The American Journal of Management and Economics Innovations
46
https://www.theamericanjournals.com/index.php/tajmei
TYPE
Original Research
PAGE NO.
46-53
10.37547/tajmei/Volume07Issue08-05
OPEN ACCESS
SUBMITTED
22 July 2025
ACCEPTED
27 July 2025
PUBLISHED
12 August 2025
VOLUME
Vol.07 Issue 08 2025
CITATION
Anna Mastykina. (2025). Algorithmic Identification of Relevant Investors
Using Machine Learning. The American Journal of Management and
Economics Innovations, 7(8), 46
–
53.
https://doi.org/10.37547/tajmei/Volume07Issue08-05
COPYRIGHT
© 2025 Original content from this work may be used under the terms
of the creative commons attributes 4.0 License.
Algorithmic Identification
of Relevant Investors
Using Machine Learning
Anna Mastykina
Founder, Taskinfinity.com, FundWise LLC Buenos Aires, Argentina
Abstract:
In this article, the problem of the low
efficiency of traditional cold communications with
venture capital funds is examined. The relevance of the
study is determined by the need to develop automated
tools for targeted search of relevant investors capable of
overcoming the limitations of warm recommendations
and expanding access to capital for startup teams
without an extensive network. The aim of the paper is to
demonstrate an algorithmic approach based on machine
learning methods for identifying relevant investors and
to investigate the integration of ML ranking with a
disciplined multistep-outreach strategy. The novelty lies
in the use of a multilayer feature architecture combining
an investment graph, thematic embeddings, soft signals
from public channels, and dynamic indicators of fund
activity, as well as in the construction of a controlled
cycle of cold communications with two follow-ups in
each three-day window. The obtained results confirm an
increase in the efficiency of the cold channel:
algorithmic selection enabled maintaining an open rate
at the level of 74
–
80%, a reply rate in the range of 10
–
17%, and provided 96 scheduled calls per quarter
without a single warm recommendation. The integration
of the ML ranking model with a structured cadence
strategy increases the controllability of the process,
turning fundraising from a lottery into a repeatable
business process with continuous model learning on
feedback data. Practical implementation includes not
only the development of an investor ranking model but
also the creation of infrastructure for large-scale
mailings: configuration of mail domains, optimization of
message templates, A/B-testing, and integration with
meeting-scheduling tools. This allows startups to
systematically increase open, click, and reply rates as
well as conversion into negotiations. The article will be
useful to startup founders, venture analysts, and
The American Journal of Management and Economics Innovations
47
https://www.theamericanjournals.com/index.php/tajmei
fundraising specialists seeking to improve the efficiency
of cold communications with investors.
Keywords:
algorithmic investor identification, machine
learning, cold outreach, fundraising, investor ranking,
multistep outreach, email marketing, link prediction
Introduction
Mailboxes of venture fund partners are overflowing: the
average open rate of business email campaigns
fluctuates between 17 and 28%, depending on the
industry, while click-through rates in most sectors barely
reach 2% (Campaign Monitor, n.d.). The low conversion
is exacerbated by the very method of investor selection.
This distribution turns traditional manual research into
a vicious circle: to get on the radar of funds, a warm intro
is needed, and intros are obtained only when the
company is already visible on the market or the founder
has previously sold a successful business.
The present article pursues two objectives. First, it
demonstrates that algorithmic, machine-learning-based
investor search can radically increase the efficiency of
the cold channel even for teams without an elite
network. Second, it shows how combining ML ranking
with a disciplined multistep outreach turns fundraising
from a lottery into a controllable process: in the author’s
campaigns, this has already yielded an open rate of 78%,
a response rate of 13%, and 96 fund calls in three
months without a single warm intro. Thus, the key thesis
of the work is formulated: in the era of available data
and cloud models, the success of capital raising is
determined not so much by the circle of personal
acquaintances as by the quality of the algorithm that
places the right investors at the top of your funnel.
