Agent Strategy & Readiness · Example roadmap
Rolling out agents, wave by wave
An example roadmap for three agents in twelve months. Each further agent starts only once the previous one is live and delivering measured results. You release budget wave by wave, and BOTFORCE runs the agents from the first go-live.
The plan
Twelve months, three waves, one operation
Discovery sets the order and the budget, the foundation is built once, and each wave begins only after the previous go-live. From the first go-live, BOTFORCE runs the agents.
- Discovery
- Foundation
- Pilot
- Build & test
- Live at an autonomy level
- Operated by BOTFORCE
- Gate G1–G4
- Value review G5 every 90 days
- Starts only after the previous agent’s go-live
The plan as a list
- Discovery: 4 weeks with BOTFORCE Discovery · €12k, month 1
- Foundation: Built once, month 2–3
- Agent 1 · Incoming invoices: Pilot, month 2; Build & test, month 3–4; Live at L3, month 5–7; L4 up to €10,000, from month 8; G1 before month 2; G2 before month 3; G3 before month 5; G4 before month 8
- Agent 2 · Order confirmations: Pilot, month 5; Build & test, month 6–7; Live at L3, from month 8; G1 before month 5; G2 before month 6; G3 before month 8
- Agent 3 · Service requests: Pilot, month 8; Build & test, month 9–10; Live at L2, from month 11; G1 before month 8; G2 before month 9; G3 before month 11
- Operated by BOTFORCE: 1 agent, month 5–7; 2 agents, month 8–10; 3 agents, from month 11; G5 before month 8; G5 before month 11
- Budget release: Wave 1 €60k before month 2; Wave 2 €26k before month 5; Wave 3 €26k before month 8
Cost control
Why not all agents at once?
Budget for the next wave only
Only the next wave is ever released. If a pilot misses its gate, you have spent its pilot budget, not the budget for three agents, and what you learned shapes the next choice.
Build once, reuse
Identity, policy, audit trail and the eval harness are built with the first agent. The next ones reuse them; in the worked example the build cost per agent drops from €36k to €26k.
Decide on actuals
Before the next wave starts, the previous agent delivers measured figures: cost per case, exception rate, eval results. The next business case uses them instead of estimates.
One pilot at a time for the business
Every pilot needs subject-matter experts for the exception queue and the golden set. In sequence that ties up one team, not three at once.
Worked example
Same break-even, far less capital at risk
Cumulative result of build, operations and savings over 24 months. In parallel the savings arrive sooner, but far more money is committed before the first agent shows whether it delivers what the business case promised.
- Wave by wave
- All at once
| Figure | Wave by wave | All at once |
|---|---|---|
| Committed before the first go-live | €72k | €155k |
| Peak capital tied up | €85k | €155k |
| Highest monthly spend | €24k | €42k |
| Break-even | Month 16 | Month 16 |
| Position after 24 months | +€144k | +€140k |
Assumptions
- Discovery assessment €12k in month 1, foundation €24k in months 2–3, the same in both variants.
- Build per agent including pilot: €36k for the first, €26k for each further one, because foundation, tools and evals are reused.
- All at once: three agents at €36k each without reuse plus 10% coordination, spread over months 2–5, all three go live in month 6.
- Operations including model cost from go-live: €3k / €2.5k / €3k per month per agent.
- Savings at full effect: €10k / €7k / €8k per month; 50% in the first month after go-live, 75% in the second.
- Undiscounted and before tax, rounded. With your figures, BOTFORCE Discovery calculates the business case per process.
Gates
Every diamond is a decision
Every gate has a question, criteria and an accountable role. BOTFORCE Discovery tracks the same gates for every process: go, review or no-go, with reasons and a date.
- G1
Qualify
Start a pilot?
- Positive three-year net present value
- Feasibility assessed
- Sponsor and owner named
- Data access settled
CoE lead
- G2
Pilot → build
Build it fully?
- Eval pass rate above target, e.g. ≥ 90% on the golden set
- Cost per case within ±20% of the model
- No critical guardrail violations
- Exception rate on plan
CoE lead
- G3
Go-live
Go to production?
- Eval suite green, security review passed
- Guardrails and stop switch active
- Monitoring live, rollback plan in place
- Obligations settled with your legal counsel
Process owner
- G4
Raise autonomy
Next level or wider scope?
- Success rate stable over weeks
- Intervention rate below threshold
- No critical incidents
- Realised value evidenced
CoE lead
- G5
Value review
Continue, expand or refine?
- Actuals for the last three months recorded
- Net benefit ≥ 80% of plan
- Cost per case within ±20%
- No open critical incidents
Sponsor · every 90 days
Plus G6: when the value is gone, an agent is shut down cleanly. Access and identity are revoked, data is kept or deleted according to retention rules, and lessons learned are recorded.
The foundation
BOTFORCE Discovery carries pipeline, business case and operations
The roadmap is not drawn on slides. Discovery assesses every process, calculates it transparently and follows it through the gates into operations.
- Pipeline
Which agent first?
Every process with its route (RPA, hybrid or agent), autonomy level and phase from assessment to operations, on heatmap and pipeline.
In the roadmapOrder of the waves
- Business case
What may the wave cost?
Build, run and operating cost per route, net benefit, payback and three-year NPV. Calculated deterministically; no figure comes from a language model.
In the roadmapBudget release per wave
- Operations
Is the agent delivering?
Monthly actuals (cases, success rate, cost per case, HITL rate, incidents) and the gates with decision, criteria and accountable role.
In the roadmapRaise autonomy (G4) and value review (G5)
Agent operations
Go-live is where our work begins
BOTFORCE does not stop at analysis and build; we run the agents too, as a managed service with a dedicated tenant for your organisation. Every further agent joins the same operation, with the same controls and one shared report.
Operated by BOTFORCE from the first go-live, month 5 in the example, and beyond the roadmap.
Monitoring and cost per case
Runtime, errors and model cost per agent, with thresholds and alerts before a budget tips over.
Exceptions with service levels
We run the queues, your experts decide the cases, with agreed response times and a clear overview.
Evals on every change
New model, new prompt, new policy: first the eval suite, then production. Every change is versioned.
Monthly report
Actuals per agent (cases, success rate, cost per case, HITL rate, incidents) as the basis for every value review.
Stop switch and incidents
Every agent can be stopped outside the model. Incidents are reviewed, and failure cases go into the golden set.
Review every 90 days
Together with the sponsor we decide: continue, raise autonomy, refine or shut down cleanly.
Example and worked example with assumed values for the fictional Aurelia Haustechnik GmbH · Not an offer and not a forecast for your organisation · Not legal advice · As of September 2026