AI portfolio governance+GTM workflow transformation
Built, run, handed over. Naanya is the practice of Lakshmi Peri.

For B2B SaaS and fintech. Big enough that AI arrived in six departments, small enough that nobody was hired to run it.

Everyone is moving.Nobody is steering.

Teams adopted AI faster than anyone could put a process around it. This is what that looks like right now, and it is where almost every company your size currently sits.

Eighteen years running enterprise transformation. Five million customers migrated across twelve countries without a disruption. Then I built the AI program operations system I now help other companies govern. I build the agents, not just the policy around them. More →

8Initiatives running 0With an owner 0Issues found 0Gate status set
Approved 0
Pending review 0
Not approved 0

What waiting costs. Duplicate subscriptions nobody consolidated. A launch that slips because readiness surfaced late. A tool touching customer data that no one reviewed, found by a customer rather than by you. None of it announces itself.

And here is what it cost you

Your hardest workflow is the one AI skipped.

AI went where one person could adopt it alone. It never reached the work that spans teams and eats the week.

Today · hunted by hand
Where is launch readiness?
Slack #launch 12
Email 8
launch-deck v7
status.xlsx
Battlecard old
Jira !
cert tracker
?
?
?
Three meetings to answer one question. You find out after go-live.
With the operating model
One screen. Always current.
Feature X LaunchReady · Sep 15
Launch readiness85% · on track
Field certificationWest 29% · flagged
Audience overlapResolved
Open decisions1 · owner named
The one real risk is surfaced. You decide before go-live.

Getting from left to right is mostly program design. The AI is what keeps the screen current without anyone maintaining it.

See the right-hand panel actually running
A working demo of the launch readiness system I built: eight agents reading the tools a team already uses, turning scattered status into one decision-ready view.
Open the demo →
Why a policy does not fix it

A policy tells nobody what to do on Tuesday.

The request that lands on your desk

A team wants to start running customer data through an AI tool.

Two depths, and I am direct about which

Govern the portfolio. Rebuild go-to-market.

Most companies attempt the list and stop there.
The inventory is the easy part. Three things are not.
  • Getting people to tell you what they are running. Nobody volunteers a tool they think will be taken away. It comes out of expense reports, SSO logs, and browser extensions, not a survey.
  • Sizing the control to the risk. A notetaker needs a retention setting. A resume screener stops entirely. Getting that wrong in either direction is expensive.
  • Making it survive the handover. Most of these die when the internal owner inherits a document instead of a cadence that is already running.

Everywhere — I govern

  • Every AI tool and initiative found, with a named owner
  • Risk tiered and routed: pre-cleared, committee, or IT and Legal
  • Intake and a review path that does not take six weeks
  • No initiative advances without an owner, a baseline, and a value hypothesis
  • One portfolio view leadership can act on

Go-to-market — I rebuild

  • Launch readiness, forecasting, campaign launch, pipeline reviews, exec reporting
  • Mapped end to end, with information-only steps removed
  • Decision rights set so approvals stop round-tripping
  • AI applied where it drafts, humans kept where they decide
  • Measured against a baseline agreed before anything changes

Finance, HR, and engineering workflows get governed, not redesigned.

How this goes

Start where it costs nothing.

STEP 1 · START HERE
Take the check →
Twelve questions, five minutes. You get a score and the three gaps costing you most.
STEP 2
Twenty minutes on a call
I walk you through your result. You leave with two or three things to do whether or not you hire anyone.
STEP 3
A scoped proposal
Fixed fee and fixed dates, both written down before anything starts. Only if it is worth doing.
01

Portfolio check

Free · 5 minutes
  • Scored the moment you finish
  • Four domains, scored and ranked worst first
  • A 20-minute readback if you want one
Take the check →
No pitch unless you ask
02

AI Portfolio Review

Fixed fee and fixed dates, set in the scope call
  • Every AI tool and initiative found, with a named owner
  • A decision on every one: approved, pending, or not approved
  • An operating model: one front door for new requests, review routed by data sensitivity, a published bar for what ready means
  • Used on live initiatives, not left as a document
  • Handed to an owner you name, with the cadence already running
Answers problem one
03

GTM Workflow Transformation

One workflow at a time · fixed fee, set in the scope call
  • One workflow at a time: launch readiness, forecasting, pipeline reviews
  • Mapped end to end, with information-only steps removed
  • Decision rights set so approvals stop round-tripping
  • AI applied where it drafts, humans kept where they decide
  • Measured against a baseline agreed before anything changes
Answers problem two

After handover, I stay on call. A new initiative nobody knows how to route, a vendor or a rule that changes, a model that needs a tune after six months of real use.

In their words

People I did the work for.

“The price implementation was a tremendous success. $75M in revenue, under 1.2% attrition, no impact on customer acquisition. Lakshmi was the glue that brought it all together.”
Tony KingSVP Pricing & Commercialization, Xero
“A pivotal leadership role organizing governance, operating models, and readiness frameworks. A major reason we delivered without any customer or business disruption across 5+ product lines and a dozen countries.”
Joe WolfIntuit

More recommendations on LinkedIn →

The deliverable

What you actually receive.

