Hartwell & Associates is a composite case study representing Phos AI Labs engineering consultancy engagement patterns. Identifying details have been changed.
Hartwell & Associates, a 34-person civil and structural engineering consultancy at $22M annual revenue, began their AI implementation in November 2025 with a clear objective: reduce proposal time so principals can bill more hours.
Eleven months later, the objective was met. The average proposal had gone from 11.2 principal hours to 4.4 hours.
The win rate had increased by 12 percentage points. The firm was pursuing 40% more proposals per year than it had been pursuing at implementation start.
But the path from week one to month eleven was not a straight line. And the detours are the most valuable part of the story.
This article describes the Hartwell & Associates implementation month by month: what was built, what was wrong, what was fixed, and what the results looked like at month three, month six, and month eleven.
The goal is not to present an idealised case but to show what a realistic implementation looks like for a firm at this scale and discipline mix.
For context on why firms like this are falling behind competitively, see why professional services firms lose proposals to AI-enabled competitors.
The firm — profile and context
Hartwell & Associates, November 2025
Size and structure:
- 34 professionals: 6 principals and associates, 12 project engineers (PE-licensed or working toward licensure), 8 designers and technicians, 5 administrative and project coordination, 3 principal-level staff (managing principal, project principal, BD principal)
- Revenue: $22M annual, split approximately 60% civil engineering (site development, drainage, infrastructure) and 40% structural engineering (building structures, bridge inspection, specialty structures)
Client base:
- Municipal public works agencies (35%)
- Private developers (30%)
- Educational and institutional owners (20%)
- Industrial and commercial (15%)
Proposal metrics at implementation start:
| Metric | Pre-implementation baseline |
|---|---|
| Annual proposal volume | 55 to 65 competitive proposals per year |
| Average proposal time | 11.2 principal hours + 6.3 BD coordinator hours = 17.5 hours per proposal |
| Win rate (all proposals) | 31% |
| Win rate (fully competitive, 4+ firms) | 21% |
| Proposals declined for capacity reasons | ~20 to 25 per year |
The specific problem
The firm was pursuing approximately 45 proposals per year and declining an estimated 20 to 25 additional opportunities per year because the principal team did not have the capacity to produce more proposals without significantly reducing billable hours.
The managing principal’s assessment before implementation:
“We are good engineers. Our work is well-regarded. We lose proposals because we do not have the time to write them as well as they deserve to be written. We are a firm of good technical minds who write proposals like it is a distraction from the real work, because it is.”
Months 1 to 2: Building the foundation
Week 1 to 2: The context pack build
The structured context pack interviews were conducted across six sessions over two weeks.
| Session | Attendees | Content | Duration |
|---|---|---|---|
| 1 | Managing principal | Capabilities matrix (civil) and engagement framing | 90 minutes |
| 2 | Managing principal | Capabilities matrix (structural) | 90 minutes |
| 3 | Structural practice lead | Structural vocabulary and project types | 90 minutes |
| 4 | Civil practice lead | Civil vocabulary and project types | 90 minutes |
| 5 | BD principal | Proposal writing standards, fee presentation | 90 minutes |
| 6 | BD coordinator | Client communication, proposal structure | 60 minutes |
Total interview time: 9.5 hours across two weeks. The first draft context pack was completed by the end of week two.
The problem with the first draft context pack
The capabilities matrix described Hartwell & Associates’ capabilities accurately, but at the wrong level of specificity for the firm’s primary client types.
The structural engineering section described capabilities in general structural engineering vocabulary (gravity load design, lateral system design, foundation design, structural assessment). The firm’s structural work, however, was primarily for two specific client types.
| Client type | What they actually want to know |
|---|---|
| Private developer (mid-rise multifamily) | “Have you done this building type, this many stories, this construction method, in this market?” |
| Public school district (K-12) | “Have you done school buildings in this state? Do you know DSA requirements? Have you done seismic retrofits on occupied facilities?” |
The first draft capabilities matrix answered neither of these questions. It answered the general question.
