Most companies think they need AI. Very few know where it will actually make money. One of the clearest examples of this I’ve come across: a $150M+ piping manufacturer in Houston whose executive team was sitting on a goldmine they didn’t know existed—until an outside team came in to help them see it.
Here’s how that team audited the entire operation, scored 11 AI initiatives, built a phased roadmap, and shipped Phase 1 in 60 days—projecting $1.2M to $2.5M in annual ROI. Not theoretical value. Real, measurable, CFO-approved value.
If you’re a mid-market company wondering where to start with AI, this is the playbook.
Why the Company Sought Outside Help
The company—let’s call them PipeTech—had been in business for 28 years. They’d built a solid operation: 400+ employees, $150M+ in annual revenue, a massive product catalog, and a customer base spanning oil & gas, construction, and industrial manufacturing.
But they had a problem. Their VP of Operations had been to three AI conferences in the past year, and every vendor was telling him a different story. One said they needed a chatbot. Another pitched predictive maintenance. A third wanted to sell them a $400K computer vision system for their warehouse.
“We know AI is important. We just don’t know what’s real and what’s hype. And honestly, we don’t know where to start.”
That was the honest admission from their CEO in the initial call. They didn’t need a vendor. They needed a strategy.
The Audit: What It Actually Looked Like
Worth demystifying because “AI audit” sounds like something that requires a team of 20 and six months. It doesn’t. Here’s what the team actually did, start to finish, in about four weeks.
Week 1: Process Mapping & Stakeholder Interviews
The team interviewed 23 people across 8 departments—everyone from the CEO to warehouse floor leads to the sales rep who’d been there 19 years and could quote pipe prices from memory. Each conversation was 30-45 minutes, built around the same three questions:
- What takes you the most time every day? (Identifies automation opportunities)
- Where do you feel like you’re guessing when you should know? (Identifies intelligence gaps)
- What information do you wish you had but don’t? (Identifies reporting and prediction opportunities)
The patterns emerged fast. Sales was spending 40% of their time on manual quoting—pulling pricing from spreadsheets, cross-referencing inventory, calling the warehouse to check stock. Customer service was answering the same 15 questions over and over. And the executive team was flying blind on which customers were at risk of churning.
Pro tip: Don’t just interview managers. The people doing the actual work know where the pain is. A warehouse lead revealed they were manually re-entering delivery data into three different systems every day. That single insight turned into a $180K/year automation opportunity.
Week 2: Data Assessment
This is where most AI initiatives die. Not because the data doesn’t exist, but because nobody knows what they have.
PipeTech had been in business for nearly three decades. That means they were sitting on 80 million+ rows of transactional data—sales orders, inventory movements, customer interactions, delivery records, supplier performance data. All of it trapped in a legacy ERP system, a handful of Access databases (yes, really), and about 200 spreadsheets that only specific people knew how to navigate.
Everything was catalogued—a data inventory mapping each data source to its quality level (clean, needs work, or unusable), refresh frequency, and accessibility. Here’s what the audit found:
- 60% of their data was usable with minor cleaning
- 25% needed significant transformation (different formats, missing fields, duplicate records)
- 15% was effectively unusable (legacy systems with no documentation, data locked in proprietary formats)
That 60% number was great news. It meant we could start building real value without a massive data infrastructure overhaul. We just needed to be smart about which initiatives we tackled first.
Week 3: Opportunity Scoring
Here’s the framework that makes this whole approach work. For every AI opportunity identified, scoring happens across five dimensions:
- Revenue impact — How much money will this make or save per year?
- Implementation complexity — How hard is this to build? (1-5 scale, where 1 is an off-the-shelf tool and 5 is custom ML models)
- Data readiness — Do we have the data we need right now?
- Organizational readiness — Will people actually use this? Is there a champion internally?
- Time to value — How quickly will we see results?
Each dimension gets a 1-5 score, and we calculate a weighted composite. Revenue impact and data readiness get double weight because they’re the biggest predictors of whether an initiative will actually succeed.
The audit surfaced 11 scored initiatives, ranging from a simple automated reporting dashboard (score: 4.6/5) to a full predictive pricing engine (score: 2.8/5—high value but the data wasn’t ready yet).
Week 4: The Roadmap
This is where the scores turned into a plan. Here’s the three-phase framework that guided the work:
Phase 1 (0–90 days): Quick wins. High ROI, low complexity, data is ready. These build momentum and prove to the organization that AI actually works.
