Most AI projects die the same death. A team builds a prototype, a demo impresses the board, budget gets approved, and then — nothing. The model sits in a notebook. The integration stalls. Six months later, someone asks "what happened to that AI thing?" and the answer is a shrug.
This is not a technology problem. It is a design problem.
At DILR, we call our approach Value by Design — the principle that every AI system must have its value pathway engineered before a single line of model code is written. Not value as a slide deck promise. Value as a measurable, traceable, auditable line on the P&L.
The 88-33-14 problem
McKinsey's State of AI 2025 report tells a brutal story:
- 88% of organisations are aware of generative AI
- 71% are actively experimenting
- 33% have anything in production
- 14% report material impact on EBIT
That funnel is not a technology gap. It is a value design gap. Companies know AI exists. They can build prototypes. What they cannot do is connect AI output to business outcomes in a way that survives contact with real operations.
AI does not fail because models are bad. AI fails because nobody designed the value chain — from model output to operational change to financial impact — before building.
What "Value by Design" actually means
Value by Design is not a framework you overlay on top of a finished product. It is a set of constraints you apply before you start building.
1. Start with the P&L line, not the model
Before choosing a model, architecture, or vendor, we ask one question: which P&L line moves?
- Is this reducing cost-to-serve? (Contact centre headcount, manual QA, data entry)
- Is this increasing revenue capture? (Missed leads, faster response, broader reach)
- Is this protecting margin? (Compliance fines, churn prevention, quality assurance)
If the answer is vague — "it improves efficiency" or "it enhances the customer experience" — the project is not ready. Efficiency is not a P&L line. Experience is not a metric. We need: "Reduces average handling time by 40 seconds per call, saving £180K annually across 12 agents."
2. Design the measurement before the solution
Every DILR deployment starts with a measurement contract — a one-page document that states:
| Element | Example |
|---|---|
| Metric | Missed call rate |
| Baseline | 22% (industry average for UK estate agents) |
| Target | < 5% |
| Measurement method | Twilio call logs vs. answered calls, weekly report |
| Owner | Operations manager |
| Review cadence | Weekly for 90 days, then monthly |
This is not bureaucracy. This is insurance. Without a measurement contract, you have no way to know if the AI is working, no way to justify continued investment, and no way to improve.
3. Place AI where humans consistently fail
Humans are brilliant at judgement, empathy, and novel problem-solving. Humans are terrible at:
- Consistency at scale — answering the 200th call of the day with the same energy as the first
- Availability — being there at 3am when a landlord calls about a burst pipe
- Speed of response — calling back a web lead within 60 seconds instead of 4 hours
- Data capture — remembering to log every field in the CRM after every interaction
Value by Design means placing AI exactly in these gaps — not replacing human judgement, but filling the holes where humans consistently drop the ball.
4. Build for graduation, not dependency
The worst AI deployments create permanent vendor lock-in. The best ones graduate — they transfer knowledge, process, and eventually control back to the client's team.
Every DILR engagement has a graduation plan:
- Month 1-3: We operate. Client observes.
- Month 4-6: Client co-operates. We coach.
- Month 7-12: Client operates. We audit.
- Month 12+: Client owns. We are available for edge cases.
How this shows up in Dilr Voice
Dilr Voice, our multi-agent voice AI platform, is a pure expression of Value by Design. Every feature maps to a value chain.
The inbound value chain
The P&L line: Revenue captured from calls that would have gone to voicemail. For a typical UK estate agent handling 15-20 calls daily with a 22% miss rate, that is 3-4 missed opportunities per day. At an average commission of £3,800, even recovering one deal per month adds £45,600 annually.
The outbound value chain
The P&L line: Cost-per-qualified-lead. A junior SDR costs £34K/year and makes perhaps 40 meaningful calls per day. Dilr Voice makes unlimited calls at a fraction of the per-minute cost, with perfect consistency on every dial.
The multi-agent value chain
Each agent has one job, one personality, and one set of tools. The caller perceives one seamless conversation. The business gets specialist handling at every stage.
The adoption gap: a visual breakdown
Stanford's AI Index 2026 and ServiceNow's Enterprise AI Maturity Index paint the same picture:
The drop from 88% to 14% is not a technology problem. It is a value design problem. Every company that falls out of the funnel made the same mistake: they started with the model instead of the outcome.
The anti-patterns we refuse to build
Value by Design is also about what we do not do:
We do not demo without a deployment plan. A demo that impresses the board but has no path to production is worse than no demo at all.
We do not measure activity, we measure outcomes. "We processed 10,000 calls" means nothing. "We recovered £180K in previously missed revenue" means everything.
We do not build black boxes. Every AI decision in Dilr Voice is logged, transcribed, and auditable. Sentiment scores, call summaries, agent switching decisions — all visible.
The DATS system: Value by Design operationalised
Our consulting methodology — Discover, Assess, Transform, Scale — is Value by Design turned into a repeatable process:
Why this matters now
The AI industry is entering its accountability phase. The hype cycle has peaked. Boards are no longer impressed by demos — they want ROI. CIOs are being asked to justify AI spend with the same rigour as any other capital investment.
Companies that built AI for the name are discovering that AI without value design is just expensive infrastructure.
Companies that built AI for the value are compounding their advantage.
The gap between these two groups will only widen.
If you are building AI and cannot point to a specific P&L line it moves, stop building and start designing. The model is the easy part. The value chain is the hard part. And the hard part is where the money is.
Start here
-
Audit your current AI initiatives. For each one, can you name the P&L line it affects? If not, it is a hobby project, not a business investment.
-
Write measurement contracts. Even for projects already in flight. It is never too late to define what success looks like.
-
Talk to us. Our Placement Diagnostic takes 4-6 weeks and tells you exactly where AI belongs in your organisation — and more importantly, where it does not.
Or try Dilr Voice yourself — $10 free credits, no card required, first agent live in 5 minutes. See Value by Design in action.