The 100-day plan for a new add-on is already full. Finance and payroll cut over, insurance and benefits move, the safety program is aligned, customers and general contractors hear about the deal, and key people sign retention agreements. Adding "AI" to that list sounds like one more systems project competing for the same people.
It doesn't have to be. Scope it to the work that decides margin, the next round of bids, and run it with named owners. This is the playbook we use to bring an add-on onto a shared AI layer in its first 100 days. It assumes a mechanical or HVAC contractor joining a platform that already runs Operon. The phases are mostly about sequencing and ownership, so the shape carries over to other trades.
Who owns what
Before kickoff, two things should be settled. The initiative is a line in the value creation plan with an owner and a review date, not an aside on the integration checklist. And the scope is one bid type: the work the add-on bids most often, such as plan-and-spec commercial HVAC, with specialty work left for later. Then name four owners.
| Role | Owns | In the first 100 days |
|---|---|---|
| Operating partner | Why this is in the plan, and the definitions the portfolio reports on | Sets the priority, clears access and time, chairs the day-45 and day-100 reviews |
| Platform CEO | The operating model: which standards are common and which stay local | Decides the common scope breakdown and verification standard, and makes this a priority for the add-on rather than a side project |
| Add-on estimating lead | Whether a verified take-off is good enough to bid | Picks the live bids, verifies, flags misses and brings the senior estimators along |
| Operon forward-deployed engineer | The deployment: ingestion, configuration and the feedback loop | Maps the add-on's drawings, templates and conventions, reviews every correction weekly and trains the team |
Two more people matter: the add-on's president, who decides whether estimators get time for this in the middle of bid season, and whoever controls the file server and job-cost exports. Name both on day 0.
Days 0–15: inventory and access
The first two weeks establish what exists, where it lives and who knows it. Resist the urge to upload the archive.
- The live pipeline. Every bid due in the next six to eight weeks, with drawings, project manuals and addenda. These are the candidates for the next phase.
- The estimating system. Templates, workbooks, assemblies, labor units, markups, scope letters and standard exclusions. Ask an estimator to walk through a recent bid from invitation to submitted number, and write down in their words how long each step took. That is your baseline.
- The archive. Where bid sets, construction sets, submittals, as-builts and closeout files live, how they are named and what is missing.
- Job outcomes. Job-cost reports, change-order logs and final margins on closed jobs, so estimates can later be set against results.
- The people. Who touches every estimate, who reviews before submission, and who knows things nobody else does.
Access is where timelines slip, so accounts, roles and data connections are set up in this window. Role-based access control decides who sees what: the add-on's estimators work on their own company's jobs, and the platform sees what portfolio reporting needs. Have the security conversation now, not in week six. We do not train AI models on customer content, each customer runs in an isolated tenant in the cloud or on-premise, data is encrypted in transit and at rest, and every verification and change is written to an audit log.
By day 15 you should have a written inventory, working access, a baseline, and three to five live bids chosen for the next phase.
Days 15–45: first workflow, live on real bids
The first workflow is take-off on live bids, because that is where the estimators' week goes and where results show up first. Choose bids of the kind the company wins most often, not the hardest job in the pipeline.
- Run the take-off on the live bid set. Equipment, duct, pipe, fittings and devices are counted across the set and reconciled against the schedules. Every item carries a confidence score and a link to the sheet and region it came from.
- The estimator verifies. Low-confidence items first, then a pass across the set. Corrections are the most valuable output of this phase, because each one teaches the deployment something about this company's drawings.
- Quantities flow into the estimate the company already uses. Don't replace the pricing workbook in week three. Exports map into the existing template, and a common structure comes later, once there is trust.
- Addenda are reviewed as changes. When sheets reissue, the estimator checks what moved instead of recounting the set.
- Run in parallel once, then stop. On the first one or two bids, the estimator's usual count and the verified take-off sit side by side. Differences get explained, not averaged. After that, the verified take-off is the take-off.
