AI Opportunity Assessment for Enterprise: 2026 Playbook

If you're in Group Strategy, the CFO office, or the Office of the CEO at a 500–5,000 employee company drafting Phase I scope for an internal AI mandate, you are almost certainly about to issue an RFP whose first line reads some variant of "AI opportunity assessment: potentials & feasibility." That phrase is doing a lot of work. It signals to the market that you want a structured catalogue of where AI can move the needle — not another strategy deck, and not a vendor pitch. This guide explains what an AI opportunity assessment actually is at enterprise scale, what should be in scope, how the deliverable should look, what it costs in 2026, and the procurement traps that quietly waste seven figures.

What an AI Opportunity Assessment Actually Delivers

An AI opportunity assessment is a time-bound enterprise engagement whose deliverable is a prioritised opportunity catalogue — every candidate AI/automation use case across the in-scope business units, scored on effort, impact, dependencies, business-calendar fit, and time-to-value, and each linked to the source evidence that produced it. Output is not a slide deck. Output is an artefact a CFO can fund against and a Group COO can govern against.

The category overlaps with several adjacent labels, which is why procurement teams keep getting sold the wrong thing. Distinctions worth getting on the record before you send the RFP:

• AI strategy — board-facing thesis on whether to invest, how much, against what hypothesis. Output: an investment narrative.
• AI roadmap — sequencing of a pre-selected set of initiatives across 12–36 months. Output: a Gantt chart.
• AI opportunity assessment — structured discovery + scoring of the full opportunity surface across BUs and functions. Output: a ranked catalogue (typically 100–300 candidates) with effort/impact/dependency scoring and source attribution.
• AI use case discovery — the discovery component of the assessment. Sometimes sold as a standalone, almost always under-scoped without the scoring layer.
• AI opportunity mapping — the visualisation layer (heatmaps, dependency graphs, cluster views) on top of the catalogue.

A serious AI opportunity assessment contains use case discovery and opportunity mapping as components. If a vendor pitches one of those three labels in isolation, you are almost certainly looking at a partial scope dressed in enterprise language. Confirm what is in and out of scope in writing before signing.

When to Run One (and When Not To)

The assessment is the right Phase I instrument when the organisation has decided AI is a priority but does not yet have a defensible answer to "where, in what order, for what return, at what risk." It is the wrong instrument when those answers already exist and the actual blocker is execution capacity.

Trigger signals (run one)

• Post-IPO efficiency mandate. The S-1 promised operating leverage. The board now wants a programmatic answer to "where are we capturing it?" An AI opportunity assessment is the standard Phase I scoping instrument.
• Direct board question on AI. The Q1 board meeting asked the CEO for "the AI plan." A 30-deck answer ages out in six months. A persistent opportunity catalogue is the durable response.
• Vendor & tool consolidation pressure. The organisation now runs 14–22 AI-adjacent tools across BUs with overlapping capabilities. An assessment surfaces the consolidation opportunities alongside the build opportunities.
• M&A integration debt. Three acquisitions in 24 months. Each subsidiary brought its own AI footprint, its own data assets, its own shadow stack. The assessment is what produces the cross-entity unification view.
• EU AI Act / governance mandate. Compliance is forcing inventory. Once you have to inventory the AI surface anyway, scoring it for opportunity costs 20% more and produces 10x the value.

Not-trigger signals (do not run one yet)

• You already have a prioritised backlog. If a recent engagement — internal or external — produced one, what you need is implementation capacity, not another assessment.
• The blocker is a single decision, not a portfolio. "Should we buy or build the contact-centre AI stack" is a vendor evaluation. Don't dress it up as an assessment.
• No executive sponsor with cross-BU mandate. An assessment whose recommendations cannot cross BU lines will surface opportunities nobody can authorise. Fix the sponsorship first.
• You're 30 days from an integration freeze, peak season, or month-end close. Discovery during a freeze produces partial data and resentful interviewees. Wait six weeks.

