Agentic AI Strategy Roadmap: Stage 2 Pipeline Templates

An agentic AI strategy roadmap is the artefact that turns a Stage 1 readiness assessment into a twelve-month, board-ready plan — a rolling roadmap with quarterly OKRs, an executive memo, a capability-prioritisation matrix, an investment phasing model, and a stakeholder communication cadence. Stage 2 of the pipeline is not glossy slides; it is the small set of templates that lets a steering committee allocate capital and hold itself accountable.

Stage 2 matters because readiness findings without a roadmap die in slides. Most programs we audit have the right diagnosis from Stage 1 — the gaps, the severity, the maturity stage — and still cannot get a quarterly budget approved because the conversation never converts into a one-page plan the executive layer can sign. The roadmap is the conversion mechanism.

This guide covers why Stage 2 exists, the twelve-month roadmap template with quarterly capability stacking, three quarterly OKR templates, the two-page executive memo, the impact-and-feasibility prioritisation matrix, the three-phase investment model, the communication cadence, and the hand-off into the data foundation work that opens Stage 3. Every section is a template you can clone, adapt, and walk into next quarter's budget review with.

The most common failure mode after a Stage 1 readiness audit is a thirty-page report that becomes a thirty-slide deck that becomes an item on next quarter's steering-committee agenda — and then quietly disappears. Findings without an artefact the executive layer can fund are findings without consequences. Stage 2 exists to prevent that decay by converting diagnosis into a small set of decisions a board can ratify in a single meeting.

What Stage 2 produces, concretely: a twelve-month rolling roadmap organised by quarter and capability stack; quarterly OKRs with three objectives and three key results each; a two-page executive memo that frames the investment ask; a capability-prioritisation matrix that ranks initiatives by impact and feasibility; an investment phasing model across three tiers; and a stakeholder communication cadence that keeps the working group, the steering committee, and the board aligned across the year.

What Stage 2 does not produce: a vendor recommendation, a hiring plan, or a technical architecture. Those belong to Stages 4, 7, and 5 respectively. Stage 2 is deliberately strategic — the output is the smallest set of artefacts that lets capital flow with confidence into the technical stages that follow.

A useful diagnostic: if your Stage 1 audit closed more than ninety days ago and Stage 2 is still being drafted, the program has a cadence problem, not a content problem. Reopen the readiness scoring, refresh the highest-severity findings, and ship Stage 2 as a rolling document rather than a finished one. The roadmap is a living artefact; treating it as a one-off deliverable is the mistake.

The twelve-month roadmap is the spine of Stage 2. It is organised by quarter rather than by month — month-by-month plans imply a precision agentic-AI programs cannot deliver, and they over-fit to the team composition at the moment of drafting. Each quarter stacks new capability on top of the previous one rather than replacing it, so the program compounds rather than churns.

The template below is a one-page roadmap with four quarters, three capability tracks, and one milestone per track per quarter. Twelve milestones total, all binary-pass at the end of each quarter. Drop it into a Google Doc or a Notion page; the format is deliberately portable.

# Agentic AI Roadmap · 12 months · rolling, re-baseline quarterly

TRACK             | Q1                          | Q2                          | Q3                          | Q4
------------------|-----------------------------|-----------------------------|-----------------------------|-----------------------------
Capability        | Foundation: 1 internal      | Scale: 1 customer-facing    | Compose: multi-agent on      | Optimise: latency, cost,
                  | copilot in production       | surface with retrieval      | top of Q1+Q2 surfaces        | governance review, re-audit
------------------|-----------------------------|-----------------------------|-----------------------------|-----------------------------
Governance        | Risk register, policy,      | Incident response runbook   | First tabletop drill,        | Annual external audit,
                  | model approval workflow     | + drill cadence agreed      | model-card discipline live   | regulatory pre-staging
------------------|-----------------------------|-----------------------------|-----------------------------|-----------------------------
Enablement        | Engineering enablement      | Governance literacy for     | Business stakeholder         | Leadership briefing
                  | for 1st cohort (20 eng)     | leads + risk owners         | fluency programme            | + succession depth

INVESTMENT PHASE  | Phase 1 — Foundation        | Phase 1 — Foundation        | Phase 2 — Scale              | Phase 3 — Optimise
GATE              | Stage 1 audit ≥ Reactive    | Q1 milestones binary-pass   | Q2 milestones binary-pass    | Q3 milestones binary-pass

Three rules govern the roadmap. First, every milestone is binary — it shipped or it did not, no half-credit. Second, every milestone has a named owner from the working group. Third, every milestone has an inheriting Stage in the pipeline — Q1 governance work feeds Stage 8, Q2 retrieval work feeds Stages 3 and 5, Q4 audit work loops back to Stage 1. The roadmap is the sequencing artefact for the whole pipeline.

