AI value review

Know which of your AI workflows pay back, and which to fix, stop or start next.

I score each AI workflow in production on value, cost, adoption and risk, and rank planned ones by payoff, effort and data.

Book a call See your AI readiness score
  • Five AI workflows scored on value, cost, adoption and risk, each ending in a keep, fix or stop call

Companies I’ve worked with

Across product, analytics, and AI.

Who this is for

For leaders accountable for what their AI workflows return (business processes where AI does part of the work and your people check the results).

CEO

• The board will ask what your AI portfolio returns.
• You need the answer in revenue, margin and customer satisfaction, including what to stop funding.

CFO or board

• You approved AI budgets across several teams.
• You need a number per workflow that you can check line by line.

COO or head of transformation

• You run the rollouts across teams.
• You need to know which processes are adopted, and which teams quietly work around them.

CPO or head of product

• Not all of the AI in the company came through your roadmap.
• You need to know which overlap with what you already pay for, and what to keep, fix or stop.

Why it matters

AI spending is rising fast, and most companies still can’t tie it to earnings.

$2.52T

Worldwide AI spending forecast for 2026, up 44% year over year.

Source: Gartner, January 2026.

37%

of organisations can attribute any EBIT impact to AI, unchanged from last year.

Source: McKinsey, The state of AI in 2026, 1,719 respondents.

6%

attribute 5% or more of their EBIT to AI.

Source: McKinsey, The state of AI in 2026, 1,719 respondents.

25%

of AI initiatives have delivered the ROI expected of them. Only 16% have scaled enterprise-wide.

Source: IBM Institute for Business Value, CEO Study 2025, 2,000 CEOs.

Without a baseline, there’s nothing to compare a result against. Without an agreed definition of “working”, every claim is an opinion. Without adoption tracking, a stalled rollout stays hidden for months. For AI placed on the market or put into service from 9 December 2026, the EU’s revised Product Liability Directive treats software as a product, so missing evidence becomes a liability question too.

What you get

A decision for each AI workflow, backed by evidence, not a strategy deck.

A scored view of every workflow, with a call for each.

• Value, cost, adoption and risk, in your own numbers.
• Cost counts run cost, cost per case, and people redoing the AI’s work.
• Keep, fix or stop, with the stop criteria written down.

One workflow scored on value, cost, adoption and risk, ending in a keep, fix or stop call
Invoice processing agent scorecard showing unverified value, partial cost evidence, no usage log, and no audit trail

An honest measurement-gap map.

  • What can’t be measured yet, such as missing usage logs or audit trails, and what it takes to close each gap.

A scoped next step, only where one pays back.

• For each workflow worth fixing or starting: what an AI delivery sprint would change, its estimated payback and assumptions.
• The sprint’s first week fixes its price in writing.

Sales quote assistant marked fix, leading to an AI delivery sprint scoped from the evidence, with estimated payback and assumptions written down
Ranked shortlist of three workflows by payoff, effort and data readiness, with a baseline measured for the first

Nothing in production yet? A ranked shortlist.

  • The 2 or 3 processes most worth starting, ranked by payoff, effort and data readiness.
  • A baseline for the top one. Part of the portfolio review.

How it works

Four steps, from your list of AI workflows to a decision for each.

1. Define “working”

A short session per workflow ties “is it working” to one number that drives a P&L line, like quote-to-close rate.

2. Score against evidence

• Evidence comes from billing exports, usage logs and short interviews.
• Where evidence is missing, the gap is a finding.

3. Make the call

Keep, fix or stop for each workflow, with what I found and where the evidence runs out.

4. Scope the next step

Where a fix or a new workflow pays back, you get the sprint scope and estimated payback.

Why it works

Each value figure is traced to numbers your CFO can check, and compared with teams that don’t have the AI workflow yet.

Driver tree and counterfactual comparison linking quote-to-close improvement to annual revenue

A driver tree breaks a claim like “this saved money” into the numbers that add up to it. A counterfactual compares teams with and without the AI workflow over the same period, so the result comes with a confidence range, like ±3%.

Example: AI agent portfolio walkthrough

An illustrative portfolio of four AI workflows, not a client, scored the way the review works.

A sales quote assistant, a support triage agent, an invoice processing agent and marketing drafting, each scored on the same four columns.

Portfolio scorecard summary with four AI workflows, their value, cost and adoption, and a keep, fix or stop call for each

Support triage lifted first-contact resolution 18%, worth €95k a year against €9.6k run cost. The missing audit trail is still flagged.

Expanded support triage agent scorecard with value, cost, adoption, and risk evidence and a keep call

Quote-to-close rose 9% (±3%), but 32% of reps don’t use the assistant. The rollout stalled at ability, on ADKAR: how far each team has got, from knowing the AI workflow to using it by habit.

Expanded sales quote assistant scorecard showing a fix call, a 9 percent quote-to-close lift and 32 percent of reps not using it

The invoice agent’s claimed 60% time saving was never checked in eight months in production. Three of five processors re-check its work by hand, and a duplicate payment went through unreviewed.

