Brantley Davidson

Ex @wpp agency vet turned founder of Prometheus | Helping growing brands with AI

[Brantley Davidson @brantley_d]

The best AI project I've seen this year isn't about efficiency.

It's a firm building a tool to capture 40 years of regulatory expertise before their guy retires.

Every mid-market company has this person. The one who just knows. The one everyone calls.

AI won't replace them. But it can make sure what they know doesn't walk out the door.

[Brantley Davidson @brantley_d]

one of our clients found $240K in pipeline sitting in accounts marked dead.

their AI flagged buying signals from contacts who had gone quiet for 18+ months. things like job changes, company news, renewed web visits.

the sales team had written those accounts off. the AI hadn't.

[Brantley Davidson @brantley_d]

diesel service company billed the wrong rep on a chunk of orders. for years.

ERP splits "order rep" vs "customer rep." HubSpot pulled one field, orders rolled up wrong. 90% of the time they're the same, which is why the 10% never got caught.

fix was one field mapping.

[Brantley Davidson @brantley_d]

a logistics client changed how they evaluate vendors because of one LinkedIn video i posted.

his words: "a short LinkedIn video by one Mr. Brantley" made API access a front of mind question on every vendor call.

one post. shifted a buying process.

[Brantley Davidson @brantley_d]

built a forecast prototype for a client. first demo, he goes "that's phenomenal, man."

then silence.

because the second it worked, the real question showed up: nobody on the team had agreed on how they actually want to forecast.

tech was the easy part.

[Brantley Davidson @brantley_d]

told a room of sales reps about an app you talk to that drops notes straight into the forecast tool.

one guy: "goodness... that'd be amazing."

reps don't want a better CRM form. they want to never fill out the form.

[Brantley Davidson @brantley_d]

talked to a manufacturer who spent 14 months picking a CRM.

6 weeks after launch, sales was still quoting out of spreadsheets.

nobody mapped quote-to-cash before buying the software. they automated a broken workflow.

[Brantley Davidson @brantley_d]

You could buy these just for the mango flavor - they are that good. But - the energy is also awesome.

[Lawrence | SwolLyfe @_TrueVoodoo]

State Performance is officially LIVE ⚡
@stateperform

Nicotine-free energy pouches powered by goBHB® ketones + 50mg caffeine.

Use code STATE30 for 30% off.

10-packs come out to just under $ 7 per puck, plus FREE shipping.

Mango 🥭 + Wintergreen ❄️

Energy. Anywhere.

Link Below⬇️
State30 is not valid on subscriptions

[Brantley Davidson @brantley_d]

seeded one manufacturer with Claude. no training plan, no rollout deck.

two months later, two guys on the floor are building dynamic offline HTML views on their own. company just posted AI staff roles.

the lesson every time: don't roll out AI. seed it and get out of the way.

[Brantley Davidson @brantley_d]

@emollick This requirement - that I 100% agree with - is what is preventing potential adopters from getting in the game.

[Brantley Davidson @brantley_d]

@hnshah This insight struck me recently pretty hard; on implementations for true agents - the most success comes when the business lead with true domain expertise provides feedback directly back to the agent. It's much more a significant partnership on these builds than other tech.

[Brantley Davidson @brantley_d]

On the applied side, it's crazy how so many of our conversations are even as simple as; let's move that paper documentation in a digital format that's a bit more automated/easy for AI to read. OR - just take pictures of the paper so AI can have the data. Then apply the loops/learning.

[Brantley Davidson @brantley_d]

@_TrueVoodoo Heck ye

[Brantley Davidson @brantley_d]

We deployed a customer service agent for a services company in April. The AI worked. Accuracy was solid. Response times dropped 40%.

The team hated it.

The senior reps felt like we were telling them they weren't good enough. The newer reps were afraid it was step one toward replacing them. The ops manager who championed the project started getting cold shoulders in the break room.

We paused the rollout for two weeks. Sat down with each team. Let them ask questions. Let them be skeptical. Rebuilt the workflow so the reps controlled when the AI drafted a response and when it didn't.

Adoption went from 20% to 85% in three weeks after that.

The technology was never the problem. Nobody had talked to the people whose jobs were about to change. That's where most AI projects actually fail.

[Brantley Davidson @brantley_d]

Three clients this quarter told me they needed to cut headcount.

After the audit, all three realized the opposite. They were understaffed.

What looked like too many people was actually too many manual steps. Their teams were buried in data entry, reconciliation, and status update meetings that existed because no system talked to another system.

When you automate the busywork, you don't find people to cut. You find people who finally have time to do the work you hired them for.

