One canonical model, every sector
Not a model per sector. Claude proposes source-field to canonical-entity mappings, a person confirms them, and the mapping is reused at the next company.
Private equity · Where sector expertise compounds
Turn sector expertise into something that compounds
Alpha Engine carries a private equity firm from sourcing a company through to proving what the firm's own operating expertise was worth. It researches and ranks targets against a mandate, builds an ontology of each business as diligence hardens it, matches the firm's solution packs to the companies they fit, and then separates the uplift the firm created from the market movement it happened to ride.
Status. In development. We are looking for two private equity firms to work with as design partners.
The business problem
A firm's operating expertise lives in the heads of its partners and dies when they leave.
Sector-focused firms win because they have seen the same problem twenty times. But the playbook exists as experience rather than as an asset — it is not written down in a form that can be matched to a new portfolio company, applied consistently, or measured afterwards.
So the same diagnosis is made from scratch at every company, the same interventions are rebuilt by whoever happens to be running the account, and nobody can say which of them actually worked.
And when an exit goes well, the firm cannot separate what it did from what the sector did anyway. Alpha and beta arrive in the same number, which makes the firm's own contribution a matter of assertion.
The platform
Five stages, one ontology underneath
Positioning
What makes it different
The ontology hardens as you learn
Inferred from public sources at sourcing, tested through the deal levers, hardened by diligence answers, then measured directly through connectors into ERP, CRM and HR.
Playbooks become matchable assets
A solution pack is structured against the ontology, so the system can tell you which portfolio companies it actually fits rather than leaving that to recall.
Alpha separated from beta
Uplift attributable to a deployed pack, held apart from sector and market movement. The firm's contribution becomes a figure rather than a claim.
Email as topology, not content
Metadata only — sender, recipient, timestamp, thread. You get the shape of how a business communicates without reading anyone's mail.
01
Sourcing and research — From a mandate to a ranked shortlist
The analyst's front end. A deal brief becomes structured search criteria, the search returns ranked candidates, and each candidate opens into an evidenced view the analyst can act on.
The problem. Sourcing against a sector mandate is manual, inconsistent between analysts, and leaves no record of why a company was passed over. The rejected pile is where the learning is, and it is thrown away.
| Functionality | What problem it solves | Impact on the business |
|---|---|---|
| Three ways inPaste an existing deal brief, build one conversationally, or skip it and set criteria directly. | An analyst with no brief has nowhere to start; one with a brief retypes it as filters. | The briefing becomes the search, and the analyst refines criteria rather than inventing them. |
| Criteria you can argue withSector, geography, revenue and EBITDA band, ownership type, growth signals, deal levers — each optional and editable. | A black-box search returns a list nobody can challenge or tune. | The analyst sees and adjusts the reasoning before anything is spent running it. |
| Two-stage searchA cheap preliminary pass returns a rough-fit list of a size the analyst chooses; the deeper pass runs only on the set they keep. | Deep research on fifty companies costs a great deal to discover that forty are wrong. | Research spend lands on the companies that survived a human filter. |
| Evidenced drill-inFit score with criterion-by-criterion reasoning, financial snapshot with confidence, ownership and deal context, value creation thesis, risk flags, comparables, SWOT. | A ranked list without evidence is a ranking nobody will defend to a partner. | Every score opens into what it was derived from. |
| Screening memoA generated memo with the drill-in, the justification and an estimate of diligence cost, sent to the deal partner to approve, reject or return with questions. | Moving a company forward depends on catching a partner at the right moment. | Approval opens the diligence workspace and loops in the operating partner automatically. |
Business value
- Rejections recorded with reasons, not lost
- Research spend aimed by a human filter
- The partner decision leaves an auditable trail
How it connects
Sourcing produces the first, loosest version of a company's ontology. Diligence hardens it, and the value creation thesis written here is what measurement later tests.
02
Ontology and diligence — A model of the business that gets more honest over time
A graph of each company — what it does, how it is organised, what it runs on — built from public research, tested through diligence, then measured directly once connectors are live. Every element carries a confidence level and a source.
The problem. Diligence findings end up in documents nobody reads again. The understanding built at enormous cost during a deal is not available in any usable form on day one of ownership, so the operating team starts over.
| Functionality | What problem it solves | Impact on the business |
|---|---|---|
| Confidence that risesEach element moves from inferred to confirmed as evidence arrives, and the view shows which is which. | A model built from public sources looks identical to one built from verified data. | Nobody acts on a guess believing it was a finding. |
| Gaps and contradictions, listedOpen gaps are tracked, and answers that contradict the public picture are surfaced rather than silently overwritten. | Diligence contradictions are noticed by whoever happens to read both documents. | Closing gaps becomes shared, assignable work between analyst and partner. |
| Three ways to answerThe target answers in the portal, the deal team uploads data room documents for extraction, or a consultant conducts the analysis and enters findings. | Diligence processes assume a cooperation level the target may not offer. | The process adapts to the deal rather than the deal to the process. |
| Connectors that measureOracle Fusion, SAP, Salesforce, Workday and Dynamics 365 through live read-only APIs, or periodic extract ingestion where a live connection is refused. | What a company says it does and what its systems show it doing are different things. | The ontology stops being a description and becomes a measurement. |
| Email metadata onlySender, recipient, timestamp and thread through Microsoft Graph. No content. | Communication topology is revealing, and reading employee email to get it is not acceptable. | You see how the business actually communicates without the privacy exposure. |
Business value
- Diligence understanding survives into ownership
- Claims separated from verified facts
- Integration refusal does not stop measurement
How it connects
The ontology is what solution packs are matched against and what the early warning system watches. Everything downstream is only as good as this.
