How it works
From AI pilot to production, one decision path at a time.
Three phases, and you can start at any of them. Each asks the same question: how do people and AI agents do this work together, and who answers for it?
Discovery → Design → Scale
Three phases with gates the business signs off.
The phases form the outer loop, which decides the operating model. Building and running the agents is the inner loop. We build the first agents with you to prove a path; production builds and Run go to your team or an engineering partner you choose. We keep the operating model current as the market and the environment change.
-
Phase 1
Discovery
The decisions in a function, ranked by value
- 1 Discount on a proposal Approve
- 2 Account priorities for the week Approve
- 3 Routing inbound leads Monitor
Business case for each path: what it is worth, and what its autonomy costs
Signed off at the Portfolio gate
-
Phase 2
Design
Who does each step, and who answers for it
Discount on a proposal
- Preparation Agent gathers the facts
- Decision Person signs off
- Execution Agent carries it out
- Consequence Feeds back into preparation
Supervision mode Approve Accountable: Sales Director
Signed off at Gate 1 and Gate 2. You own the design.
-
Phase 3
Scale
Live with a supervision mode and a named owner
Your team or an engineering partner you choose builds and runs; HAR Works keeps the operating model current
- Build
- Run
Live Approve Owner: Sales Director
Signed off at Gate 3
-
Living operating model
Reassess
Operating reviews keep the operating model current as the market and the environment change. What they find feeds the next Discovery and Design.
- Phase 1
Discovery
We map the decisions in a function and ask three questions of each: how often, the same way every time, and what an error costs.
- What you get
- A ranked list of decision paths, a named accountable person for each, and the paths worth designing first.
- Signed off at
- Portfolio gate
- Who does it
- HAR Works with your function, HR and IT
- Phase 2
Design
For each chosen path we first describe the path and the data and systems it needs. Then we design who does each decision, how closely people supervise it, and the human and agent roles.
- What you get
- A design your team or an engineering partner you choose can build safely. You own it at Gate 2.
- Signed off at
- Gate 1 and Gate 2
- Who does it
- HAR Works with the accountable person, HR and IT
- Phase 3
Scale
We name the build at Gate 3 and hand it to your team or an engineering partner you choose. After go-live, evidence from operation reopens the split and the supervision when things change.
- What you get
- Paths live with a supervision mode and a named owner, kept current as the market and the environment change.
- Signed off at
- Gate 3, then operating reviews
- Who does it
- Your team or an engineering partner you choose builds and runs; HAR Works keeps the operating model current
Gates
Four gates, each with one aim.
At each gate the business signs off what the gate covers, and the next phase starts after it. Evidence from operation can send a path back to Gate 2, where it is approved the same way as the first time.
- Stalled pilot Where you are now
- Discovery Function Sprint, 4–6 weeks Portfolio gate
- Design Who does what, who supervises, who answers. You own the design. Gate 1 and Gate 2
- Build and run Live with a supervision mode and a named owner Gate 3
- Reassess The operating model kept current as the market and the environment change Operating reviews
Evidence from operation feeds the next round, and can send a path back to Gate 2
- HAR Works with your function, HR and IT
- HAR Works with the accountable person, HR and IT
- Your team or an engineering partner you choose builds and runs; HAR Works keeps the operating model current
| Gate | What the business signs off | Its aim |
|---|---|---|
| Portfolio gate | Which paths to design, defer or drop, and in what order | The paths worth changing are chosen, with what their autonomy costs. |
| Gate 1 | The decision path, and the data and systems it needs | The path and its inputs are known well enough to design on. |
| Gate 2 | The design record, the human roles and the agent roles | The path can be built safely, with an accountable person who has the capacity to carry it. |
| Gate 3 | The implementation plan, which names the build and who does it | The release can go live without exposing what is still unresolved. |
People in every phase
The human side runs through every phase.
What leaves people’s work, what stays, what changes, and how much human work the target state needs. HR co-owns this part, and headcount is an output of the design.
- Human roles are approved at Gate 2, together with the agent roles.
- Supervision takes time, so the design checks it against the hours people have.
- Freed capacity counts when it is placed: released hours become value when they go to named work.
- Disagreement about a decision is recorded with both positions.
Commitments
You own the AI operating model.
The same team answers for the designed supervision working in operation. That is why we commit to the following.
-
You own the design
The operating model, the specifications and the data are yours.