The 4 Inputs Every Assessment Needs

Before issuing the RFP, assemble these four inputs internally. Any partner worth hiring will request them in the kickoff. Having them ready compresses the engagement by 1–2 weeks and surfaces scoping gaps before they become change orders.

Input	What it contains	Owner	Why it matters
Workflow inventory	List of recurring workflows per BU and central function, with cadence (daily/weekly/monthly), headcount touched, primary system of record	COO / Group HR Ops	Defines the denominator the assessment will score against
Current vendor map	All AI, automation, RPA, analytics, and SaaS vendors in use per BU, with contract value and renewal date	CIO / Procurement	Surfaces consolidation opportunities and prevents recommending what's already deployed
Business calendar	Month-end close, peak season, integration milestones, regulatory filings, board cadence for next 12 months	CFO Office / Group PMO	Critical input to time-to-value scoring; an opportunity that collides with peak season is not deployable
Governance constraints	EU AI Act classification, data residency rules, model risk policy, existing risk steering committees, acceptable-use posture	Group Legal / Risk / Trust & Safety	Filters out opportunities that are technically attractive but blocked by policy

The most common procurement failure mode is treating these inputs as a Phase I deliverable rather than a Phase I prerequisite. If you let the vendor build them for you, you've paid consulting day rates for what your COO and CIO already know.

Three-Channel Discovery: Interviews + AI Surveys + Passive Telemetry

The single decision that most determines the quality of an AI opportunity assessment is the discovery method. Talking to executives produces an executive-shaped view of the organisation. Talking to executives and the people doing the workflows and reading the telemetry produces a workflow-shaped view. The first method is what tier-1 firms typically scope. The second is what a serious 2026 AI use case discovery exercise looks like.

Channel 1 — Deep interviews

• Format: 60–90 minute semi-structured interviews, 30–50 roles total: ELT, BU MDs, functional heads, customer care leads, operations managers, data engineering leads, product owners per BU.
• Output: coded transcripts mapped to an opportunity ontology — not narrative quotes. Every coded insight is traceable to an interview ID.
• What it captures: strategic intent, cross-BU dependencies, political constraints, recent failed initiatives the deck won't mention.
• What it misses: the line-level workflows where most of the ROI lives. The 850 employees not in the interview list.

Channel 2 — AI-driven asynchronous surveys

• Format: conversational, 5–7 minute async prompts delivered in the respondent's working language. Adaptive follow-ups based on first answers. Not multiple choice. Not Likert scales.
• Output: structured workflow-level data from the full workforce, clustered and themed automatically, every cluster traceable to source responses.
• What it captures: the workflows leadership doesn't see — the reconciliation Excel that runs every month-end, the customer escalation pattern in one country, the agent onboarding flow nobody owns.
• What it misses: strategic context. Surveys without interviews produce a tactical opportunity list with no narrative.

Channel 3 — Passive telemetry

• Format: read-only signals from systems already in use — calendar density, document repetitive-open patterns, email response variance, support ticket clustering, ITSM ticket categories. No new tools deployed. No invasive monitoring.
• Output: objective signals that confirm or contradict what interviews and surveys reported.
• What it captures: the gap between what the organisation says it does and what it actually does. Workflows that consume disproportionate time but nobody complains about because they've been normalised.
• What it misses: intent. Telemetry without interviews produces a pattern map with no "so what."

The three channels are designed to triangulate. Each channel alone produces a biased view. Cross-validated, they produce an opportunity catalogue where every entry has multi-source evidence and the procurement team can defend the prioritisation against any internal challenge.

How to Score and Prioritize the Opportunities

Discovery without scoring is a list. Scoring is what turns a list into an executable backlog. A defensible scoring matrix for an AI opportunity assessment runs along five vectors. Each opportunity gets a score, the scores roll up to a composite, and the composite is filterable so the CFO, COO, and each BU MD can produce the view they need from the same underlying data.

• Effort — euros, FTE-weeks, vendor dependencies, data-readiness lift. Score 1–5.
• Impact — revenue gained, cost avoided, risk reduced, NPS/CSAT improvement. Score 1–5.
• Dependencies — how many other initiatives must ship first. Score 1–5 (lower is better).
• Business calendar fit — does deployment collide with peak season, month-end, integration freeze? Score 1–5.
• Time-to-value — weeks until first measurable outcome. Score 1–5.