The Portfolio Review produces one document your leadership can act on. These are its sections.

01
Initiative inventory
Every AI tool and initiative found, with a named owner, the data it touches, and a risk tier.
02
Decisions
Approved, pending review, or not approved on each one, with the reasoning and what has to change to move it.
03
The operating model
One front door for new requests, a review that routes by data sensitivity, a published bar for ready, and who decides what.
04
Go-to-market findings
The workflows AI never reached, ranked by time recoverable, with the specific step that is costing you.
05
Value baseline
Hours converted to capacity, backfill avoided, or spend removed. Each figure attested by the person who owns that process.
06
First ninety days
Sequenced, with owners and dates. Not a list of everything wrong.

I will walk a redacted example through on the call, so you can see the format before you commit to anything.

Boundaries

What I do not do.

  • Red teaming, model evaluation, or security testing
  • Legal opinions on regulatory exposure
  • Independent audits or certification
  • Training or fine-tuning models. I build agent workflows on top of existing ones.
  • Workflow redesign outside go-to-market
  • Buying or reselling AI tools
  • Standing in as your permanent AI owner
  • Anything that cannot be handed over

Better you know before the first call than after the contract. When these are needed, I say so and help scope them to someone qualified.

What I need from your team

An executive sponsor, and about two hours a week from four or five people who know how the work actually runs. Read access to the tools already in use. No engineering time, and no new licences to buy before we start.

How I handle your data

Under your NDA and inside your tenancy wherever possible. Nothing is copied to personal accounts or consumer AI tools, nothing trains a model, and everything is returned or destroyed at the end. If a step would put your data somewhere you would not choose, I raise it before doing it.

Background

I built it before I advised on it.

Fourteen years at Intuit running enterprise transformation, most of it in go-to-market. Five million customers migrated across twelve countries without a disruption.

Then I built an AI program operations system in production. Eight specialised agents on LLMs handling planning, risk, dependency tracking, and executive reporting. Weekly leadership update prep went from about ten hours to under thirty minutes.

Single orchestrator, scoped skills, not a network of independent agents. Multi-agent demos better. Single orchestrator is far easier to audit and gate. If you cannot explain who approved what, you do not have governance. You have a demo.

Questions

The things people ask first.

How long does this take?
A Portfolio Review is a matter of weeks, not months. A workflow rebuild is longer, because it has to run through a real cycle to be worth anything. Exact dates are set in the scope call and written down before we start, so this does not become open-ended. Those dates assume interviews get scheduled in the first week and read access is granted early. If that slips, the end date moves by the same amount, and I will say so at the time rather than at the end.
What does it cost?
Fixed fee, agreed before we start, so there is no meter running and no invoice you did not expect. These are five-figure engagements, not six. If that is going to be wrong for you, you will know on the first call rather than three weeks into a proposal cycle.
Can we start smaller?
Yes, and often that is the better route. A short first phase covering the inventory and the interviews, priced on its own. At the end you have something useful whether or not it goes further, and the next phase gets scoped against what we actually found rather than what either of us assumed.
What if the scope changes partway through?
What is in and what is out gets written down before we start. Anything outside that is a separate conversation and a separate number, raised when it comes up rather than absorbed quietly and billed later.
We already have an AI policy. Is this still relevant?
Usually more relevant, not less. A policy tells people what is allowed. It does not tell a manager what to do on Tuesday, and it does not tell you what is currently running or who owns it. Most of the work here is turning a policy into something that operates.
Who from our side needs to be involved?
An executive sponsor, and roughly two hours a week from four or five people who know how the work actually runs. No engineering time, and nothing new to buy before we start.
Do you need access to our systems?
Read access to what is already in use is enough for the Review. Nothing is copied into personal accounts or consumer AI tools, nothing trains a model, and everything is returned or destroyed at the end. If a step would put your data somewhere you would not choose, I raise it before doing it.
Will you sign an NDA?
Yes, before anything substantive is shared. Your paper is fine.
We are smaller or larger than that band. Does it still work?
The size range is shorthand for a condition: AI arrived across several departments and nobody was hired to run it. Below roughly two hundred people there is usually not enough surface area to need this. Above a few thousand you likely have an internal function, and the work is different.
Can you just fix one workflow?
Yes. Some engagements start there, particularly if you already know which one is bleeding. The Review exists to find that out when you do not.
What happens when you leave?
A named internal owner, a cadence already running, and documentation written for them rather than for me. If that handover cannot happen, the engagement was not worth doing.
Written up

Six questions to answer before funding another pilot.

Each one answered with a working framework rather than an opinion.

All six frameworks →

Find the one you did not know was running.

Somewhere ahead of this: one list, one owner, a decision on every initiative, and a launch cycle that does not eat a week. Start with five minutes. The result is yours whether or not we ever talk.