-
Platform independence
The design is independent of any platform up to the engineering specification. Production builds run in your environment, on the platform you choose, and your team or an engineering partner builds them.
-
Fixed price per phase
Each phase is priced for an agreed scope before work starts. The living operating model is offered as a pilot, on a monthly subscription.
-
Free to continue or stop
Each phase is a decision of its own. The deliverables stay with you if you do not order the next phase.
Scope
What we do, and what we leave to others.
HAR Works owns the operating model, which is the method, the design decisions, and keeping them current as the market and the environment change. The technology that runs the agents in production stays with you and your partners.
What we do
- Develop and own the HAR method, and develop certification and quality checks for partners
- Sell fixed-price Discovery and Design, starting with commercial work
- Deliver flagship engagements ourselves, to keep learning and keep the proof current
- Run the living operating model, offered as a pilot on a monthly subscription, whatever platform your agents run on
- Name the build at Gate 3 and hand it to your team or an engineering partner you choose
- Design roles, capacity and trust as part of the change
- Build the first agents with you to prove a path
What we leave to others
- Building production agent systems Your team or an engineering partner you choose
- Running agents, technical operations, uptime and service levels Your team or an engineering partner you choose
- Choosing the platform and the vendor You
- Large HR changes, such as reorganisations Your HR team or an HR partner
Where you stand
The HAR adoption model: five levels and a wall.
The name of each level tells the business impact; the criterion tells the capability it takes. We work with organisations at the wall between level 2 and level 3.
- Level 1: Individuals experiment
- Level 2: The organisation learns from shared tools
- The wall: Don’t Scale Chaos
- Level 3: Measured business impact
- Level 4: Business impact scales across functions
- Level 5: Impact becomes self-reinforcing
- Personal productivity Levels 1–2
- Organisational productivity Levels 3–5
| Level | What you hear | What it takes |
|---|---|---|
| 1 · Individuals experiment | Some of us use AI in our own work. | Individual trials and learning. There is no shared tool and no business impact. |
| 2 · The organisation learns from shared tools | Everyone has Copilot, and we don’t know what it gives us. | Broad licensing, training and personal productivity, with no measured business impact. Tools and automations used in shared work also sit here when they have no owner and no measured business result. |
| The wall · Don’t Scale Chaos | Our pilots don’t make it to production. | Data, integrations, ownership, supervision and strategy are missing. Crossing the wall takes one business problem solved end to end, with a decision path described, a named accountable person, a supervision mode that follows from the consequence, and the result measured. |
| 3 · Measured business impact | One business problem is solved with AI, and the result is measured. | The organisation can design one workflow for people and agents. A decision path has an owner, its supervision mode follows from the consequence, and the result is measured as a business outcome. |
| 4 · Business impact scales across functions | We take the same operating model from one function to the next, and the result shows in several of them. | The organisation can take one operating model for people and agents into several functions. Shared rules cover accountability, decision rights, supervision and access. Anchors and capacity are known, there is one view of all agents, thresholds are agreed at design time, and the first review takes place 30 days after go-live. |
| 5 · Impact becomes self-reinforcing | Everyday feedback and evidence of results keep changing the division of work and the agents’ autonomy, and each round improves the next. | The organisation closes the feedback loop. Evidence from use changes the agents’ autonomy in both directions, the division of work, the agents’ life cycle and where resources go, without a separate development programme. People decide on the changes and answer for the loop. |
Questions
Questions about the approach.
Do you build the agents?
We build the first agents with you to prove a path; production builds and Run go to your team or an engineering partner you choose. We name the build at Gate 3 and keep the operating model current as the market and the environment change.
What does the living operating model add after go-live?
Models improve, people change roles and the business moves on, so a design starts to age once it is live. Regular reviews read the evidence from operation and propose changes to the split of work or the supervision mode. A named person in your organisation decides.
Who owns the design?
You do. The operating model, the specifications and the data are yours, and they stay with you if you stop after any phase.
Which AI platform do you recommend?
We leave that choice with you. The design is independent of any platform up to the engineering specification. Your team can build it on the platform and with the vendor you choose.
How do you measure savings?
On one decision path, before and after, by cost per accepted outcome, time and error rate. We agree how to measure it in the Sprint, and then we measure it.
Can we stop after one phase?
Yes. Each phase is a decision of its own, and the deliverables stay with you.
Start with a 30‑minute maturity check.
We place your company on the five-level adoption scale, match your situation to a typical buying trigger and recommend one next step. No charge, and no obligation to continue.