Most people evaluating AI consulting proposals are doing it for the first time. The AI consulting market has no standard taxonomy — firms range from slide-deck strategists to engineers who deploy production systems handling hundreds of decisions per day. They all use similar language. They all claim to deliver ROI. And most buyers cannot tell the difference from a proposal document alone.

This matters because the failure rate is not a technology problem. RAND's analysis of more than 2,400 enterprise AI initiatives found that 84% of failures were caused by leadership and process issues — not the models, not the infrastructure, not the data quality alone. Which means the firm you choose and the contract you sign are more predictive of your outcome than the technology stack they propose to use. The proposal evaluation you are doing right now is upstream of everything else.

80%+
of enterprise AI projects fail to deliver intended business value
RAND Corporation, 2024
42%
of companies abandoned most AI initiatives in 2025 — up from 17% the prior year
S&P Global, 2025
84%
of AI failures caused by leadership and process issues, not technology
VentureBeat, 2024
5%
of AI pilot programmes achieve rapid revenue acceleration (MIT Project NANDA, 2025)
MIT NANDA, 2025

Step 1 — Classify each proposal before comparing any criteria

You cannot compare a strategy proposal to an implementation proposal on the same scorecard. They are different products wearing similar packaging. The first step in any AI consultant proposal evaluation is to identify which type each proposal represents — and confirm that you are receiving the type you actually need.

Type A
Strategy Consultant
Delivers
  • Use case audit and prioritisation
  • AI readiness assessment
  • Vendor comparison and selection support
  • Implementation roadmap document
  • Business case and ROI modelling
Evaluate on: discovery methodology depth, use case prioritisation logic, credential relevance to your industry. Engagement ends when the document is delivered.
Type B
Implementation Consultant
Delivers
  • Working AI workflows in production
  • Integrated pipelines with your systems
  • Deployment and monitoring setup
  • Team training and handoff documentation
  • Acceptance-tested, measurable outcomes
Evaluate on: production portfolio evidence, milestone payment structure, named team technical credentials, specific acceptance criteria. Engagement ends when the system generates promised outcomes.
Type C
Hybrid
Claims to deliver
  • Strategy and implementation in one engagement
  • Discovery phase → roadmap → build → deploy
  • Most common format for mid-market companies
  • Widest quality range of the three types
  • Risk: strategy-only consultants who inflate scope
Evaluate on both criteria sets. Confirm the team has both strategic and engineering credentials — not just consultants who have learned to mention "implementation." Ask for separate portfolio evidence for each phase.

"The title 'AI consultant' covers everything from someone who builds PowerPoint decks about AI strategy to someone who deploys production systems handling 200+ calls per day. These are fundamentally different jobs wearing the same label."

Saksham Solanki — How to Hire an AI Consultant: A Practitioner's Guide, April 2026

Step 2 — Read these four sections first in every proposal

A 40-page AI consulting proposal contains four sections that carry the most signal about actual delivery capability. Read these four before anything else. They will tell you more in 15 minutes than the rest of the document will in an hour.