03
Solution packs — The firm's playbooks, written down and matched
A library of operating playbooks — approaches to a particular business challenge, or to raising a particular kind of maturity. Each is structured against the canonical model and the maturity model, so the system can match it to the companies it fits.
The problem. Operating expertise is applied by whoever remembers it. There is no library, no way to tell which company a playbook suits, and no record of what it was supposed to achieve.
| Functionality | What problem it solves | Impact on the business |
|---|---|---|
| Three sourcesRed Bear-authored generic packs, packs the firm writes itself with conversational help or a document template, and packs requested from the Red Bear account team with a brief. | A library that only the vendor can fill never reflects the firm's own edge. | The firm's expertise goes in alongside ours, in the firm's own words. |
| Senior review before liveEvery pack, whatever its source, is reviewed and approved by a senior partner, who can interrogate it conversationally first. | An unreviewed playbook applied across a portfolio is a liability. | Nothing reaches an analyst without a partner having read it. |
| Matched, not rememberedPacks are evaluated against a company's ontology to show where they would add value. | Matching a playbook to a company depends on who is in the room. | The fit is proposed from the model and argued with by a person. |
| Applied with a predictionOn application the system proposes a KPI baseline from connected data, the analyst confirms it, sets dates and owners, and the predicted uplift is recorded. | Interventions start without a baseline, so afterwards there is nothing to measure against. | Predicted against realised becomes answerable, because the prediction was written down first. |
Business value
- Expertise held as an asset rather than as memory
- Fit proposed from evidence
- Every deployment carries a prediction to test
How it connects
Packs are matched against the ontology and measured by attribution. A pack that never produces measured uplift is a pack the library should lose.
04
Measurement and attribution — Separate what you did from what the market did
Continuous measurement of deployed packs against their recorded baselines, with sector and market movement held apart from the uplift attributable to the intervention.
The problem. A portfolio company improves during a bull run in its sector. Some of that is the firm; some is the tide. Reported as one number, the firm's actual contribution is unprovable — to an LP, to a prospective investor, or to itself.
| Functionality | What problem it solves | Impact on the business |
|---|---|---|
| Baseline before interventionThe KPI baseline is proposed from connected data and confirmed before the pack goes live. | Retrospective baselines are chosen, consciously or not, to flatter the result. | The comparison is fixed before anyone knows how it turns out. |
| Alpha held apart from betaSector and market movement are modelled separately from pack-attributable uplift. | A rising sector makes every intervention look effective. | The firm can state its own contribution and show the working. |
| Predicted against realisedThe uplift recorded at deployment is compared with what actually happened, and the gap feeds learning. | Nobody revisits what an intervention was supposed to achieve. | The library improves because its predictions are scored. |
| Reporting built on itPower BI over Microsoft Fabric, including an LP quarterly export carrying the attribution. | Attribution that exists only inside the tool never reaches the people who need it. | The measurement arrives in the reporting the firm already runs on. |
Business value
- A defensible answer to what the firm was worth
- Pack library scored by outcome
- LP reporting carrying real attribution
How it connects
Measurement scores the packs, and feeds the predictive learning that improves the next match.
05
Early warning — Know before the board pack tells you
Monitoring across connected operational data and external signals — news, market feeds, social posts — ranked by materiality against the company's ontology, with a triage view and a suggested action.
The problem. Portfolio deterioration surfaces at the quarterly review, which is between one and three months after it started. The window in which it was cheap to fix has usually closed.
| Functionality | What problem it solves | Impact on the business |
|---|---|---|
| Internal and external signalsConnected ERP, CRM and HR data alongside news, stock and social feeds. | Operational problems and reputational ones arrive through different doors and are watched by different people. | One watch across both, on the same company. |
| Materiality ranked against the ontologyThresholds are ontology-aware, so a signal is weighted by what it touches in that specific business. | Generic thresholds produce alert fatigue, and alert fatigue produces ignored alerts. | Fewer alerts, each of which is about something that matters here. |
| Triage, not a notificationThe alert opens a view showing the signal, its materiality, the evidence, what it touches in the ontology, and a suggested action — deploy a pack, raise a question, or dismiss with a rationale. | An alert that says only that something happened leaves the work undone. | The response starts with a proposal rather than a blank page. |
Business value
- Deterioration visible between board packs
- Alerts weighted by this business, not a generic rule
- Dismissals recorded with reasons
How it connects
Early warning reads the ontology and the measurement stream, and its most common suggested action is to deploy a pack.
