How many scoring dimensions should a model have?

Enough to separate the questions, few enough to explain a score in one breath. Misano uses five — market, financials, moat, management, risk — plus strategy fit as a separate axis. Past seven dimensions the weights stop meaning anything.

Do I need special software for a deal pipeline, or is a spreadsheet enough?

A spreadsheet works up to a few dozen live companies and one user. It breaks where the value is: consistent scoring across hundreds of companies, automatic registry data, rescoring old passes when the thesis moves, and an audit trail of why each no was a no. That gap is what pipeline tooling — Misano included — actually sells.

How do I stop bad data from producing confident scores?

Gate on data quality explicitly: extraction guards (unit scale, decimal commas, impossible margins), a data-completeness signal on each company, and hard gates that block scoring when critical inputs are missing. A confident score on garbage inputs is worse than no score.

What should happen to companies I reject?

They go to a watching stage with the reason for the pass recorded — not into a folder. Sectors and companies move; a periodic rescore of the watching list against the current strategy version is the cheapest source of proprietary dealflow most funds ignore.

Running a deal pipeline: sourcing, screening, and scoring against your thesis

Three failure modes quietly cost small funds the deals they should have done: the screening tax (a full workday scoring a company that didn't fit — you knew after an hour), curated inbound (your dealflow is what sellers, bankers and advisors chose to show you), and the buried no (the company you passed on two years ago has grown into your thesis, and nothing reminds you). A pipeline is the system that fixes all three.

Stages: make the funnel explicit

01Universe — everything that could conceivably fit: registry sweeps, web search, inbound, conference lists. No judgement yet, just capture with a source tag.
02Analysed — the company has been through the scoring model once. Most of the universe dies here, cheaply.
03Watching — doesn't fit today, plausibly fits later. This is where buried no's go to stay alive.
04Shortlist — actively pursued: management contact, first data requests.
05Due diligence — data room open, advisors engaged, real money on the line.
06Portfolio — owned, and now a monitoring problem rather than a sourcing one.

Scoring: derive the model from the strategy, then stop negotiating with it

A scoring model is only useful if it is written down before you see the company. Misano derives it from the fund strategy itself: upload the strategy document or a 200–10,000 character brief, and the AI extracts target sectors, anti-patterns, and a five-dimension scoring model calibrated to the thesis — market, financials, competitive moat, management, and risk, composited into a 0–100 score, with strategy fit rated as a separate axis rather than blended in.

Two design choices matter more than the dimension list. First, separate what math computes from what judgement assesses: financial ratios and risk gates are deterministic code, market attractiveness and moat are AI judgement with cited evidence — and the score breakdown shows which is which. Second, hard gates: a company below a floor on any critical dimension cannot compensate with charm elsewhere. A composite score without gates is how a great story with terrible unit economics reaches your shortlist.

The discipline that compounds

Score everything with the same model. The value of a 0–100 score is ranking a hundred companies against each other; ad-hoc criteria destroy exactly that.
Version the strategy. When the thesis changes, re-derive the model — and note which scores were produced by which version.
Rescore the pile of no's. Companies grow, sectors move, your view moves. Misano rescores the watching list against the current thesis on demand — your old 'no' isn't always still a no.
Meter the funnel. Companies analysed, share filtered out, shortlist conversion — Misano publishes its own live funnel at misano.ai/statistics as an honesty exercise.