Section A Technical approach and proposed solution
This section reveals whether the proposal was written about your company or copied from a template. A genuinely customised technical section names your current tools, references the workflow audit findings, and explains why the proposed approach fits your specific data environment and constraints. A template section describes how "AI can transform operations in your industry" without naming a single system you told them about.
✓ Passing language
"Given your existing Salesforce CRM and manual quoting workflow described in discovery, we propose a retrieval-augmented generation layer that reads deal history and generates first-draft quotes, reducing that 4-hour process to under 20 minutes."
✕ Template language
"Leveraging our proven AI methodology, we will identify high-value automation opportunities across your operations and implement cutting-edge AI solutions aligned with industry best practices."
Section B Deliverables, timeline, and pricing structure
Deliverables should be named in specific formats with specific delivery dates. "Recommendations and insights" is not a deliverable. "A 12-workflow automation system deployed to your production Zapier/Make environment with documented runbooks for each workflow, delivered by Week 10" is a deliverable. The pricing structure is equally revealing: implementation work that is 100% time-and-materials with no milestone accountability creates a perverse incentive — the longer delivery takes, the more the consultant earns. Fixed-fee or milestone-based implementation pricing signals confidence in delivery.
✓ Credible deliverable
"Deliverable 3: Customer inquiry triage automation — live in your Zendesk instance, routing 80%+ of common inquiries to correct queue without manual intervention. Delivery: Week 8. Payment gate: 25% of project fee on client acceptance testing."
✕ Vague deliverable
"Phase 2 deliverables include AI implementation support, workflow optimisation guidance, and recommendations for maximising AI ROI across your customer service operations."
Section C Proposed team and credentials
Named individuals with verifiable credentials are the single most important quality signal in an implementation proposal. "Our experienced AI team" without names cannot be due-diligenced. A named senior engineer with a LinkedIn profile showing specific production AI deployments can be. The bait-and-switch pattern — senior engineers shown in the sales process, junior engineers assigned to the project — is caught at the proposal stage by requiring named team members and confirmation that those individuals will work on your project specifically. Ask the question directly in your follow-up: "Are the individuals named in this proposal the ones who will actually execute the work?"
✓ Named team with evidence
"Lead: [Name], AI Systems Architect, 7 years production AI deployment. Previously: built the customer intent classifier at [Company] handling 50,000 daily queries. LinkedIn: [link]. Day-to-day contact and weekly call lead."
✕ Unnamed team
"Our delivery team consists of experienced AI engineers and project managers who will be assigned based on project requirements and current availability."
Section D Success metrics and measurement plan
This section reveals whether the consultant plans to be accountable for outcomes or just for activity. Specific, measurable success metrics with a pre-deployment baseline and a target tied to a delivery date are the sign of a consultant willing to be measured. Vague metrics like "improved efficiency" and "increased AI adoption" are the sign of a consultant who wants to be measured only on effort. The pre-deployment baseline is particularly important: without it, there is no way to determine whether an outcome was achieved. A proposal that does not specify how the baseline will be measured before project start cannot demonstrate improvement after project end.
✓ Specific outcome metric
"Baseline (Week 0): average time to process a customer quote request = 4.2 hours, measured across 200 historical tickets. Target (Week 12): 45 minutes or less for 85%+ of requests. Measurement: Zendesk ticket resolution time report, automated weekly export."
✕ Vague outcome claim
"Upon project completion, your team will experience significantly improved operational efficiency, reduced manual workload, and accelerated AI adoption across key business functions."
Skip the proposal lottery

Find AI consultants already evaluated on these criteria

TechRadiant verifies AI consultants and agencies on production portfolio depth, deliverable specificity, and documented outcome track records — the criteria this evaluation framework defines. Verified consultants in our index have already been assessed so you start with a shortlist rather than a lottery.

Share your AI brief → Browse verified AI consultants

Step 3 — Apply the disqualifiers before you score

Six conditions disqualify an AI consulting proposal before scoring begins. These are not minor concerns to be noted in an evaluation — they are structural problems that predict delivery failure at rates high enough to justify elimination without further assessment. If a proposal triggers any of these, remove it from the comparison before applying the scoring matrix.