What it looks like
The screens, drawn to the current specification
Drawn from the specification rather than captured from a running build, and each is labelled as such in its own frame. Showing you a prototype screen built to a superseded model would be worse than showing you a drawing of the right one.
Finding and assessing a company
Deal brief → search criteriaConceptual interface
Brief, as pasted
“Looking for founder-owned industrial distributors in the Gulf, 150 to 600 million, where we can apply the procurement playbook we used at Hadid.”
Criteria proposed — all optional, all editable
SectorIndustrial distribution · MROedit
GeographyGCC, with UK bolt-on potentialedit
RevenueAED 150m – 600medit
EBITDA margin8% and aboveedit
OwnershipFounder-owned · family holdingedit
Deal leversProcurement · pricing discipline · branch networkedit
A preliminary pass returns a rough-fit list of a size you choose. The deeper research runs only on what survives your filter.
Candidate · Gulf Industrial Supply CoConceptual interface
78
Fit against mandate
Criterion by criterion, with the evidence behind each
| Criterion | Assessment | Evidence | Status |
|---|---|---|---|
| Sector fit | Strong | SIC and revenue mix from filings, 2026 | verified |
| Revenue AED 310m | In band | Audited accounts FY25 | verified |
| EBITDA 11.2% | Above threshold | Derived from filings — see working | inferred |
| Founder-owned | Yes | Shareholding register, Mar 2026 | verified |
| Procurement lever | Likely | Supplier concentration, 3 sources disagree | conflict |
ShortlistReject with reasonDeeper researchShare with partner
Building a picture of the business
Ontology · Gulf Industrial Supply CoConceptual interface
ProcurementConfirmedDiligence answer + ERP read
Branch operationsConsultedManagement interview, not yet measured
Pricing governanceInferredPublic sources only — open gap
IT estateContradictionAnswer says 4 systems; connector shows 9
Open gaps are assignable between analyst and partner. Closing them is shared work, not a report somebody files.
Applying expertise, and proving what it was worth
Apply pack · Procurement consolidation v4Conceptual interface
Baseline — proposed from connected dataDirect spend AED 214m · ERP, 11 Oct 2026
Baseline — supplier countNo connected source — recorded as a gap
Predicted uplift180–320 bps on direct spend, over 18 months
Owner at the companyHead of procurement — named
Sponsor at the firmOperating partner — named
Frozen at application. Later revisions keep every prior version, so the comparison is against what was actually believed at the time.
Attribution · Procurement consolidationConceptual interface
Observed improvement410 bps
Pack · 230 bpsSector · 180 bps
Pack-attributable: 150–310 bpsSector and market: 120–240 bps
Predicted at application180–320 bpsRealised, pack-attributable150–310 bps
Predicted against realised, over the same period, using the prediction as recorded at application.
Evidence
What a design partner measures
Alpha Engine has no published results, because it has no customers yet. These are the measurements the first engagement produces, each from the firm's own portfolio against a baseline fixed before any intervention starts.
| Measure | How it is produced |
|---|---|
| Attribution split | Uplift attributable to deployed packs, held apart from sector and market movement |
| Predicted against realised | The gap between the uplift recorded at deployment and what actually happened |
| Sourcing funnel | Candidates researched, shortlisted, taken to screening and converted, with rejection reasons |
| Ontology confidence | Proportion of each portfolio company's model confirmed rather than inferred |
No values are quoted here because none have been earned yet. We publish no benchmark we cannot reproduce from a live engagement.
Read further
10 documents on this product
Every one is a working draft, and says so. No form, no email address — the link downloads the PDF.
- Data sheet: Suite overviewINV-DS-00 · v3.0
- Data sheet: FindINV-DS-01 · v2.1
- Data sheet: PredictINV-DS-02 · v2.1
- Data sheet: ValidateINV-DS-03 · v2.1
- Data sheet: DecideINV-DS-04 · v2.1
- Data sheet: OwnINV-DS-05 · v2.1
- Data sheet: LearnINV-DS-06 · v2.1
- Data sheet: ProtectINV-DS-07 · v2.1
- Data sheet: Solution packs and maturityINV-DS-08 · v1.0
- Data sheet: Human-in-the-loop review and predictive learningINV-DS-09 · v1.0
Next step
Two design partners, roughly six months to a working platform
We are looking for two sector-focused private equity firms to build this with. A design partner gets the platform configured to their canonical model and their own packs, and the roadmap shaped around what their portfolio actually needs.
Your packs, not ours
The generic library is a starting point. The work is getting the firm's own expertise into a form the system can match and measure.
One portfolio company first
Connectors, ontology and a single pack deployment with a confirmed baseline, before any breadth.
Integration is the project
Not just systems: process, culture and training integration alongside them. That is where these programmes succeed or fail.
Everything reusable comes back
Each engagement puts templates back into the shared library, so the next one costs less and this does not drift into consultancy.