AI Plan and Build
Built on Real Production AI

Pre-close diligence. Twelve-week production sprints. Portfolio-wide platforms. Three offerings under one practice. Built on the AI Intel Service we operate in RenovationRoute. Run by a senior engineer, not a partner who hands off to associates.

Why Most AI Engagements Disappoint

Big consulting firms produce slide decks. Strategy, roadmap, use case shortlist, tech stack rec. Twelve weeks later the team has decks and no shipped product.

Boutique AI shops produce demos. Works on toy data, breaks on real data, has no evals, no fallbacks, no cost controls, no security thinking. Pilot stuck in pilot.

Internal teams want to do it themselves and often can, but they have not shipped a production LLM system before. They underestimate latency. They underestimate prompt injection. They overestimate the model. Their first attempt does not survive contact with real customers.

We sit in between. Senior engineer who has shipped production AI, paired with the internal team, with a clear end state and a fixed scope.

Three Ways We Engage

Pre-close, post-close, or across the portfolio. Pick the one that maps to your situation, then read the detail below. Every engagement is fixed scope and priced on a call, because AI work varies too much to put a number on before we understand it.

Pre-Close

AI Diligence, Before You Close

Standard cyber DD misses AI. Shadow LLM usage, prompt injection exposure, IP risk from training data, and the "AI feature" that is one model deprecation away from breaking. Most diligence providers check for SOC 2 reports and firewall configs. They do not look at the OpenAI API key in the engineering team's shared vault that has been processing customer data for eight months with no logging.

The result is that AI risk surfaces six months post-close, when a customer notices their data showed up in another customer's output, the model vendor changes terms, or the team's "AI feature" turns out to be a thousand-line prompt nobody can maintain. AI Diligence catches this pre-close, done by someone who has shipped production AI and knows what failure modes to look for. The deliverable is a risk-ranked report your deal team can use in IC and post-close planning.

What We Cover

Six workstreams over one to two weeks, depending on target complexity.

01

AI surface mapping

Where AI is used in the product, in internal tooling, in customer-facing automation, and in the team's daily workflow. Includes the shadow usage nobody admits to.

02

Model and vendor risk

Which models, which vendors, what terms, what data residency, what happens at renewal. Lock-in exposure and deprecation risk.

03

Data and IP exposure

Customer data flowing to model providers. Training data provenance. Indemnification on outputs. IP contamination from open-source models with restrictive licenses.

04

Security exposure

Prompt injection paths. Output validation. Access controls on AI features. Whether the team's "AI feature" can be tricked into leaking customer data or executing tools it should not.

05

Value creation feasibility

The AI story in the CIM is usually optimistic. We tell you which post-close AI plays are realistic in twelve months and which are wishful thinking. Feeds directly into your 100-day plan.

06

Team capability assessment

Does the engineering team have anyone who has shipped production AI. Honest read on whether the post-close plan needs new hires, a sprint partner, or both.

How AI Diligence Works

Day 0: Kickoff and access

Thirty-minute call with the deal team. NDA in place. We get read-only access to the data room and a list of target-side technical contacts.

Days 1 to 4: Document review

Architecture diagrams, vendor contracts, security questionnaires, prior audits, AI vendor lists, model usage logs if available. We surface what the data room shows and what it conspicuously does not.

Days 5 to 7: Technical interviews

Three to five sessions with the target's CTO, lead engineers, and security lead. Specific questions, not generic checklists. We get the real story on how AI is actually used.

Days 8 to 10: Synthesis and report

Risk-ranked findings, post-close remediation priorities, 100-day plan input, and a clear read on the value creation AI story. Delivered as a written report and a thirty-minute review with the deal team.

What You Get

Written diligence report

Twelve to twenty pages. Executive summary, risk register, findings by workstream, post-close priorities, appendix with technical detail. Format works for IC and for board.

Risk register

Every material finding ranked by likelihood, impact, and time to remediate. Each item maps to a concrete post-close action.

100-day plan input

A short section that feeds directly into the operating partner's post-close plan. Specific AI work the portco should do in days zero to one hundred.

Deal team review session

Thirty to sixty minutes with whoever needs to be in the room. Q&A on findings, scenarios, and how the AI posture should change the deal.