The revision
The structural practice lead and the BD principal spent a focused two hours rewriting the capabilities matrix with two client-type-specific capability sections.
- Developer section: building type, height, construction method, delivery method experience
- Institutional section: occupancy type, regulatory framework experience, DSA knowledge, seismic assessment experience
The revised capabilities matrix was loaded into the context pack on day 18.
The project portfolio library sprint
The BD coordinator blocked four days to build the project portfolio library. The plan was two days.
What was discovered: the firm had no consistent internal record of project details in a usable format. The CRM had project names and clients. The network drive had proposal packages, some complete, some incomplete. The principals’ personal files had their own reference project descriptions, inconsistent in format and completeness.
The sprint:
| Day | Activity | Result |
|---|---|---|
| 1 | Identified 75 candidate projects; eliminated 20 (too old, too small, or not used as references) | 55 projects to enter |
| 2 | Interviewed three principals for structured descriptions of 15 highest-priority projects (45 minutes per session) | 15 high-priority entries complete |
| 3 | Entered the 15 high-priority entries; began researching remaining 40 from available documentation | 15 entries in library |
| 4 | Completed 35 of the remaining 40; flagged 5 requiring additional principal input | 50 entries in library |
The BD coordinator’s assessment after the sprint:
“I’ve worked here for seven years and this is the first time we have had a complete picture of what we’ve built. It’s embarrassing and also kind of useful to finally have it.”
What the library sprint revealed
Three significant gaps in the firm’s reference project record:
- The firm had done 12 K-12 educational projects — a significant body of work — that had never been systematically tracked by school type, grade level, or state regulatory framework
- Three of the firm’s most significant structural projects had never been entered into any client-facing reference document because the principals had used them verbally in presentations and never captured them in writing
- The civil engineering reference library had no stormwater master planning projects documented, despite the firm having completed six of them, because stormwater work had been systematically under-represented in proposals as less prestigious than infrastructure work
The library sprint fixed all three gaps.
Months 2 to 4: First proposals and early calibration
The first AI-assisted proposal (week 5)
The project: a 35,000 SF municipal recreation center for a city in the firm’s primary market area. Mixed civil and structural scope. A procurement the firm had pursued twice before and lost both times.
Section-by-section evaluation:
Project understanding section: strong. The AI correctly identified the site constraints from the scope document summary the project engineer provided and drafted a project understanding section the BD principal described as “better than what we usually produce.”
Qualifications section: weak. The AI selected three reference projects from the library, but two of them were for private developers, not municipal clients. The municipal recreation center client was looking for public sector experience.
The fix: added a client type keyword tag to the portfolio library entries (public sector municipal / public sector institutional / private developer / industrial) and added a client type input field to the qualifications workflow. Problem resolved in 45 minutes.
Technical approach section: mixed. The project engineer provided the methodology framework and the AI drafted the narrative. The BD principal described the first draft as “technically accurate but written for a structural engineering audience, not a municipal public works audience.”
The fix: added a “client technical vocabulary level” field to the workflow inputs.
| Input option | AI output |
|---|---|
| Technical peer (professional engineers) | Full technical vocabulary, code and standard references |
| Owner or manager (non-PE technical) | Plain language with selective technical terms |
| Public or institutional (non-technical) | Business language, outcomes-focused, minimal jargon |
Problem resolved in 30 minutes after deployment.
First proposal outcome: the firm was shortlisted for the first time in three prior attempts. They did not win (awarded to the incumbent), but feedback from the procurement coordinator mentioned their project understanding was “notably specific.”
Month 3: The first same-day submission
A drainage improvement project for a repeat municipal client, an RFP received on Monday morning. In previous years: the proposal went out Thursday. This year: submitted Tuesday afternoon.
The BD principal’s note in the project log:
“First time in my career I have been ahead of an RFP deadline for a competitive proposal. Not just on time: ahead of it.”
