Paul Atherton of Highspire and Reece Barnes of Adaptive work through how to tell useful AI from noise, which back office workflows are actually worth automating, and how to use a large language model to cut estimating time. Includes four free downloads you can put to work the same day.
Watch part one of the series here.
Jump to a section
Time | Topic |
06:54 | The five levels of a self-managed company |
11:05 | How to think about AI, and how to spot the noise |
14:46 | Four workflows worth automating |
17:11 | What manual AP actually costs you |
20:53 | Getting your team on board |
25:50 | The three types of AI, in plain English |
31:08 | AP automation |
36:45 | Receivables, draw packages and cash flow |
40:06 | WIP and catching profit fade |
45:58 | Four ways estimating burns you |
47:27 | Using a large language model for estimating |
49:07 | The four free downloads |
52:06 | Build it yourself or buy it |
58:38 | Live Q&A |
What we covered
How to tell useful AI from noise
AI needs context specific to construction. Cash flow, project life cycles and job costing do not work like other industries.
Before evaluating any tool, identify the process first. Is it repetitive, and how much manual effort does it take today?
Four workflows worth automating
Chasing approvals
Manual data entry into your general ledger or ERP
Receipt tracking and coding
The month end crunch
What manual AP costs
Roughly 12.5 minutes per invoice, counting every hand that touches it, not just the typing.
At 1,500 invoices a month, that is around 312 hours.
The three types of AI
Optical character reading pulls information off a document.
Large language models learn from your data and find the patterns.
Agentic AI takes the action. Adaptive uses all three together.
WIP and profit fade
WIP is three inputs: AP data, receivables data, and context from the field on percent complete.
Automate the first two and WIP stops being a quarterly scramble. The conversation shifts from how do we build the report to what do we do about what it shows.
Estimating
Class D estimates are where large language models are strongest today. Use them for proposals in the 20 to 30 percent accuracy range.
Teams are taking scope review, class D budgets, material pricing and sub bid comparison from one to two weeks down to about a day.
Free downloads
Four resources referenced in the session, shared by Highspire:
Example proposal for a residential construction company. Use it to show AI the output you want.
Build your own CFO in Claude. Get a second read on your business the way a CFO would.
Class D estimates in Claude. All the prompting included. Drop it in and hit enter.
Proposal builder for Claude. All the prompting to produce a finished proposal.
Your hosts
Paul Atherton, CEO and Co-Founder, Highspire
Reece Barnes, Head of Construction Network, Adaptive
Questions: [email protected]