The composite is a weighted average; weights are set by the executive sponsor before scoring begins so the prioritisation cannot be reverse-engineered to favour any one BU. Below is a concrete worked example from a mid-cap multi-BU enterprise.

Opportunity	BU	Effort	Impact	Deps	Cal. fit	TTV	Composite
Tier-1 support deflection (LLM triage)	Customer Care	2	5	2	4	5	4.4
Month-end reconciliation co-pilot	Finance	2	4	1	5	4	4.2
Agent onboarding generation	Sales Ops	3	4	2	4	4	3.8
Dealer dashboard insight layer	Distribution	3	4	3	3	3	3.4
RFP response generator	Sales	2	3	1	5	5	3.8
Contract review automation	Legal	3	4	2	4	3	3.4
Predictive churn scoring	CS	4	5	4	3	2	2.8
Pricing engine v2 (ML)	Revenue Mgmt	5	5	5	2	1	2.4
Knowledge-base unification	Group IT	3	3	2	5	3	3.4
HR ticket auto-resolution	Group HR	2	3	1	5	5	3.8

What the worked example illustrates: the highest-impact opportunity (predictive churn scoring) does not necessarily top the composite, because dependency burden and time-to-value drag it down. The opportunities that actually deserve Q1 capital are the ones with high impact, low effort, and clean calendars — not the most technically ambitious. A scoring discipline is what surfaces this; a deck almost never does.

What the Deliverable Looks Like

The most important question to put on the RFP: "What, physically, do we have in our hands at the end of the engagement?" The credible 2026 answer is a persistent intelligence layer — not a final report. The deliverable behaves more like Google Maps for the AI surface area than like a McKinsey binder.

What it contains

• Heatmaps of opportunity density by business unit, function, and theme.
• A knowledge graph of workflow dependencies across BUs (where one vertical's data asset is upstream of another's reporting cadence; where one acquisition's tooling can be lifted into the parent).
• A filterable catalogue of 100–300 ranked opportunities, each scored on the five vectors above and tagged by BU, function, theme, vendor footprint, governance class.
• Full source attribution — every opportunity clickable through to the specific interview ID, survey cluster, telemetry pattern, and document section that produced it. Auditable. Defensible to a board challenge.
• An AI opportunity mapping view — visual cluster diagrams that let leadership tell the strategic story without reverse-engineering it from a spreadsheet.

How different roles use it

• CEO opens the macro heatmap and identifies the 3–5 strategic clusters that warrant a board-level narrative.
• CFO filters by "savings >€500K under 6 months, effort <€500K" and builds the FY27 funding submission.
• BU Managing Directors drill into their vertical, see cross-BU dependencies before approving their own roadmap, and avoid building what a sister BU already deployed.
• Group Risk / Trust & Safety filters by governance class and produces the AI Act conformity view in one click.

The structural reason the intelligence layer is the deliverable: it is persistent. Your teams update it after the engagement ends. New opportunities surface from operations, get scored against the same ontology, and feed the same governance cadence. A deck is dead from day one. An intelligence layer compounds.

Timeline & Team Setup

Phase I of an AI opportunity assessment at enterprise scale runs 8 weeks, with a structural commercial gate at day 30. The gate is what separates an assessment engagement from open-ended advisory work.

Days 1–30 — Pilot cluster

• Week 1: kickoff, scoping confirmation, input ingestion (workflow inventory, vendor map, calendar, governance constraints), interview roster lock-in.
• Weeks 2–3: three-channel discovery on a single chosen cluster (one vertical or one shared-function group). Interviews run in parallel with async surveys and telemetry collection.
• Week 4: first intelligence layer instantiation. Pilot catalogue of 30–60 scored opportunities. Operator validation of the top 10.

