Most AI implementations fail not because the technology doesn't work, but because organisations spend six months planning and then try to deploy everything at once. By the time they're ready to build, the business context has shifted, the champions have moved on, and the appetite for change has evaporated.
At Iyara Labs, we've built a different approach. Our Audit → Rank → Ship framework compresses discovery, prioritisation, and first deployment into three weeks. You get a working system in production before your previous vendor has finished their discovery questionnaire.
Here's exactly how it works.
Phase 1: The Audit (Days 1–3)
The audit is not a strategy workshop. It's fieldwork.
We spend three days inside your operations, interviewing the people who actually do the work, not just the people who manage them. We watch how tasks get handled. We ask uncomfortable questions: "How long does this take you?" "What do you do when the information is wrong?" "What would happen if we removed this step entirely?"
What We're Mapping
The central output of the audit is what we call the human effort graph: a visual map of where time is being spent across your organisation, annotated by task type:
- High-frequency, predictable tasks: these repeat dozens or hundreds of times per day and follow consistent patterns (routing enquiries, extracting data from documents, filling forms from structured inputs)
- Decision tasks with clear rules: approvals, escalations, or responses that follow deterministic logic most of the time
- Creative or judgement-heavy tasks: these stay with humans; we flag them but deprioritise them
We're explicitly looking for tasks with two properties: high repetition and predictable inputs/outputs. Those are the tasks where AI delivers ROI in weeks, not years.
Stakeholder Interviews
We speak with three types of stakeholders:
- Operators: the people doing the work day-to-day. They know where the friction is. They know which tools break, which processes have workarounds, and where they're copy-pasting the same thing forty times a day.
- Managers: they know what's expensive, what's slow, and what they've tried before. They also hold the keys to data and integration.
- Decision-makers: one or two conversations to understand strategic priorities, risk tolerance, and what "success" looks like at the executive level.
Deliverable: The Opportunity Matrix
By day three, we produce a ranked opportunity matrix: a structured document listing every automation or AI opportunity we identified, each scored across three dimensions: estimated time/cost impact, data availability, and implementation complexity. This is not a slide deck. It's a working document we use in the next phase.
Phase 2: The Rank (Days 4–5)
The opportunity matrix typically contains eight to twenty items. We can't build all of them in three weeks, and we shouldn't try. The Rank phase is where we apply rigour to the list.
The Scoring Model
Each opportunity is scored on three axes:
1. Impact (0–10) How much time, cost, or revenue does this opportunity affect? We convert estimates to a common unit (usually hours per week and cost per task) so we can compare apples to apples. A task that takes one person two hours per day scores higher than one that saves an analyst thirty minutes per week, even if the latter feels more technically impressive.
2. Build Effort (0–10, inverted) How complex is this to build? We consider: data availability (does clean, structured data already exist?), integration requirements (does it plug into an existing system or require custom connectors?), model selection (is this a standard language task or does it require fine-tuning?), and build time. Lower effort scores higher.
3. Risk (0–10, inverted) What are the consequences of errors? A system that drafts internal summaries has low risk; a system that sends customer-facing messages has higher risk; a system involved in financial decisions has the highest. We weight risk heavily for first deployments; a failure in the first system poisons the well for everything that follows.
Selecting the Top 3–5
We apply the scores, review the output with the client, and select the top three to five opportunities. For each selected item, we produce a one-page brief covering:
- Success metrics: what does "working" look like? (e.g., "lead response time under 2 minutes, 95% of the time")
- Data requirements: what data does the system need, in what format, and is it available today?
- Integration points: which systems does this connect to, and what's the access path?
- Estimated build time: in working days
Deliverable: The Roadmap
The Rank phase closes with a costed, prioritised roadmap. Every item has a clear definition of done. The client reviews it, we align on the first build, and we move immediately.
Phase 3: The Ship (Weeks 2–3)
This is where most consultancies stop: they hand you the roadmap and invoice. We build.
What "Ship" Means
The highest-impact item from the roadmap goes into immediate development. We build the real system, not a proof of concept, not a demo, not a sandbox prototype. We connect to real data sources, integrate with the tools your team actually uses, and test against real edge cases.
By the end of week three, your team is using the system in their actual daily work.