Phase 2 (90–180 days): Strategic builds. Medium complexity, higher value. These require some data work and cross-functional coordination.
Phase 3 (180+ days): Transformational bets. High complexity, highest potential value. These are the game-changers, but they need the foundation that Phase 1 and 2 build.
For PipeTech, Phase 1 included three initiatives:
- Live operations dashboard — Real-time visibility into sales pipeline, inventory levels, and delivery status. Estimated value: $320K/year in time savings and faster decision-making.
- VIP customer reporting — Automated weekly reports for their top 50 accounts, flagging order anomalies and churn risk. Estimated value: $280K/year in retained revenue.
- Quote automation assistant — An AI tool that pre-fills quotes by pulling from historical pricing and current inventory. Estimated value: $190K/year in sales productivity.
Shipping Phase 1 in 60 Days
Here’s the part most AI strategies skip—the actual delivery. A roadmap that lives in a slide deck is worth nothing. The team committed to shipping Phase 1 working software in 60 days, not 90. Why? Because momentum matters more than perfection at this stage.
They started with the live dashboard because it had the highest score and the cleanest data path. The ERP system had an API (barely documented, but it existed), and the key data points could be pulled without major transformation work.
Week 1-2 was data pipeline setup. The team built connectors to their ERP, cleaned and normalized the key fields, and set up a refresh schedule. Nothing fancy—just reliable.
Weeks 3-4 were dashboard development. A modern BI stack powered five views: sales pipeline, inventory status, delivery tracking, customer health scores, and an executive summary. Each view was designed with the specific user in mind—the warehouse manager sees different things than the CFO.
Weeks 5-6 were user testing and iteration. This is where it got real. The VP of Sales looked at the customer health scores and immediately spotted three accounts they didn’t know were at risk. One of those accounts was worth $1.2M in annual revenue. They called the customer that afternoon.
“We’ve been flying blind for years. I didn’t know we could see this. Why didn’t we do this sooner?”
By week 8, the dashboard was live across the organization. The VIP reporting was automated and running. And the quote assistant was in beta with three sales reps.
The ROI: $1.2M to $2.5M
Here’s how those numbers break down, because skepticism about AI ROI claims is well-earned.
The $1.2M floor represents just Phase 1 value, conservatively measured:
- $320K in operational time savings (measured by reduction in manual reporting hours)
- $280K in retained revenue (based on the at-risk accounts flagged in the first month)
- $190K in sales productivity (measured by reduction in average quote time)
- $180K in process automation (the warehouse data re-entry problem and similar quick fixes)
- $230K in faster inventory decisions (reduction in overstock and stockout events)
The $2.5M ceiling includes Phase 2 and 3 projections: predictive demand planning, automated procurement optimization, and the full pricing engine. These are modeled but not yet proven—which is why we present both numbers.
Key principle: Always present ROI as a range, not a point estimate. Your CFO will respect you more, and your projections will be more defensible. The floor should be based on what you’ve already shipped or are actively building. The ceiling includes future phases with higher uncertainty.
What I’d Tell You If You’re Starting This Journey
First, don’t start with the technology. Start with the business problems. Every vendor will tell you that their tool is the answer, but if you don’t know the question, you’ll waste six figures finding out the hard way.
Second, talk to the people doing the work. Executives know what they want to achieve. Frontline employees know what’s broken. The magic is in connecting those two perspectives.
Third, score everything. If you can’t quantify the impact, you can’t prioritize. And if you can’t prioritize, you’ll end up with a pilot graveyard—a bunch of half-finished experiments that never made it to production.
Fourth, ship fast. Your first Phase 1 deliverable doesn’t need to be perfect. It needs to be useful. The moment someone in the organization says “I can’t do my job without this anymore,” you’ve won.
And finally, build the roadmap to survive change. The AI landscape shifts every quarter. Your roadmap needs to be structured enough to drive accountability but flexible enough to incorporate new capabilities as they emerge. That’s why I use phases instead of fixed quarterly plans—the phases define what you’re trying to achieve, and the specifics can evolve as the technology does.
PipeTech is now six months into their AI journey. Phase 2 is underway, the dashboard has become the first thing their leadership team checks every morning, and they’ve already exceeded the conservative $1.2M projection. The quote assistant alone saved their sales team 600+ hours in the first quarter.
The ROI was always there. They just needed someone to help them see it.
If your company is sitting on years of data and wondering where AI fits, the answer is probably closer than you think. You don’t need a moonshot. You need a map.