Disqualifying conditions — check before scoring
  • 1
    Technology recommendations before workflow audit
    A proposal that names specific AI tools or platforms without first demonstrating understanding of your workflows is solving an interesting problem, not your problem. The most reliable red flag in AI consulting is a provider who presents tools before auditing workflows (AI Smart Ventures, May 2026). A consultant who recommended a specific LLM in the first conversation — before reviewing what your team actually does and where time is spent — is not conducting a consulting engagement. They are selling a product.
  • 2
    Deliverables described as "recommendations," "insights," or "strategic guidance"
    These are descriptions of advisory conversations, not named deliverables. Research across close to 1,000 organisations shows that buyers who do not verify a named deliverable before signing consistently report that the engagement produced no usable output (AI Smart Ventures, April 2026). A credible proposal names the primary deliverable before the price and specifies the format of each output.
  • 3
    100% time-and-materials pricing for implementation work with no milestone accountability
    Advisory work is appropriately hourly. Implementation work is not. A consultant who prices implementation hourly has a financial incentive to take longer — and no contractual accountability for outcomes. Senior AI practitioners almost universally charge fixed-fee for implementation because their value is in the speed and quality of shipping, not the hours logged (Lilach Bullock, July 2026). T&M for implementation signals either low confidence in delivery or misaligned incentives.
  • 4
    No named team members with verifiable production AI experience
    "Our experienced AI team" cannot be due-diligenced. A named senior engineer with a verifiable LinkedIn history of production AI deployments can be. If the proposal does not name the individuals who will work on your project, you cannot verify that those individuals exist, have relevant experience, or will be assigned to your project specifically. This is the most common mechanism for the bait-and-switch: senior engineers in the sales process, junior developers in delivery.
  • 5
    Discovery phase longer than 4 weeks with no defined deliverable
    Discovery is legitimate — 2–4 weeks of assessment to define requirements, evaluate data quality, and identify risks. When discovery stretches to 8–12 weeks with no concrete output, the firm is billing to learn about your business rather than applying expertise they should already have (Groovyweb, July 2026). A discovery phase should produce a named document: a scoped requirements document, a data readiness assessment, or a prioritised use case list — not a transition into "Phase 2 planning."
  • 6
    Custom model training recommended as the first step for a standard use case
    If the first recommendation is "we need to train a custom model," the consultant is solving the interesting problem, not the right problem. Most business AI use cases are better served by existing models with the right system design around them than by custom training pipelines (Saksham Solanki, April 2026). Custom model training adds significant cost, timeline, and ongoing maintenance burden. A consultant who recommends it for a standard document processing, classification, or generation use case is optimising for their own technical interest, not your ROI.

Step 4 — The side-by-side comparison table

Replace "Proposal A" and "Proposal B" with your consultants' names. Apply the same criteria to both. The comparison is most useful when done in writing — verbal impressions from discovery calls are not reliable for distinguishing polished presenters from strong deliverers.

Evaluation criterion
Proposal A
Proposal B
Proposal classification
Type (Strategy / Implementation / Hybrid)
[Strategy / Implementation / Hybrid]

Confirm before scoring — mismatched types produce invalid comparisons
[Strategy / Implementation / Hybrid]

If types differ, re-evaluate whether you are comparing the right thing
Deliverables and scope
Primary deliverable named with format
✓ Pass example: "12 deployed automation workflows with runbooks, delivered to production by Week 10"
✕ Fail example: "AI strategy and implementation support across key business functions"
References your specific workflows
✓ Pass example: Proposal names your CRM, your current process times, your team structure from discovery
✕ Fail example: Proposal could be addressed to any company in your industry — no specifics from your discovery call
Success metrics with baseline
✓ Pass example: "Baseline: 4.2 hours per quote request. Target: 45 minutes by Week 12, verified via Zendesk report"
✕ Fail example: "Significant improvement in operational efficiency and team productivity"
Data preparation scoped
✓ Pass example: Includes data audit, quality assessment, and pipeline setup as explicit line items — allocates 40%+ of timeline to data work
✕ Fail example: No mention of data quality or preparation — assumes data is ready without auditing it
Pricing and accountability
Pricing structure for implementation
✓ Pass: Fixed-fee or milestone-based with named payment gates tied to acceptance criteria
✕ Fail: 100% time-and-materials for implementation with no milestone accountability or outcome gates
Milestone payment gate language
✓ Pass: "25% due on client acceptance of Phase 2 deliverables, defined as: [specific criteria]"
✕ Fail: "Monthly retainer of $X — engagement continues until project is complete"
What happens if outcomes aren't met
✓ Pass: Specific remediation clause — additional work at no charge, or defined exit rights for underperformance
✕ Fail: No mention of underperformance; "we'll work to address any concerns that arise"
Team and credentials
Named individuals with verifiable credentials
✓ Pass: Named lead engineer with LinkedIn link and specific production deployment history
✕ Fail: "Our experienced AI team will be assigned based on project requirements"
Production portfolio evidence
✓ Pass: Named case study with specific outcome metrics — "reduced processing time from X to Y at [company type]"
✕ Fail: Generic industry claims — "we have helped hundreds of companies across sectors transform with AI"
Verifiable references provided
✓ Pass: Direct client contact provided — name, role, direct email, engaged within last 18 months
✕ Fail: Testimonials in the proposal or "references available on request" — not yet provided
Capability transfer and post-engagement
Internal capability transfer plan
✓ Pass: Explicit handoff plan — documentation, training sessions, run-book ownership transfer to named internal contact
✕ Fail: No mention of what happens after delivery — ongoing retainer implied but not scoped
Exit clause and contract terms
✓ Pass: 30-day written notice clause; IP ownership explicitly transferred to client on final payment; no lock-in to proprietary tooling
✕ Fail: 90-day minimum commitment; IP terms absent or vague; solution built on vendor-proprietary platform

