An unsolicited proposal by Steps Ventures. Not affiliated with or endorsed by Aligned Climate Capital.
Chasing NDA signatures. Reformatting memo drafts. Market sizes nobody trusts. Stale CRM fields. Site diligence that takes days per site. LP impact reports assembled by hand each quarter. None of it is running at Aligned today. Deal ops is worked out end to end, one click away, and I walk through the rest live. Below: what each fix looks like, and honest build times for one person, a small team, and the whole fund. This is the kind of work a forward-deployed AI engineer, embedded with the fund, ships week after week.
Nothing to adopt and nothing to learn. Your CRM stays the system of record, the house memo format stays the format. The work just stops needing a person. One of the three, deal ops, is specified end to end on its own page. The other two and the live solar screen I walk through in a working session.
An NDA goes out and someone has to remember the rest: nudge the counterparty, file the signed copy, flip the stage, notice the data room went quiet. The fix: e-sign, data-room, email, and calendar signals update the CRM on their own, with a dashboard of current and historic flow, a provenance log on every write, and one-click intervention when anyone wants it. Nobody gets nudged for a thing they already sent.
Decks pasted into a chat tool come back shapeless, unsourced, and confidently padded, and an associate reformats them by hand. The fix: the house template encoded once, firmographics and contacts pulled live, every claim tagged to its source, open diligence items flagged rather than papered over. Any deck in, an IC memo out, the same shape every time.
Walked through in a working session.
In the demo market, same-year analyst estimates disagreed by 63 percent with scope never stated, and that pattern repeats wherever the IC debates the number instead of the deal. The fix: top-down and bottom-up side by side, every source graded for reliability, and the assumptions exposed as levers that move in the meeting.
Walked through in a working session.
A first pass on a site takes days, and the kill criterion tends to surface last. The demonstration scored five real, publicly announced 1 to 10 MWac projects from live public siting layers, about 20 seconds per site, on DC-basis capacity factors from NREL PVWatts. A worst-risk-first read, not a verdict. A pre-screen, not a bankability verdict: offtake, interconnection queue position, tax equity, and EPC remain data-room questions.
| Grade | Project | Developer | Location | +10% ITC | Solar CF |
|---|---|---|---|---|---|
| B | Glor Rd | PowerBank (fka SolarBank) | Attica area, Wyoming Co., NY | Yes | 13.2% |
| B | North Main | PowerBank (fka SolarBank) | Warsaw area, Wyoming Co., NY | Yes | 13.1% |
| B | Yarmouth | New Leaf Energy | Yarmouth, Cumberland Co., ME | No | 15.1% |
| B | Manlius landfill | PowerBank (fka SolarBank) | Manlius, Onondaga Co., NY | No | 13.4% |
| B | Rock Cut | Trajectory / Summit Ridge | Rockford, Winnebago Co., IL | No | 15.5% |
The two highest-CF sites both miss the IRA energy-community adder, a single geospatial fact that moves the return. The adder and a P90 haircut are toggles in the model. Locations are geolocated from public announcements and indicative until real parcel coordinates are used.
The impact report for foundation LPs gets stitched together by hand every quarter, from spreadsheets that disagree. The demonstration computes it instead: estimated annual impact of the five screened projects, from public data, by the same engine that screens the deals. Pointed at the operating fleet, it would run on metered production, mapped to EDCI, on whatever cadence the LPs want.
Every estimate below is padded, because the right way to run this is small promises and early delivery. One person means a single analyst or asset manager using it day to day. Small team means the venture desk or the ASP desk. Fund-wide means everyone, under your SSO, with the audit log.
| The fix | One person | Small team | Fund-wide |
|---|---|---|---|
| IC memos in the house format | 1 week | 2 to 3 weeks | 4 to 6 weeks |
| Market-sizing harness | 1 week | 2 weeks | 3 to 4 weeks |
| Site developability screen | 1 week | 2 to 3 weeks | 4 to 6 weeks |
| Deal ops on the CRM | 2 weeks | 3 to 5 weeks | 6 to 10 weeks |
| Data-room classification and extraction | 2 to 3 weeks | 4 to 6 weeks | 8 to 12 weeks |
| LP impact on metered production | 3 to 4 weeks | 4 to 6 weeks | 6 to 10 weeks |
Estimates assume read-only access arrives in week one, and they include the unglamorous parts: shadow-mode baselining, security review, permissions that mirror the CRM, and a provenance log on every write. In most rows the first useful output lands inside two weeks. Hours saved get measured against the shadow-mode baseline, not modeled.
None of this is a product to buy or a platform to migrate to. It is a way of working: a forward-deployed AI engineer embedded with the fund, who takes one nuisance at a time, wires it to the real systems, measures the result, and moves to the next. The demonstrations on this page came together from public data in days. Inside the walls, with real access, the cadence is a fix landing every few weeks.
Single tenant, your SSO, your cloud account, an audit log on every write. Data never trains a shared engine and is never reused for anyone else. Information barrier in writing. The shared piece is method, never your data or your deals.
Shadow mode first, a measured baseline, one signal wired end to end before the next is touched. Anything outbound can be held for approval per signal until it has earned trust. Nothing is declared done until it runs on your real deal flow.
A scoped first build: one nuisance, wired to your systems, fixed fee against a week-one baseline. We pick it in one conversation. No pilot theater, no platform to buy, no seat to fill.