Day 30 — Go/no-go gate

The single most important contractual clause in the SOW. Before signing, agree in writing: minimum number of validated opportunities, minimum aggregate addressable saving, minimum percentage backed by both interview and telemetry evidence. If the day-30 proof falls short of agreed criteria, the engagement stops and there is no fee for the remaining scope. Without explicit criteria the gate is theatre; with them it is the most useful procurement protection available.

Days 31–56 — Scope expansion

• Weeks 5–6: three-channel discovery extended to remaining in-scope BUs and central functions.
• Week 7: full catalogue scored. Cross-BU dependency graph built. Vendor consolidation opportunities surfaced.
• Week 8: intelligence layer handover. Executive read-out. Governance design for the catalogue's ongoing life.

Team setup (yours, theirs)

• Executive sponsor (your side): CFO or COO. Not the CIO. Cross-BU authority required.
• Operating partner (your side): Head of Data & Engineering, or Group Director Transformation. Day-to-day counterparty.
• BU liaisons (your side): one per in-scope BU, ~10% time over the 8 weeks. Their job is to unlock access, not to do the discovery work.
• Engagement lead (partner side): senior operator who has run the workflow type, not a consultant who read about it.
• Discovery team (partner side): 2–3 interviewers, 1 telemetry analyst, 1 survey-design lead, 1 catalogue engineer.

Cost Reference (2026 Enterprise Rates)

Pricing for an AI opportunity assessment at 5–15 BU scope, two working languages, 8-week Phase I duration varies by roughly an order of magnitude depending on partner archetype. The table below is the reference grid we use in scoping calls.

Partner archetype	Phase I fee (5–15 BU)	Deliverable	Day-30 gate?
Tier-1 strategy firm	€800K–€2M+	Top-down assessment, ~200-page deck, ~30 use cases, board-grade narrative	Rare
Big-4 advisory	€400K–€1.2M	Hybrid assessment with vendor recommendations, often bundled with implementation arm	Sometimes
Operator-led assessment firm	€150K–€450K	Bottom-up three-channel discovery, persistent intelligence layer, 100–300 ranked opportunities, source-attributed	Yes (structural)
Build in-house (Chief AI Officer + 3 FTE)	€0 cash + 9-month time cost + €800K–€1.2M annual run-rate	Whatever the team produces; depends entirely on hiring quality and tooling	N/A
Single freelance "AI consultant"	€30K–€120K	Coverage limited to 1–2 BUs; not an enterprise-scale assessment	N/A

The honest read: tier-1 firms win where the board wants a brand-name cover sheet on the deliverable. Operator-led firms win where the buyer wants an artefact that survives Phase II and into Phase III execution. Building in-house is the right call only if your hiring market is favourable; if it isn't, you spend nine months recruiting and have nothing in the CFO's hands at the next board meeting. A single freelance consultant is the right answer for a single-BU pilot, not for an enterprise AI opportunity assessment.

Common Pitfalls

Five traps recur across enterprise AI opportunity assessment engagements. Each one is avoidable if the RFP and SOW are scoped against it explicitly.

1. Over-reliance on executive interviews

Talking to 50 executives produces a 50-executive-shaped opportunity map. The other 850–4,950 employees are the ones running the workflows where most of the ROI lives. Demand a discovery method that reaches beyond the leadership layer — in writing, in the SOW.

2. Single-method discovery

Interviews alone produce strategy-shaped insight with no operational grounding. Surveys alone produce a tactical list with no narrative. Telemetry alone produces a pattern map with no intent. The serious version of AI use case discovery uses all three in parallel and cross-validates. If a vendor pitches "we'll do 40 interviews," ask what their plan is for the other two channels.

3. Deck-as-deliverable

If the deliverable is a deck, the engagement is dead the day it ends. A 200-slide PDF cannot be filtered, cannot be updated, cannot be drilled into for source evidence, and cannot answer a CFO question that wasn't anticipated by the consultants. Insist on an intelligence layer artefact your teams own and can update.