Expanded invoice processing agent scorecard showing an unverified value claim, no usage log, an unreviewed duplicate payment and a stop call

Pricing

Prices are “from”: your fee is fixed in writing before I start, set by the workflows and systems in scope.

Single workflow

From €2,900

  • One AI workflow in production
  • Value, cost, adoption and risk, scored in your numbers
  • A keep, fix or stop call, with the evidence behind it
  • A list of measurement gaps and how to close them
  • 1 to 2 weeks
Book a call

Portfolio

From €5,900

  • Up to five workflows, or one that spans two or more systems
  • Every workflow scored on the same four columns, with a call for each
  • Nothing in production yet? The 2 or 3 processes most worth starting, ranked
  • A board memo and a roadmap for closing the measurement gaps
  • 3 to 4 weeks
Book a call

Larger portfolios

A fixed quote after the call

  • More than five workflows, or several business units
  • The same scoring, calls and board memo as the portfolio review
  • Want someone to own the portfolio afterwards? See the fractional AI lead
Book a call

Systems are the software a workflow connects to, like your CRM or ERP. Prices exclude VAT. The review fee counts in full toward a sprint booked within 60 days. Know which workflow needs work? Start with the AI delivery sprint.

Who I work with

I’m Ravi, and I lead every review, with senior advisors and specialists chosen for your processes.

Ravi Ayyala

Ravi Ayyala

Founder and product AI lead

  • Leads the review, translates from business to AI, the build if it makes sense, and user adoption.
  • Ex-LinkedIn, ex-Blacklane (Uber) product leader.
  • MBA (Oxford), AI trained (Stanford), Engineering (NUS)
Linde Scheers

Linde Scheers

Marketing and growth advisor

  • Advises SmartProduct and selected client work on positioning, adoption and distribution.
  • Former Google leader and former CMO at eBay Germany.
  • Currently CMO at ING Germany.
Linde Scheers work and education
Shankar AVSB

Shankar AVSB

Finance and strategy advisor

  • Advises SmartProduct and selected client work on investment thinking, business outcomes, and technology.
  • Technology investor with experience at ABB and Infosys and study at Stanford.
Shankar AVSB work and education

FAQ

Everything else worth knowing before you book a call.

Claude Code helps your team set up AI workflows. It doesn’t tell you which three numbers matter to your board, or which teams have adopted what’s in production. The review answers both.

• The fee is fixed in writing before I start, and is the same whatever the call.
• You keep the scorecard and the method, so your own team or anyone else can do a fix.

• A sponsor who makes the keep, fix or stop decisions, about 2 hours in total.
• Each workflow owner, about 2 hours for the definition session and review.
• Read access to billing exports and usage logs, where they exist.
• 3 to 5 interviews of 30 minutes with people who do the work.

Access is agreed upfront, read-only where possible. I show you anonymised example scorecards before you share anything real, and nothing proprietary leaves your environment without your agreement.

No. It weighs value and risk together. The risk read covers data sensitivity, reliability evidence, audit trail and incidents. It draws on AIUC-1 (the standard for AI agent security, safety and reliability) and Munich Re’s due-diligence areas for insuring AI. Gaps come in a form your risk and compliance team can pick up. It’s not a certification.

It depends on what the AI could get wrong, not whether it’s customer-facing. For products, including AI systems, placed on the market or put into service on or after 9 December 2026, the directive covers harm to people: death or injury (including medically recognised psychological harm), damage to property that isn’t used exclusively for professional purposes, and destruction or corruption of data not used for professional purposes. It doesn’t cover damage confined to your own business property or records. Three questions worth asking: could a malfunction plausibly injure someone? Could it destroy or corrupt a person’s own data that isn’t used for work? Could it damage property that isn’t purely a business asset? If any answer is yes, being able to show how the system was tested, monitored and updated matters more for anything placed on the market or put into service on or after that date. The review’s risk read flags where that evidence is missing. This is general orientation, not legal advice, so bring these three questions to your legal team rather than starting from scratch.

You tell me what’s in production or planned, and what worries you most. Licences count when they are tied to a named process. I filter that list with three questions:
• Does it move revenue, margin or cost-to-serve?
• Does it need AI, or would something simpler do?
• Does something you already pay for handle it?

No. A sprint is a fixed-length, time-boxed piece of work with a measured result, as in agile. If you already know which workflow needs work, go straight to the sprint. Its week one runs the same check on that one workflow and ends in a go or no-go. The review is for when you’re weighing several, or don’t know where to start.

That’s common, and it’s a finding, not a blocker. You get exactly what’s missing and what it takes to close each gap.

A defensible stop protects you and your board as much as a keep does. It frees budget for work that can pay back.

• You keep the scorecard and the method behind every number, to run again next quarter.
• Where a workflow is worth fixing or starting, the next step is an AI delivery sprint.
• The review fee counts in full toward a sprint booked within 60 days.
• Want someone to own the whole portfolio? That’s the fractional AI lead.

1 to 2 weeks for a single workflow, 3 to 4 weeks for a portfolio of up to five.

Find out which AI workflows pay back, and what to fund next.

A free 30-minute call, no slides. It ends with a yes or no on the review.

Book a call See your AI readiness score