One observation from running this format across roughly twenty engagements: tracks of three are the cap. Adding a fourth track — typically marketing, finance, or a hyped capability nobody on the working group can actually own — sinks the roadmap. Keep it at three tracks and put everything else in the steering-committee risk paragraph.

OKRs translate the quarterly roadmap into a small, public scorecard the working group reviews weekly and the steering committee reviews monthly. The template is deliberately tight: three objectives per quarter, three key results per objective, nine measurable lines total. Programs that draft more than nine lines consistently miss more than half of them; programs that hold to the nine consistently ship six to seven.

The four-quarter OKR template below shapes one objective per track (capability, governance, enablement) with three key results each. Re-baseline at the end of every quarter — do not roll incomplete OKRs forward unchanged. If a key result missed, either it was the wrong KR or the program lacks the resourcing to hit it; either answer is a Stage 2 conversation, not a silent carry.

One discipline that separates OKR templates that work from those that decay: every key result is a number, a date, or a binary. "Improve retrieval" is not a KR; "retrieval recall ≥ 0.85 on the May eval set by 30 June" is. The act of writing KRs in that shape forces the working group to disambiguate the work, and the disambiguation is most of the value.

OKRs are reviewed weekly inside the working group and reported monthly into the steering committee. The format is a single traffic-light page — green, amber, red — with one line per KR. Anything more elaborate gets skimmed; anything less elaborate is invisible. The discipline is to keep the format constant across quarters so the committee learns to read it at a glance.

The executive memo is the artefact the board reads. Two pages is the cap — pages three through twenty get skimmed at best. The structure is borrowed from the corporate-strategy memo format used at large operating companies: problem, ask, plan, risk. Every section is two to four paragraphs. The whole memo fits on two sheets of A4 with eleven-point body type and a one-inch margin.

The template below is the literal skeleton we hand to working groups. Every section has a target length and a question it answers. Copy the skeleton, fill the sections, share with the sponsor for one revision pass, then send. Memos that go through more than two revision passes are usually memos that needed a tighter Stage 1 audit underneath them.

# Agentic AI Strategy — 12-month plan
TO:    Board / Steering committee
FROM:  AI working group · sponsor: <named exec>
DATE:  <quarter, year>
RE:    Twelve-month strategy & investment ask

------------------------------------------------------------
1. PROBLEM  (3-4 sentences · what is at stake)
Where the business is today on agentic AI capability. The
Stage 1 readiness score and maturity stage. The two or three
gaps that, if unfixed, will cost a quarter in the next year.

2. ASK  (one paragraph · the headline number)
Total twelve-month investment. Capital split by phase
(foundation / scale / optimise) and by category (people /
tools / external). The single sentence the board ratifies.

3. PLAN  (one page · the twelve-month roadmap, condensed)
Q1: foundation. Q2: scale. Q3: compose. Q4: optimise.
Three tracks (capability, governance, enablement). One
milestone per track per quarter. Named owners. Investment
phasing gates between phases.

4. RISK  (3-4 sentences · what could go wrong)
Top three risks with mitigation. Drift risk on tools and
models. Skills lag risk. Governance pre-staging for
regulation. The trigger conditions for re-baselining.

5. DECISION  (one sentence · what you are asking for)
"Approve the twelve-month plan and the Phase 1 capital
allocation; re-decide Phase 2 at end of Q2."

One forward projection worth naming: as regulatory pressure on agentic AI increases through 2026 and into 2027, the executive memo is also the document that gets cited in regulator inquiries and external audits. Programs that draft the memo well in Stage 2 already have the regulatory artefact in hand; programs that skip it spend two to three times more retrofitting under timeline pressure later. Treat the memo as both an investment ask and a governance artefact from the first draft.

Capability prioritisation is the conversation that decides which of the twenty plausible agentic-AI initiatives makes it into the twelve-month roadmap. The lightest framework that works in this context is a two-by-two matrix on impact and feasibility, with each candidate placed in one of the four quadrants. The matrix is not a precise instrument — it is a discussion structure that forces the working group to disagree explicitly rather than implicitly.

Impact is scored on a three-point scale: meaningful business outcome (3), useful but contained (2), interesting but not yet tied to outcome (1). Feasibility is scored the same way: ship-in-one-quarter (3), ship-in-two-quarters (2), ship-in-three-plus-quarters (1). Anything that scores 3×3 is roadmap-required. Anything below 2×2 is roadmap-excluded for this year. The contested band is everything in between.

One discipline that separates working groups using the matrix well from those using it as theatre: the placement of each candidate is recorded with the names of the people who agreed to it and the date. Re-baseline at the end of each quarter — if an initiative moved quadrants because feasibility changed or impact resolved, that movement is itself an item to discuss in the steering committee, not a silent edit.

