vapi, Retell, Bland or Synthflow?
Compare Dilr Voice with Vapi, Retell, Bland and Synthflow on published prices
Each comparison sets published prices and features side by side, including when the other platform is the better pick.
What AI problems does DILR fix?
The four AI problems businesses call us about most
Faster code, flat delivery
AI made your developers quicker and nothing ships sooner. The review queue absorbed the gain, and incidents went up with it.
The AI bill keeps climbing
Token prices fell and your invoice still grew. Usage outruns unit price when nobody owns the total.
Great demo, broken in production
The agent worked once in front of an audience. Without evals, permissions and observability it cannot be trusted with customers.
Nobody can find the answer
The policy exists, in a PDF, somewhere. McKinsey puts information-hunting at about a quarter of the working day.
Where is DILR.ai based?
London, with a working day that overlaps New York and Mumbai
Registered in England as Dilr.ai Ltd, Companies House 16842656. London concentrates finance, healthcare and regulation, so the compliance questions in stage 03 are ones we have already answered.
| City | 09:00 to 17:00 local, in UTC |
|---|---|
| New York | 14:00 to 22:00 |
| London | 09:00 to 17:00 |
| Mumbai | 03:30 to 11:30 |
- 3 working hours shared with New York
- 2.5 working hours shared with Mumbai
Is our AI ready for the EU AI Act?
We track enterprise AI compliance changes for UK organisations in a public changelog, updated monthly
Quick answers
Answers to common questions about DILR.ai
What it does, which product fits, what it costs, and how compliance is handled.
- DILR.ai is a London AI company that designs, deploys, and governs enterprise AI systems. It ships four product lines, Dilr Voice (AI voice agents for inbound and outbound calls), Dilr Mira (private clinical language models that run on your own hardware), Cognibl (work management where people and AI agents share one board and nothing reaches done without attached proof) and Dilr Academy (an AI tutor that builds interactive courses, licensed to universities and colleges), plus the DATS AI consulting system.
- If phone calls are the bottleneck, start with Dilr Voice. If you are a university, college or training provider that wants courses built and taught rather than bought off a shelf, Dilr Academy. If clinical documents cannot leave your infrastructure, Dilr Mira. If the problem is that AI agents and people are working the same backlog with no record of what was actually proved, Cognibl. Not sure where AI fits at all? That is what the consulting practice exists for.
- Both, deliberately. The products run on their own (each with its own pricing, detailed on its page), and the DATS consulting system exists for enterprises that need AI placed inside existing systems and governed properly. The consulting practice uses the same production experience the products are built on.
- SMBs that need calls answered (estate agents, clinics), universities, colleges and training providers that want AI-built courses, and regulated enterprises (healthcare, financial services) that cannot send data to hosted APIs. If you are unsure where AI fits, the 4 to 6 week Placement Diagnosticmaps it against your P&L.
- Pricing is published per product, not quoted on request. Dilr Voice starts at $0.14 a minute with $10 of trial credits and no card, plus a monthly fee per phone number. Cognibl is free for up to 10 members, then $5 per user a month, and Dilr Mira ships as open weights under Apache-2.0. Dilr Academy is the one exception: it is licensed per institution and scoped on a call, because what a university buys is a course programme rather than a seat. Consulting starts with a free 30-minute consultation.
- Yes. DILR works with businesses of every size, from a two-person office to a regulated bank. A small business can start self-serve the same day: Dilr Voice runs as an AI receptionist from $0.14 a minute with $10 of trial credits, no card and no contract. The consulting practice exists for the cases where the right placement is not obvious.
- Governance is designed in stage 03 of DATS, not retrofitted after a pilot. DILR maps each placement against the EU AI Act and UK guidance and publishes a compliance changelog as the rules change. Dilr Voice carries per-country call compliance rules for 8 countries, and Dilr Mira runs on-premise or air-gapped when documents cannot leave your infrastructure at all.
Free first consultation
Book a free 30‑minute AI consultation
Bring the process that costs you most. We will tell you whether AI belongs near it, what it would take and what it would be worth. If the answer is no, we will say that too.