Engagement shape: pre-LOI or pre-close, one to two weeks, flat fee. We can move on a one-week timeline when the deal is live. Final scope and price set on the intro call.

Post-Close

The 100-Day AI Sprint

Pick one high-ROI AI use case. Ship it to production. Hand it off to the portco team. Twelve weeks, fixed scope, one senior engineer. This is the offering for PE-backed portcos with a 100-day value creation plan that names AI. The deliverable is working software in production, not a roadmap.

Most AI consulting engagements deliver a slide deck and nothing ships. Most portco engineering teams are good but have not built production LLM systems, so their first attempt does not survive contact with real customers. The sprint solves both: one senior engineer who has shipped production AI works alongside the portco team, picks the right use case, builds it correctly, and leaves the team able to keep operating it.

Week by Week

Twelve weeks. Three phases. Everything is on a calendar from day one.

Week 1: Use case selection

On-site or remote workshop with the portco CEO, CTO, ops leaders, and any sponsor operating partners. We surface every plausible AI use case, score by value, time to ship, and risk, and pick the sprint target. The other top candidates become the post-sprint roadmap.

Weeks 2 to 4: Architecture and prototype

Model selection (closed vs open, frontier vs cheap, hosted vs self-hosted). Data flow design. Eval framework set up before any prompt work, so we have a way to measure changes. Working prototype on real portco data by end of week four. Sponsor and portco review.

Weeks 5 to 8: Production build

Integration with existing systems (CRM, ERP, ticketing, whatever it touches). Cost controls and rate limiting. Prompt injection defenses. Observability for prompts, outputs, costs, and latency. Human-in-the-loop where needed. Fallback paths for when the model is wrong, slow, or down.

Weeks 9 to 10: Hardening and rollout

Staged rollout to a real user subset. Eval results against the framework set up in week three. Performance and cost tuning. Security review. Sign-off with the portco team and sponsor.

Weeks 11 to 12: Handoff and roadmap

Production cutover. Runbooks the on-call team can actually use. Training sessions with the portco engineering team so they own it. Written roadmap for the next two use cases ranked by value. Final report to the sponsor.

What Ships

Every deliverable is something the portco team can use after we leave.

Production AI service

The actual working software, deployed in the portco's environment, integrated with the systems it needs to touch. Live, with real users.

Eval framework

A repeatable way to measure whether changes to prompts, models, or pipelines make the output better or worse. Without this, every change is a guess.

Observability and cost controls

Dashboards for prompt traffic, output quality, latency, and spend. Alerts that fire before the bill becomes a surprise.

Security and governance baseline

Prompt injection defenses, output validation, access controls, data handling rules. Documented for the sponsor and the portco's security or compliance lead.

Runbooks and on-call docs

The portco team has to operate this after we leave. They get runbooks for the failure modes that actually happen, not generic templates.

Ranked next-use-case roadmap

The two or three use cases that did not make the sprint, with rough effort and value estimates. The portco can take this to the sponsor as the year-two AI plan.

Real Differences

Why This Is Not What McKinsey Sells

One senior engineer, not a pyramid

You do not get a partner who hands off to a manager who hands off to two analysts. You get the person who built RenovationRoute's AI Intel Service, on your engagement, full stop.

Shipped before the sprint ends

Strategy firms produce a roadmap and call it a win. We produce running production code and a team that can keep operating it. Different work product, different value.

Security baked in, not bolted on

Prompt injection, data leakage, output validation, IP risk. We handle these inside the build, not as a separate workstream that gets cut when the budget tightens.

Honest about what AI cannot do

Some use cases do not need AI. Some need a rule, a script, or a better workflow. We will tell you when that is the case, even if it shrinks the engagement.

Engagement shape: post-close, twelve weeks, fixed scope, one senior engineer end to end with production cutover at week twelve. Sponsor-paid or portco-paid, both work. We do not bill by the hour. Final scope and price set after the intro call, based on use case complexity and integration depth.

Portfolio-Wide

The Portfolio AI Platform

Ten portcos. Ten separate model bills. Ten different prompt injection postures. Ten teams reinventing the same eval pattern. When AI shows up in three or more portcos, the portfolio starts paying a tax: higher per-portco spend because nobody negotiates as a portfolio, inconsistent risk that surfaces in cyber DD when a portco gets sold, and slower time to value because every team starts at zero.