One client moved two warehouse coordinators off manual inventory counts and onto supplier relationship management. Revenue from their top 20 vendors went up 11% in one quarter.

Mid-market companies don't have a headcount problem. They have an allocation problem.

[Brantley Davidson @brantley_d]

Mid-market CFOs are approving six-figure AI budgets right now.

Then the invoices hit and nobody on the ops team has time to actually use what got built.

I see this every month. The budget exists. The bandwidth doesn't.

A $40M manufacturer approved $180K for an AI-powered quoting tool. Good idea. Real ROI case. But the three people who were supposed to learn the new system are also running production schedules, fielding customer calls, and covering for a vacancy they haven't filled since January.

The tool launched. Adoption after 60 days: one person uses it. The other two are still emailing spreadsheets.

AI doesn't fail because the technology is wrong. It fails because nobody budgeted the hours for the humans who have to change how they work.

[Brantley Davidson @brantley_d]

Got a call last month from a distributor in the southeast. 200 employees, $60M revenue.

Their compliance director is retiring in September. He's been there 31 years.

Everybody in the building knows him as the guy who just knows. OSHA regs, DOT filings, state-by-state variance on hazmat storage. None of it is written down.

The CEO told me: "We've tried to get him to document it. He's been saying he will for five years."

So we're building a system that sits in on his calls, reads his emails (with permission), and structures what he knows into something searchable.

We're not replacing him. We're making sure September doesn't wipe out three decades of institutional memory.

Every mid-market company has this person. Most won't realize the risk until they're already gone.

[Brantley Davidson @brantley_d]

@businessbarista We run into this constantly. The 'no' goes away the second you can show them where the data is actually flowing and what the spend returned. An uninformed no is the only responsible answer they had...

[Brantley Davidson @brantley_d]

The oversight piece is where I'm seeing mid-market companies break down first. Once they start trusting it (and they will, faster than you planned for), the human validation just stops.

The real skill isn't building the workflow, it's designing one that accounts for how people actually behave when they stop paying attention and how to provide guardrails to keep what's built functioning and not business impacting.

[Brantley Davidson @brantley_d]

I spent a decade inside WPP. I know the old consulting model cold.

You scoped a project. Staffed a team. Billed hours. Delivered a deck. Moved on.

The client got a strategy document and a handoff meeting. Then they were on their own.

That model made sense when the work was campaigns and brand positioning. It falls apart with AI.

AI projects don't end at the deck. The model drifts. The data changes. The team that was supposed to adopt it has questions in week three that nobody scoped for.

So we do something different at Prometheus. We embed. Our team sits inside the client's operations. Same Slack channels, same standups, same KPIs.

Old model: scope, staff, bill, deliver, leave.
New model: audit, embed, build, measure, stay.

The old model sold deliverables. The new model sells outcomes over time.

Most mid-market companies don't need another strategy deck. They need someone who's still there when the AI breaks on a Tuesday.

[Brantley Davidson @brantley_d]

Friday afternoon, 4pm. I'm on a call with the CEO of a mechanical contractor in Mississippi. 200 employees. He's telling me about their AI rollout.

He says, "IT is handling it."

I ask what that means specifically. He says his IT director picked a platform, is running a pilot with the estimating team, and will report back in 90 days.

I ask if the IT director has ever sat in on an estimating meeting. He says no.

I ask if the estimating team was involved in choosing the platform. He says no.

I ask who defined what success looks like for the pilot. Long pause. "I think IT is figuring that out."

This is the pattern I see most often in mid-market companies right now. The CEO decides AI is important. That's the right instinct. Then the CEO hands it to IT because AI is technology, and IT does technology.

The thing is, AI projects in operations live or die on workflow knowledge. The estimators know how bids actually get built. The project managers know which reports matter and which ones collect dust. The AP clerk knows that one vendor always sends invoices with the PO number in the wrong field.

IT knows infrastructure, security, integrations. All critical. But they rarely sit close enough to the daily work to define what "better" looks like for the people using the tools.

What ends up happening is IT builds something technically sound that nobody uses. The estimating team gets a tool that doesn't match how they actually put bids together. The project managers get a dashboard that tracks metrics they don't care about. And three months later the CEO asks why adoption is low and everyone points at everyone else.

The companies I've seen get this right do something different. They pick one team with a real problem. They put someone from that team in the room from day one. IT is there for guardrails and integration. But the people who own the workflow own the project.

It's slower. It's messier. There are more opinions and more meetings. But the thing that gets built actually gets used.

I told that CEO in Mississippi something I've said a dozen times this year: the IT director got handed a project with no problem definition and no success criteria. He was set up to build something nobody asked for.