4. No day-30 go/no-go gate

An AI opportunity assessment without a structural commercial gate is an open-ended advisory engagement with a structured-discovery cover story. The gate is the only mechanism that genuinely de-risks the buyer. If the partner refuses to put one in the SOW, that is the answer to the question "is this engagement scoped against outcomes or against time-and-materials?"

5. Missing operator validation

An opportunity is not validated because a consultant said it was. It is validated because someone who has actually run that workflow at enterprise scale stress-tested the assumption. The discovery team must include senior operators, not just analysts. Ask in the RFP for the named operator who will validate the top opportunities and what they have shipped before.

SUPALABS First-Party Data

FAQ

What is the difference between an AI opportunity assessment and an AI strategy engagement?

An AI strategy engagement decides whether to invest in AI, how much, and against what investment thesis — usually a board-facing artefact. An AI opportunity assessment sits one layer below: it surfaces, scores, and prioritises the specific opportunities across every in-scope BU and function, and produces a ranked catalogue the CFO can fund against. Strategy outputs a thesis. An assessment outputs a backlog with effort/impact/dependency scoring and source attribution. If you have already had a strategy engagement and nothing scaled, the missing layer is almost always the assessment — not another strategy refresh.

How long does an enterprise AI opportunity assessment take?

Phase I runs 8 weeks at typical 5–15 BU enterprise scope: days 1–30 are a deep-dive pilot on one chosen cluster with a hard go/no-go gate at day 30, then days 31–56 extend the three-channel discovery to the remaining BUs and central functions. A subsequent implementation-planning phase typically adds 4–6 weeks if the gate is passed. If a partner is pitching a 6-month "assessment" with no gate in the first 30 days, you are being sold an open-ended advisory engagement dressed as an assessment. Reject that scope.

What should an AI opportunity assessment cost at enterprise scale?

At 5–15 BU scope, two working languages, 8-week Phase I: expect €150K–€450K from an operator-led assessment firm, €400K–€1.2M from a Big-4 advisory firm, and €800K–€2M+ from a tier-1 strategy firm. The differential reflects the structural choice: tier-1 firms staff with senior consultants and produce a board-grade deck; operator-led firms build a persistent intelligence layer and structurally stake the Phase II fee on day-30 proof. Both are defensible procurement choices depending on what the board needs from the cover sheet.

Who should sponsor the assessment inside the organisation?

The sponsor that produces the best outcomes is the CFO or COO, not the CIO or CTO. Cross-BU mandate, control of the budget, and authority over operating-model decisions are all required for the assessment to produce actionable recommendations. The CIO/CTO is a critical participant but a problematic sole sponsor — assessments sponsored only out of IT tend to drift toward tooling decisions rather than workflow redesign. The strongest setup is a CFO/COO executive sponsor, the existing Head of Data & Engineering as operating partner, and a clear escalation path to the CEO for cross-BU prioritisation conflicts.

How do we know the partner is doing AI opportunity mapping properly, not just listing use cases?

Three structural tests. First, is the discovery method genuinely three-channel — deep interviews plus async surveys reaching beyond the executive layer plus read-only telemetry? Second, is every opportunity in the catalogue traceable to the specific interview, survey cluster, or telemetry pattern that produced it — or are the recommendations composite industry benchmarks dressed up as bespoke insight? Third, does the deliverable contain visual AI opportunity mapping — heatmaps, dependency graphs, cluster views — that let executives see structural patterns, not just read a list? A serious assessment passes all three. A rebadged consulting engagement fails at least one.

Can we run an AI opportunity assessment ourselves, in-house?

Technically yes, structurally rarely. The internal-build path requires a Chief AI Officer plus 2–3 senior FTE, runs roughly €800K–€1.2M annual fully loaded, and consumes 6–9 months before producing a first usable catalogue — assuming you can hire the right people in a tight 2026 market. The case for building in-house is strongest when the organisation will run continuous assessments year after year and the team can be redeployed to implementation between cycles. The case against is that most enterprises need the first catalogue in the CFO's hands inside one quarter, and an external AI opportunity assessment is the fastest path to that artefact — with the option to operationalise the catalogue in-house once it exists.

Sources & References