A portfolio AI platform fixes all three. It is not a software product. It is a thin layer of shared infrastructure, governance, and patterns, plus a senior partner who keeps it current, so the next sprint compresses from twelve weeks to six.

What the Platform Covers

Pick the pieces that matter for your portfolio. Most sponsors start with three or four and grow.

Shared LLM gateway

One place every portco routes traffic through. Per-portco usage, cost, and rate-limit controls. Portfolio-level vendor negotiation. Easy to swap models without rewriting every portco app.

Eval framework

A common way to measure AI output quality across portcos. Reusable test sets for common patterns (extraction, classification, summarization). Every portco does not rebuild this from scratch.

Governance baseline

A short, opinionated set of rules: what data can go to what model, prompt injection defense expectations, output validation requirements, vendor risk approval process. Real policies, not a forty-page deck.

Reusable use-case patterns

The five or six AI use cases that come up in most mid-market portcos (intake, scoping, search, support deflection, sales ops) implemented as starter patterns each portco can adapt.

Observability and cost reporting

Portfolio-level dashboards on AI spend, traffic, output quality, and incident counts. The sponsor sees what is happening without chasing every portco for a monthly update.

Operating partner advisory

Monthly working sessions with the value creation team. We bring the cross-portfolio view (what is working, what is failing) so operating partners can spot patterns earlier.

How the Engagement Runs

Month 1: Portfolio assessment

Quick read on every portco that uses AI today or plans to. What is in place, what is missing, where the biggest exposures are. Drives the platform build priorities.

Months 2 to 4: Initial platform build

Stand up the gateway, eval framework, governance baseline, and observability for the first set of portcos. Phased rollout so we are not boiling the ocean.

Ongoing: Operate and extend

Monthly working sessions with operating partners. New portcos onboard onto the platform. Pattern library grows as we learn what works. Quarterly reviews with the sponsor on portfolio AI posture.

On demand: Compressed sprints

Portcos on the platform get faster, lighter 100-Day Sprints, because the shared infrastructure and patterns do a chunk of the work up front.

Engagement shape: portfolio-wide, recurring, with a twelve-month initial term so we can build something real rather than stitch together monthly fire drills. Scoped to the size of the portfolio, the platform components in scope, and the cadence of operating-partner engagement. Usually makes sense at five or more portcos doing AI.

The Proof

Built on Production AI We Operate

We do not just talk about production AI. We run it. Every pattern we sell is one we have shipped, broken, and fixed under real load.

The AI Intel Service in RenovationRoute

A Python and FastAPI microservice that takes a homeowner's plain-language project description and produces a structured scope a contractor can bid on. Live in production. Real users, real money on the outputs.

Every production question got answered the hard way

Evals before prompts. Model selection by tradeoff, not by brand. Output validation against a schema. Prompt injection defenses. Cost caps in the gateway. Fallback paths for when the model is wrong, slow, or down. The patterns we sell are the patterns we shipped here first.

Read the full technical case study

What we built, what we broke, and what we fixed. Honest about what AI can and cannot do in a production system handling real money.

Who Hires Our AI Practice

PE deal teams

Evaluating a target with AI exposure. Want a one-to-two week read before close on what is real and what is wishful in the CIM AI story.

PE operating partners

Have a portco with AI in the value creation plan and need real production traction before the next board meeting. Sprint shape works for this.

Portco CEOs and CTOs

Engineering team is good but has not shipped a production LLM system. Need a partner who pairs with the team for twelve weeks and leaves them able to ship the next one.

Sponsors with 5+ AI portcos

Multiple portcos doing AI in different ways, inconsistent risk, no shared platform. Portfolio engagement makes sense at this scale.

Mid-market companies under board pressure

Customers, investors, or the board expect AI. The team has not shipped one. Sprint shape works here too.

Pilots stuck in pilot

Built something six months ago. Works on demo data, fails on real data. Need someone to take it the last mile.

Talk to someone who has actually shipped this

Thirty-minute call. We figure out which of the three offerings fits, or whether none of them do, and scope it before you commit. No pitch.