Step 5 — The weighted scoring matrix

Score each criterion 1–5 for each proposal (1 = absent or disqualifying, 5 = clear and specific). Multiply the score by the weight to get the weighted score. Total the weighted scores to get the final comparison number. A score above 80 on 100 indicates a credible proposal; below 60, walk away or request significant revision before proceeding.

Criterion
Weight
Score A (×wt)
Score B (×wt)
1. Deliverable specificity
Named outputs with formats and delivery dates. The primary test of accountability.
25%
2. Proposal customisation
Technical section describes your environment specifically. Template language scores 1; full customisation scores 5.
15%
3. Team credentials
Named individuals with verifiable production AI experience. "Experienced team" without names scores 1.
15%
4. Pricing and accountability structure
Milestone-based with acceptance criteria. 100% T&M implementation pricing scores 1.
15%
5. Success metrics quality
Specific, measurable outcomes with baseline and target. Vague efficiency claims score 1.
15%
6. Discovery methodology
Structured workflow audit before solution design. "We'll discover requirements as we go" scores 1.
5%
7. Data and compliance approach
Addresses your specific data environment, quality requirements, and regulatory obligations. Generic data statements score 2.
5%
8. Capability transfer plan
Explicit handoff documentation, training, and internal ownership plan. Implied ongoing retainer with no transfer plan scores 1.
5%
Weighted total (max 100)
100%
/100
/100

2026 pricing reference — what you should expect to pay

Before evaluating pricing in any proposal, establish whether the quoted number is reasonable for the scope and firm type. The ranges below are 2026 market rates for established consultants and agencies. Outliers at either end warrant investigation — unusually low rates suggest junior staffing or offshore-but-presented-as-onshore; unusually high rates from non-Big-4 firms without clear justification suggest premium positioning without premium delivery.

Firm type Typical rate / scope Engagement structure Watch for
Solo AI specialist $80–$200/hr · $15K–$50K fixed projects Fixed-fee preferred; hourly for advisory Capacity risk — one person, no redundancy. Ask what happens if they are unavailable.
Boutique AI consultancy $150–$300/hr · $25K–$150K projects Milestone-based preferred; monthly retainer for ongoing advisory ($5K–$25K/mo) Named vs assigned team — confirm the people quoted are the people delivering
AI-first implementation agency $50K–$300K+ project engagements Fixed-fee milestone projects; dedicated team retainer for scale Offshore teams presented as onshore — ask for team location explicitly
Big 4 / Strategy firm $300–$600/hr · $100K–$500K+ engagements T&M for strategy; fixed-fee rare; junior-heavy staffing with senior partner oversight Blended rates hiding junior staffing; strategy deliverables priced as implementation work
Offshore AI agencies $22–$50/hr · comparable project scope at 30–60% lower cost Fixed project or dedicated team; milestone-based Apply the same evaluation criteria; rate advantage disappears if ramp-up and quality gaps inflate hours

For the broader AI agency and consultant evaluation framework — including the specific questions that surface whether a consultant builds for your long-term capability or for their ongoing dependency — see our AI agency briefing guide and our AI consultants vs in-house teams cost comparison.