Each week, the forward-deployed engineer goes through every correction with the estimating lead and fixes the patterns in configuration: a local engineer's symbol legend, a naming convention for duct systems, a schedule format the models haven't seen before.
The phase is done when the add-on submits a bid whose quantities came from the verified take-off, and the estimator would do it again.
Days 45–75: knowledge base from historical jobs
With take-off live, turn to the archive, selectively. The aim is not to digitize everything the company has ever built. It is to make the jobs that resemble the current pipeline answer questions.
- Choose the jobs. Closed jobs of the types the company is bidding now, ideally won jobs with complete files: bid set, estimate, change orders, job-cost report and as-builts.
- Link estimate to outcome. What was bid, what changed and what it cost. Change orders that trace back to scope the original take-off missed are the most useful records in the archive.
- Capture the senior estimators' review. When senior people verify historical take-offs and note why a job went the way it did, their judgment is recorded against the sheets it applies to.
By day 75 the knowledge base should answer questions like these, to take a few examples:
- What did we carry for test and balance on our last few lab jobs, and did it hold?
- Which scope items came back as change orders most often on this general contractor's projects?
- On jobs designed by this engineering firm, where did our take-off differ from what we installed?
Those questions used to go to one person. Now they go to the history, and the person reviews the answer. We wrote about why that matters in The retiring expert is a data problem.
Days 75–100: portfolio reporting and the next add-on
The last phase turns one company's deployment into a portfolio asset.
- Common definitions. What counts as a bid, a verified take-off, a scope gap and a recovered change order. The operating partner owns the definitions, and every company reports the same way.
- The first portfolio view. The add-on appears beside the platform and earlier add-ons, on the same measures, in the monthly operating review.
- The playbook, updated. Every convention mapped, template translated and mistake avoided goes into the playbook, so the next add-on starts further ahead.
- The next add-on, scheduled. The estimating lead from this add-on joins the next kickoff.
The most credible person to tell a newly acquired estimator that this works is another estimator who sat in the same seat three months earlier.
What to measure
Measure against the add-on's own baseline, not a benchmark from somewhere else. Most of these are directional within 100 days, and some take quarters.
- Use on bid day. Was the submitted number built on the verified take-off? Logins measure curiosity. Bid-day use measures trust.
- Estimator capacity. Bids turned per estimator, and bids declined for lack of time.
- Time to quantities. From documents received to verified quantities, and how long an addendum takes to absorb.
- What verification catches. Corrections made, and scope gaps found before bid day that the old process would have found after award, or never.
- Answers from history. Questions the knowledge base answers that used to need a particular person.
- Margin fade (a lagging measure). Bid margin against final margin on jobs bid with the new workflow. It takes quarters to read, so record bid margins from the first bid.
Common failure modes
- Starting with the archive. Months go into digitizing old jobs, nothing touches a live bid, and the estimators decide it is an IT project.
- No named owner at the add-on. If the estimating lead isn't accountable for the definition of done, nobody is.
- Parallel runs that never end. One or two side-by-side bids build trust. Ten build resentment, because the estimator is doing the job twice.
- Standardizing on day one, or never. Forcing the platform's template in week three loses the estimators. Never converging means the portfolio still can't compare itself on day 300.
- Leaving the senior estimator out. The person with the most judgment has the most to verify. Make them the reviewer who sets the standard, not a user to be won over.
- Bundling it with the ERP cutover. The same people get pulled into both, and the one tied to month-end close wins. Sequence them.
- Treating it as the vendor's project. Our forward-deployed engineers carry the deployment work. The outcome has to be owned by the platform and the add-on.
Day 101
On day 101, the add-on's estimators bid off verified quantities, its job history answers questions, and the portfolio sees it on the same measures as every other company. The next add-on starts on a playbook that is one company better. That is integration reaching the work, the argument we make in The missing layer in the industrial roll-up. The model behind it is on Operon for Private Equity.
