Flagship Case Study · Structural Engineering
Thirty years of project history, finally usable. An assessor can pull the right precedent in seconds.
A 50 plus person California structural engineering and design build firm now has every foundation report, plan set, permit, and slope assessment in one place its team can actually ask. The knowledge that used to live in a few experts' heads, and in folders nobody could search, is available to everyone. Heed showed leadership what became possible against their real project data, then scoped the build. The proposed implementation sits with leadership for review.
multiple CA offices
Salesforce migration
for one secure index
software against real data
A firm that had outgrown its own filing system.
The client is a 50 plus person California structural engineering and design build firm headquartered in the northeast LA foothills, in business for more than 30 years, Inc. 5000 recognized, and the engineer of record on work that defines the state's risk profile: foundation evaluations, hillside and slope stabilization, seismic retrofits, retaining structures, and rehabilitation of complex or historic buildings. The pipeline never sleeps because the ground never stops moving.
Eighteen months earlier they had hired a Salesforce consultancy to put every project, contact, and opportunity on a single CRM. They were 85 percent through the build, roughly $400K in on licenses, integrations, and consulting hours, when leadership did the math. The consultancy quoted another $300K and six months to finish, and even then the system would track opportunities and log accounts but never tell a project manager where the foundation report from a 2019 assessment was filed.
Two of the leadership team had used AI tools daily for over a year. They had seen what large language models do with unstructured data, so they asked a different question. What if the answer was not a better CRM, but a different kind of system entirely.
The data was everywhere except where they needed it.
The diagnosis was clear within the first hour. The firm did not have a CRM problem. It had a knowledge problem. Project information was scattered across SharePoint folders organized by year, by region, and by lead engineer. Microsoft 365 held client correspondence. Salesforce held opportunity records. Plan sets, permits, soil reports, and as-builts lived in PDF libraries. Jobsite photos sat on personal phones and shared drives. Budgets and time entries lived in QuickBooks Time. Personal notebooks held the rest.
An assessor preparing for a hillside inspection needed to know whether the firm had touched the property before, whether prior recommendations were implemented, what the budget looked like on similar slope projects, and what the permit history was. None of it was in one place or searchable across systems. Every question started with a SharePoint search, an email search, a Salesforce search, a phone call to whoever last touched the project, and a coffee.
Across 20 active users at a $75 per hour blended rate, that is more than $300K per year in lost capacity. Capacity that should have been spent on engineering judgment, not on hunting for files.
Leadership could finish the migration and still have none of this, or ask whether a different architecture could put the firm's whole history within reach of every engineer. They chose to ask.
Two weeks, real data, working demo.
Every Heed engagement begins with a Proof of Concept against the client's actual data. Not a slide deck. Not a vendor demo against a fake dataset. Real records, real documents, real photos, real meeting transcripts. The POC proves the architecture and the integrations work before anyone signs a production budget.
We ran a two-week sprint. The technical choices were deliberate.
- Salesforce, read-only. We built a connector that reads from Salesforce live, mid-migration. The client had already invested in cleaning up their account and opportunity data. We did not ask them to rip any of it out. We let them keep that work and read from it.
- SharePoint and Microsoft 365. Direct connectors to where the team already lived. The firm did not need to move documents into a new repository. The system met them where their files already existed.
- Anthropic Claude for reasoning. The reasoning agent layer answers questions like "show me every hillside project in Pacific Palisades since 2018 where we recommended caissons" with citations back to source documents.
- OpenAI for image processing. Jobsite photos, foundation cracks, slope conditions, retrofit progress shots. The system tags, indexes, and surfaces them by project, by date, and by visual signal.
- Document intelligence. Plan sets, permits, soil reports, and budgets parsed into structured data the agents can reason over.
- Meeting transcript analysis. Internal meetings transcribed, summarized, and tied back to the projects they reference.
- BCF compliance dashboard. A Business Control Function view that surfaces project status, budget variance, and permit aging for leadership in one place.
- Secure project knowledge index. Everything searchable, every answer cited, every access logged.
Two weeks in, the system was running against five real projects from the firm's active portfolio. Leadership could ask it questions in plain English and get cited answers in seconds, with an audit trail back to the source.
We named it Project IQ, then scheduled the first demo.
Two demos. One validated thesis.
The first demo was 45 minutes. We asked leadership for a question they would normally hand off to an assistant, then ran it against a live job, the Levine project. The system surfaced three change orders that had never made it into the central record and exposed roughly a 40 percent gap between what the active file held and what the project had actually become. Someone in the room said the migration consultancy had not produced anything that useful in 18 months.
A week later we expanded the corpus to roughly 25 active jobs and showed search depth across regions, audit logging, role based access, and the BCF compliance view. A hillside assessor could now prep a site visit in 10 minutes instead of 90, with every answer cited back to source documents.
