Scale
The number moved and the workflow held. Extend it to the rest of the function.
The delivery method
From repeated task to a working operation.
Start with one workflow and the people who run it. We map the real work, agree the controls and build a measured pilot before deciding what deserves to scale.
One accountable team. A measured baseline. A documented decision at day 90.
Explore the process
Five phases. A clear decision at every handover.
Choose a phase to see the work, a sample deliverable and the approval needed to move forward.
Decision and roadmap
Start with the work. Not the technology.
We sit with the people doing the job, follow the real handovers and find the exceptions. Together, we choose one workflow worth improving.
What you take away
- Exception map per candidate process
- Scored opportunity matrix
- 90-day plan with a named first workflow
01 / FoundationMap the current process
01 / InputA request arrives
Channel, context and source record
02 / HandoverThe team checks the detail
Rules, judgement and missing information
Map the exceptions before automating
03 / OutputA record moves forward
Next action and accountable owner
The control gate
Agree the opportunity and baseline before committing to a build.
Discovery authorises a decision. Design requires a signed boundary. Production access and wider rollout need their own approvals; completing one phase never silently authorises the next.
Inside the engagement
Review the control commitments Know what your team receives.
Each phase has defined work and tangible outputs. Open a phase for the detailed scope; timing and dependencies are agreed in your project plan.
01Foundation
What we do
- 4–6 stakeholder interviews across the affected functions
- Silent shadowing — one hour per role, watching the actual screens
- Export and analysis of your CRM, ticket, call, WhatsApp and ERP logs
- An exception map per process: what follows the happy path, what does not
- A costed baseline per process using observed times, not self-reported ones
- A scored opportunity matrix — value against feasibility
What you receive
- Costed baseline, one page per process
- Exception map per candidate process
- Scored opportunity matrix
- 90-day plan with a named first workflow
- A written do-not-automate list
You leave with a roadmap you could hand to any vendor. If nothing here clears the bar, we say so in writing and you owe us nothing further.
02Architecture
What we do
- Process redesign against the exception map, not the org chart
- Agent and workflow design, with the human approval points named
- Integration architecture against your real systems and their real limits
- An automation boundary matrix: runs alone / needs approval / never automated
- Data handling, residency and retention design for your jurisdiction
- The business case, in your finance team's format
What you receive
- Solution architecture document
- Automation boundary matrix, signed off by your process owner
- Integration and data-flow specification
- Acceptance tests, written before the build
- Business case with payback period
Nothing gets built until the boundary matrix is signed. If your process owner will not sign it, the process is not ready and we say so.
03Pilot
What we do
- Build against the signed architecture, in your environment
- Real integrations from day one — no mocked systems
- Human-in-the-loop controls, approval queues and escalation paths
- Audit logging and observability wired in before go-live, not after
- Side-by-side running against the manual process
- Training for the team who will actually use it
What you receive
- A working automation in production use by one team
- Measured before-and-after against the agreed metric
- Audit trail and monitoring dashboard
- Runbook and escalation procedure
- Scale / pause / redesign / stop recommendation
At day 90 there are four honest outcomes: scale it, pause it, redesign it, or stop. Stop is a real option and we will recommend it if the number did not move.
04Production
What we do
- Hardening: retry logic, failure modes, rate limits, graceful degradation
- Rollout across teams, sites or entities
- Adjacent workflow extension where the architecture already supports it
- Change management: champions, training, and a named internal owner
- Handover documentation and admin training
- 90 days of hypercare after go-live
What you receive
- Production system across the full function
- Named internal owner, trained
- Complete technical and operational documentation
- Quarterly value review, in your finance team's numbers
- 90-day hypercare included
You own the outputs, documentation and configuration. We do not hold deliverables hostage; continued operation depends on the third-party accounts, licences, hosting and source systems named in the project documents.
05Run
What we do
- Monitoring, alerting and drift detection
- Fixes when upstream systems change
- A committed number of new or amended automations each month
- Prompt, model and cost optimisation as the market moves
- Quarterly value review against the original baseline
- Roadmap for the next cycle
What you receive
- Uptime and outcome reporting
- Monthly automation credits
- Quarterly value review with your finance team
- Next-cycle scope
Model and telephony costs are passed through at cost, on your accounts where possible. We do not mark up inference.
The day-90 decision
Evidence decides what happens next.
Before the build, we agree the measure, its baseline and the threshold for success. The review ends in a written recommendation your team can act on.
- Observed work and system logs establish the baseline
- The process owner agrees the acceptance criteria
- The pilot tests failures and exceptions as well as routine work
Pause
It works, but something upstream — data, a system, a team change — has to land first.
Redesign
The approach was wrong but the opportunity is real. We rebuild it once, at our cost.
Stop
The number did not move. We say so, we write down why, and you stop spending on it.
Prepare for the first conversation
Bring the workflow. We will work through the rest.
There is no need to arrive with an AI specification. The most useful starting point is a repeated task and someone who knows how it really runs.
- The person responsible for the process
- The systems and handovers involved
- A typical example, with sensitive information removed
- Where the work slows down or needs a judgement call