Investment phasing is the mechanism that lets a steering committee commit capital without committing to a bet-the-farm decision. The standard agentic-AI phasing model is three tiers: foundation, scale, optimise. Phase 1 covers Q1 and most of Q2 and is the largest capital allocation in absolute terms — it pays for the infrastructure, the governance setup, and the first enablement cohort. Phase 2 covers Q2 ramp through Q3 and pays for the second customer-facing surface and the multi-agent composition work. Phase 3 covers Q3 ramp through Q4 and pays for the optimisation and audit cycle.

Each phase has a gate before the next phase opens. The Phase 1 to Phase 2 gate is the binary-pass on Q1 milestones plus the Stage 1 audit re-scoring at end of Q1. The Phase 2 to Phase 3 gate is the binary-pass on Q2 milestones plus the Stage 1 audit re-scoring at end of Q2. If a gate fails, the steering committee re-decides the next phase rather than releasing it automatically. The gate is what makes the phasing model defensible.

A few notes on common variants. Programs that start from a higher Stage 1 maturity score — Proactive or above — can compress Phase 1 and re-allocate capital toward Phase 2 and Phase 3. Programs that start from Ad-Hoc or Reactive should expect Phase 1 to consume closer to fifty-five percent of the twelve-month budget and should resist any pressure to compress it. The gate discipline matters more than the headline split: if Phase 1 milestones do not pass, Phase 2 does not open, regardless of how the percentages were drafted.

For working groups that want a deeper view of the surrounding stack, the broader Digital Applied AI transformation service catalogue describes how Stage 2 phasing connects to the downstream stages — vendor selection, prototype, production deploy — and the typical engagement shape we use to support mid-market and enterprise programs through the pipeline.

The communication cadence is what keeps the roadmap a live artefact rather than a quarterly memorial service. Three rhythms run in parallel: the working group meets weekly, the steering committee meets monthly, and the board reviews the program quarterly. Each rhythm has a fixed agenda, a fixed artefact, and a fixed length — drift in any of those is the most reliable leading indicator of program drift overall.

The weekly working-group update is the simplest of the three: a single page with the OKR scorecard (traffic-light), the top three risks for the week, the top three decisions needed from outside the group, and a one-line status from each track owner. Distribute it in writing by Thursday so the working-group meeting can be a thirty-minute discussion of the items in red and amber rather than a status-readout.

# Stakeholder communication cadence · Stage 2

CADENCE      | AUDIENCE             | ARTEFACT                       | LENGTH
-------------|----------------------|--------------------------------|--------
Weekly       | Working group        | OKR scorecard (1 page)         | 30 min
Bi-weekly    | Sponsor + working    | Sponsor brief (½ page)         | 15 min
             | group leads          | + 2 decisions needed           |
Monthly      | Steering committee   | OKR scorecard + roadmap delta  | 45 min
             |                      | + Phase gate status            |
Quarterly    | Board                | Two-page exec memo (refreshed) | 30 min
             |                      | + investment ask if any        |
Quarterly    | All-hands (eng + bz) | One-page program update        | 15 min
             |                      | + Q&A                          |

One observation from running this cadence across engagements: the artefact that most often gets dropped under pressure is the sponsor brief. It is the lowest-status of the five, but it is the artefact that keeps the executive sponsor aligned with the working group between steering committees. Sponsor briefs that stop landing are a leading indicator that the sponsor is disengaging — react quickly when they slip.

The all-hands update is the artefact most working groups under-invest in. Engineering and business teams outside the working group form their view of the program from rumour otherwise, and rumour is rarely flattering. A one-page quarterly update with a refreshed OKR scorecard and a Q&A line keeps the broader org aligned and surfaces concerns early.

Stage 2 closes by handing two artefacts to Stage 3 — the twelve-month roadmap and the Q1 OKR set. Stage 3 owns the data foundation work: source inventory, lineage, classification, retention, training-data provenance, evaluation datasets, ground-truth labels, drift monitoring. Stage 3 is the most under-budgeted stage in most programs, and the cleanest hand-off from Stage 2 is the strongest defence against that under-investment.

The hand-off itself is a single working-group session — typically ninety minutes — that walks the Stage 3 owner through the Q1 and Q2 milestones that depend on data work, the governance artefacts that Stage 3 must satisfy, the budget envelope inside Phase 1 and Phase 2 that Stage 3 can spend against, and the Stage 1 readiness findings on the data domain that Stage 3 must close. Document the session, archive it, and proceed.

For working groups now ready to begin the next stage, the Stage 3 playbook is the immediate next read: Agentic AI Data Foundation Templates — Stage 3 covers the source inventory, RAG-readiness assessment, lineage and provenance, evaluation datasets, and PII-handling templates. For working groups still finalising Stage 1, the readiness-audit templates are at Stage 1 — Readiness assessment.