The proposed Phase 1 implementation sits with leadership for review. For context, the same firm had received discovery quotes from a Big 4 consultancy at $150K to $250K, with 12 to 16 weeks before any working software would exist. Heed arrived with working software in two weeks against their actual data.
Phase 1 scope. Fixed fee. Three week window.
Phase 1 is deliberately scoped. Production deployment on secured infrastructure with a fixed fee paid in two milestones (kickoff and production handoff), a two to three week build window, and a 30 day post launch tuning period. Final scope and pricing sit with leadership for review.
Production deployment
Project IQ deployed on secured Azure infrastructure. Cloudflare Zero Trust at the edge. Microsoft Entra ID single sign-on for the 20-user cohort. ISO 42001-aligned controls and full audit logging from day one.
Live system connectors
Salesforce read connector with managed API. SharePoint and Microsoft 365 connectors. QuickBooks Time integration. Custom internal data sources. The team kept working in the tools they already used.
Corpus expansion
From the 5 demo projects to the 20 to 30 active jobs the firm runs concurrently. Document intelligence indexed plan sets, permits, soil reports, and budgets. Image processing tagged jobsite photos.
Role-based training
Three training tracks. Field assessors learned how to surface prior site history before inspections. Project managers learned BCF compliance views. Leadership learned executive query patterns.
Security and governance
Cloudflare Zero Trust gates every session. Entra ID SSO ties identity to the corporate directory. Audit logs capture every query and every document touched. Aligned to ISO 42001 controls for AI governance.
Tuning window
The first 30 days post-launch are reserved for tuning. Query patterns get observed, prompt scaffolds get refined, source weighting gets adjusted, and edge cases get hardened against the firm's real workflow.
How the Phase 1 math is scoped.
Phase 1 Math, Projected
- 20 user cohort × 4 hours per week recovered × $75 per hour blended rate × 50 weeks
- = approximately $300,000 in projected annual capacity recovery
- Validated against the Levine project: 3 missing change orders surfaced in seconds, ~40 percent data gap exposed
- Working software in 2 weeks against real data
For comparison, the Big 4 consultancy bid for discovery alone was $150K to $250K and 12 to 16 weeks, with no software produced during that window. The Heed discovery produced working software in two weeks against real records, at a fraction of the discovery cost. Final implementation scope and pricing sit with leadership for review.
The next phase is scoped around retiring legacy Salesforce licenses and the ancillary tools they support. The decision sits with leadership. The discovery surfaced enough validated gaps to make the conversation a quantitative one rather than a hopeful one.
The scope: from CRM migration to operations platform.
The proposed scope reframes the conversation from finishing a Salesforce migration to building the operations platform the firm actually needs. The replacement does everything Salesforce was supposed to do, plus everything Salesforce was never going to do.
What replaces Salesforce
- Project pipeline and stage management
- Opportunity tracking with forecast roll-up
- Account history and contact management
- Activity logging and team collaboration
What Salesforce was never going to do
- AI search across the full project history with citations
- Image processing of jobsite photos for retrofit progress and condition flagging
- Document intelligence on plan sets, permits, and soil reports
- Meeting transcript capture and project linkage
- BCF compliance dashboard with budget variance and permit aging
- Reasoning agents that answer engineering questions in plain English
One platform, all of it, secured under Cloudflare Zero Trust and Entra ID, governed by ISO 42001-aligned controls. Built specifically for how this firm actually operates, not how a generic CRM thinks it should.
The Heed Method
Every engagement starts with a POC against your real data.
Most AI projects fail not because the technology did not work, but because nobody checked whether they were worth building. We do not quote from a menu or open with a discovery deck. We build a working Proof of Concept against your actual data in two weeks, and we show you the result before you sign a production budget.
The POC proves the architecture, validates the integrations, and produces the ROI numbers. It is the reason this firm moved from a stalled migration to a scoped implementation conversation in two demos instead of two quarters.
Keep reading.
Companion case study
How an Encino estate and family law firm consolidated eight tools into one secure employee dashboard.
Read the case study →The offering
Custom apps and dashboards built around your team's actual workflow, not a CRM template.
See the offering →The platform
The Heed Operations Platform that replaces fragmented CRMs and disconnected tools.
Explore the platform →The ROI math
How to calculate whether an AI project is worth building before you commit a dollar.
Read the post →Case study library
Downloadable PDFs, lead magnets, and the full Heed resource library.
Browse resources →Talk to Heed
Bring us your tools list, your migration receipts, or your scattered SharePoint. We will tell you what is buildable.
Book a call →Mid-migration? Tired of Salesforce? Bring us your tools list.
Most operations problems are not CRM problems. They are knowledge problems wearing a CRM costume. A 20-minute discovery call costs nothing, often produces a rough ROI estimate before you hang up, and tells you whether a Heed POC would beat the next $300K of consulting.