Month 4: The quality calibration milestone
By month four, three patterns were visible in the adoption tracking log:
| Pattern | Cause | Fix |
|---|---|---|
| Civil engineering scopes requiring less editing than structural | Civil practice lead’s context pack entries were more detailed from session 4 | Structural practice lead session added in month 5 |
| Project understanding sections consistently the strongest section | Most carefully calibrated in month 1 | Maintained |
| Fee presentation sections the weakest | AI fee narratives not calibrated to different procurement methods | Three-hour session with BD principal; procurement-method-specific fee guide built |
Month 4 metrics:
| Metric | Pre-implementation | Month 4 |
|---|---|---|
| Average principal proposal time | 11.2 hours | 5.8 hours |
| Submission timing | Last 30% of window | First 45% of window |
| Proposals pursued per year (annualised) | 55 | 68 |
| Win rate (all proposals) | 31% | 34% |
Months 5 to 11: Compounding results
Month 5 to 6: The structural engineering catch-up
The structural practice lead blocked a full afternoon for a context pack revision session. The output:
- Expanded structural capabilities matrix with DSA-specific experience section and seismic assessment project references
- New structural project portfolio entries for the three uncaptured major projects
- Structural technical approach narrative standards: a 200-word guide to how the structural practice describes its engineering approach to institutional clients
Result: structural engineering proposal sections improved by an estimated 30% in quality rating by the BD principal at month six.
Month 7: The win rate second wave
A 12-month analysis of proposal outcomes at month seven showed the win rate improvement breaking into two waves:
Wave 1 (months 2 to 4): win rate improvement primarily attributable to earlier submission timing. Competitive evaluations where the firm submitted in the first 25% of the window: won at 40%. Previous rate: approximately 31%.
Wave 2 (months 5 to 7): win rate improvement from qualifications matching. The portfolio library — now with client-type tags and the three previously uncaptured structural projects — was producing more specifically relevant reference project selections.
Month 8: The staff retention discovery
At the month-eight team check-in, two senior project engineers made unprompted comments about the AI implementation.
Engineer 1: “I feel like I’m finally doing engineering instead of writing about engineering.”
Engineer 2: “The proposal work used to take my whole Friday and I’d go home thinking about the budget I hadn’t billed. Now I finish the proposal section by noon and bill the afternoon. It’s a different job.”
The managing principal confirmed in a separate conversation that both engineers had been fielding recruitment calls. Neither had disclosed this to the principal group. Both had declined the calls.
“We haven’t had turnover in our PE group in two years. Part of that is compensation. Part of it is that the job got better.”
Month 11: Final metrics
| Metric | Pre-implementation | Month 11 | Change |
|---|---|---|---|
| Average principal proposal hours | 11.2 hrs | 4.4 hrs | -61% |
| Proposals pursued per year | 55 | 77 | +40% |
| Proposals submitted in first 25% of window | 15% | 58% | +43pp |
| Win rate (all proposals) | 31% | 43% | +12pp |
| Win rate (fully competitive, 4+ firms) | 21% | 35% | +14pp |
The financial impact
| Source | Amount |
|---|---|
| Additional wins from improved win rate (~8 to 9 additional contracts × $320K average contract value) | ~$2.7M additional annual revenue |
| Recovered principal billing capacity (524 hours × $175/hour billing rate) | $91,700 per year |
| Total first-year financial impact | ~$2.79M |
| Total investment (engagement fee + tool costs) | $42,475 |
| ROI | ~65× |
What the managing principal would do differently
”We would have built the project portfolio library before the context pack.”
“The context pack was built first because that is how the Phos AI Labs team structured it. But when we started running proposals and the reference project selection was poor, we fixed the context pack.
“If we had built the library first, the context pack calibration would have been better from the start because we would have known what our actual project base looked like.”
The lesson: for professional services firms with a significant reference project base, building the project portfolio library before or simultaneously with the context pack produces better calibration in the first proposals.
”We would have invested more of the structural practice lead’s time in the first two weeks.”
“The structural practice lead is the busiest person in the firm. We scheduled him for two sessions rather than four because we were being respectful of his time.