What the Build Looks Like
Our standard build sequence for a two-week sprint:
- Days 1–2: Architecture, data pipeline setup, API access and credentials, development environment configured
- Days 3–7: Core system build: the AI logic, integrations, and primary user interface
- Days 8–10: Testing against real data, edge case handling, error logging, and monitoring setup
- Days 11–12: Internal review, bug fixes, performance checks
- Days 13–14: Client onboarding, team training, go-live, handover documentation
The Handover Package
We do not build systems you're dependent on us to maintain. Every delivery includes:
- Architecture documentation: what the system is, how it works, and what it connects to
- How-to guides: written for the people who will manage it, not for engineers
- Edge case register: documented known-failure modes and how the system handles them
- Escalation paths: what happens when the AI gets something wrong, and who is responsible
- Runbook: how to update, monitor, and extend the system over time
We then run a half-day training session with the team that will own it. The goal is that within 30 days of delivery, your team can modify the system themselves for routine changes.
Two Examples From Our Work
WhatsApp Lead Qualification: Real Estate Agency
During a three-day audit with a UAE real estate agency, we found their top friction point: agents were manually responding to dozens of WhatsApp enquiries per day, spending the first three to five messages just gathering basic qualification information before they could determine if a lead was worth pursuing.
We ranked it first: high repetition, predictable inputs (location preference, budget, timeline), low risk (agents review and confirm before progressing), and clean integration path via WhatsApp Business API.
Two weeks later, the agency had a live WhatsApp bot that qualifies leads automatically, enriches them with CRM data, and routes them to the right agent with a pre-completed brief. Agents now spend their time on leads that are ready to view, not on information gathering.
Customer Onboarding Agent: B2B SaaS
A SaaS company was losing users in the first 72 hours of their trial, not because the product was bad, but because their onboarding flow required users to configure several steps manually before seeing value. Support tickets for onboarding issues were consuming two engineers' time each week.
We audited their onboarding flow over three days, ranked an AI-guided onboarding agent as the top opportunity, and shipped it in two weeks. The system walks new users through configuration, detects where they're stuck, and surfaces contextual help, all without engineering involvement.
Their CTO told us we shipped before they had finished evaluating vendors.
Why This Works
Most AI projects fail for a predictable reason: they try to be comprehensive before they're operational. Long discovery phases, extensive requirements documents, and phased rollouts that never reach phase two.
The Audit → Rank → Ship framework works because it forces three things:
- Triage first. Not every process should be automated. The Rank phase forces you to choose, which means the build starts with the highest-value opportunity.
- Ship something real. A working system changes the internal conversation from "can AI help us?" to "what should we build next?" The first ship creates momentum.
- Compress the timeline. Project fatigue is real. When the gap between "we decided to do this" and "this is live" is three weeks, you don't lose people.
The framework also surfaces something less obvious: the audit itself (even if you never build anything) is valuable. Companies routinely tell us they learned more about their own operations in three days of structured observation than in years of internal analysis.
FAQ
How long does the audit phase actually take? Can it be done remotely?
The audit takes three working days for most organisations. Remote audits work well for digital-native businesses: we run structured interview sessions over video, review screen recordings of workflows, and access documentation in shared drives. For operations with significant physical components (warehouse, in-person service), we prefer at least one day on-site combined with two days of remote follow-up.
What if our highest-priority opportunity is too complex to ship in two weeks?
We see this occasionally. If the top-ranked item can't be shipped in two weeks as a complete system, we modularise it: identify the highest-value sub-component that can ship independently and produces value on its own. In practice, over 90% of the opportunities we identify can ship as a working v1 in two weeks. Complex enterprise integrations occasionally push to four weeks.
Do we need to prepare our data before the engagement starts?
No. Data readiness assessment is part of the audit. We find out what data exists, where it lives, and what condition it's in. We've worked with clients who have clean, well-structured data and clients who have data spread across three spreadsheets, an old CRM, and someone's email inbox. The data situation affects which opportunities rank highest; it doesn't prevent the process.
What happens after the first system ships? Do you help with the rest of the roadmap?
The roadmap we produce in Phase 2 covers three to five opportunities. After the first ship, most clients continue with us to build the next item on the list. Some prefer to build internally from that point, using the architecture we've documented as a template. We support both paths. The handover documentation is designed so that a competent internal team can take the system from there.
How is this different from what a large AI consultancy does?
Large consultancies typically front-load discovery (sometimes three to six months of it) before any building begins. Their engagements are designed for enterprise governance structures where that timeline is acceptable. We work with startups and scaleups where speed of execution is itself a competitive advantage. We're also building systems, not reports. Our deliverable at the end of three weeks is a live system your team is using, not a strategy document recommending one.
Ready to see the Audit → Rank → Ship framework applied to your business? Talk to us about your AI consulting engagement →