“The result was four months of structural proposals that required more editing than they should have.”
The lesson: the input investment required from the busiest principal is worth protecting.
The engagement partner who says “I only need four hours of your time” is producing a four-hour context pack. The one who says “I need eight hours of your time to produce something that works” is producing a different result.
”We would have started tracking proposal outcomes from month one.”
“We did not have a structured proposal outcome tracking system before the implementation. We could not compare month-eleven win rates to pre-implementation win rates with confidence because the pre-implementation data was incomplete.
“We had to reconstruct the baseline from CRM records and email search. It took a week.”
The lesson: start the proposal outcome tracking log on the same day the implementation begins. Three fields are sufficient to start: proposal ID, submission date, and outcome (win / shortlist / no award / declined).
Common questions on the Hartwell & Associates implementation
”How does the Hartwell model translate to a smaller firm (10 people vs. 34)?”
The model scales down well. A 10-person engineering firm typically has:
- One to two principals managing all proposals (rather than three)
- A smaller project portfolio library (20 to 30 entries vs. 50)
- One practice area rather than two (civil or structural, not both)
The implementation is faster (six weeks vs. eleven to reach full deployment), the context pack build requires fewer sessions (three instead of six), and the project portfolio library sprint takes two days instead of four.
The time recovery and win rate improvement are proportionally similar.
”What happened to the proposals where AI did not help?”
At month eleven, approximately 15% of proposals still required significant principal time beyond the AI-assisted baseline.
These were universally the firm’s most technically complex proposals: large public infrastructure projects, proposals requiring the managing principal’s specific technical judgment, and projects where the client’s requirements fell outside the capabilities matrix’s calibrated range.
For these proposals: AI assisted with the sections it performs well (project understanding, qualifications, standard technical sections) and the managing principal wrote the technically differentiating sections directly. The AI saved time even on these proposals, just less than the 61% average.
”Did the client comment on the quality of the AI-assisted proposals compared to previous submissions?”
The procurement coordinator for the municipal recreation center project mentioned that the project understanding was “notably specific.” No other client feedback specifically referenced proposal quality.
The absence of negative feedback is the baseline expectation. The Foundation-loaded proposal reads as the firm’s best work, not as AI output. Clients evaluating proposals do not distinguish between a manually written proposal and a Foundation-loaded AI-assisted proposal. They distinguish between a specific, relevant, well-structured proposal and one that is not.
Ready to build the proposal system that Hartwell & Associates built, with the calibration mistakes avoided from the start?
The Hartwell & Associates implementation produced a 61% reduction in principal proposal time and a 12-percentage-point win rate improvement over eleven months.
The investment: $42,000 in engagement fees plus $475 per month in tool costs. The return: $2.79M in the first year.
The path was not straight. The first context pack was wrong. The project portfolio library build took twice as long as planned. The structural practice was undercalibrated for four months.
The lessons are specific: build the project portfolio library before or alongside the context pack, invest the principal’s time fully, and start tracking outcomes from day one.
Path one: start the proposal outcome tracking log today. Three fields: proposal ID, submission date, outcome. Track every competitive proposal from this point forward. In six months, you will have a baseline that makes the ROI of any AI implementation measurable rather than estimated.
Path two: bring in a partner. We have run 400+ AI engagements. Clients include Zapier, Coca-Cola, Medtronic, Dataiku, and American Express. Phos AI Labs applies the Hartwell & Associates lessons to the next engineering consultancy’s build sequence: project portfolio library built alongside the context pack, the busiest practice lead’s time fully invested in Phase 1, and proposal outcome tracking started from week one. Thirty minutes, no deck. Start here
Related articles
- Why AI Tools Don't Stick in Non-Tech Companies
- Can AI Handle Industry-Specific Integrations?
- How to Give AI the Right Context About Your Business
- The Four Phases of a Phos AI Labs Engagement
- How to Share AI Memory Across Different Models
- The Operations Workflows in Your Distribution Business Most Ready for AI Right Now