Materials and Methodology
The study of algorithmic identification of relevant
investors is based on the analysis of eleven sources,
including industry reports on email marketing,
Campaign Monitor, and SendGrid, cold-email campaign
statistics, Chatelaine, Pipeful, and Calendly, and the
author’s campaign data. The theoretical basis comprised
works on the construction of investment graphs and link
prediction using Crunchbase (Piloterr), analysis of the
role of soft signals from public channels (Kaiser &
Kuckertz), and dynamic metrics of fund activity (Venture
Capital Association; Te et al.).
Methodologically the study combined comparative
analysis of deliverability and conversion
—
comparing
average open and reply rates with industry benchmarks
to results of algorithmic selection; extraction of
multilayer features (Piloterr), thematic embeddings
from startup descriptions, clustering of investor tweets
and accounting for dynamics of rounds and headcount
(LinkedIn; Venture Capital Association; Te et al.);
systematic review of cold-email campaign practices
—
optimization of templates following the fifteen-words
plus one metric plus links to Deck and Calendly principle
and configuration of DMARC/SPF/DKIM DNS to improve
domain reputation (McGee).
Results and Discussion
Early-stage fundraising operates almost under the same
laws as the classic B2B sales funnel. At the entry stage, a
bulk of investment partner contacts is followed by email
opening, reply, scheduling a call, and, ideally, a term
sheet. The problem is that statistics for cold emails
remain ruthless: on average, the median reply rate
drops to 8.5
–
10% for most campaigns; only the top
quartile of sequences exceeds the 20% reply threshold
(Chatelaine, 2024). With such conversion, a founder
relying on random responses is doomed to receive few
calls after hundreds of emails, while each day of delay
reduces the traction effect and intensifies competition
for funding attention.
For a long time, this statistic was compensated by
personal intros; however, the math here is also
relentless: some deals come from former colleagues or
business acquaintances, others from secondary
recommendations. For teams without an elite network,
this scenario creates a systemic gap between the need
for capital and the real ability to reach a partner.
The algorithmic approach fundamentally changes the
mechanics of the top of the funnel. In the author’s
campaigns, the open rate is consistently maintained at
74
–
80%, and the reply share fluctuates between 10%
and 17%, as shown in Fig. 1. Each email also converts
into a call, yielding a total of 96 meetings over one
quarter without a single warm intro. The main effect
manifests not only in the growth of individual metrics
but in the controllability of the process: instead of
sending emails into the void, a clear sequence of touches
is formed where each event
—
opening, click, reply
—
returns to the model as a training signal.
The American Journal of Management and Economics Innovations
48
https://www.theamericanjournals.com/index.php/tajmei
Fig. 1. Results of one of the author’s outreach campaigns (compiled by the author)
The key to such robustness is data and discipline. First,
sample breadth: for the algorithm to select a relevant
top ten, it requires a pool of several hundred funds
enriched with features of stage, ticket size, syndicates,
and public activity. Second, update frequency: the
investment graph changes weekly, and the model drifts
if new deals are not incorporated. Finally, strict
adherence to cadence
—
two follow-up touches every
three days
—
retains the investor in the funnel longer
than a single email and provides the algorithm with
additional feedback points. Without this operational
discipline, even the most accurate scoring collapses into
statistical noise; with it, fundraising transforms from a
one-off campaign into a repeatable process where
iterative model improvement directly reflects in top and
mid-funnel metrics.
Algorithmic investor identification relies on a four-layer
data corpus that combines both structured deal
registries and soft signals from public sources. The
foundation is an investment graph: historical rounds,
exits, and co-investors are collapsed into a directed
multigraph where nodes represent funds and startups,
and edges are labeled by deal type and date. The public
Crunchbase export alone contains approximately 3.5
million company records, each with round → investor
and startup → exit linkages, yielding millions of edges for
training link prediction and ranking models (Piloterr,
n.d.). This is supplemented by commercial datasets: for
example, QuantumLight builds scoring based on 700,000
VC-backed companies, demonstrating that graph scale is
critical for recommendation quality (Thornhill, 2025).
The next layer comprises portfolio descriptors. For each
company,
the
model
extracts
thematic
and
technological features from descriptions, tags, and stack
words; texts are converted into embeddings and
aggregated into a fund’s niche temperature.
To capture implicit shifts in capital focus, the graph is
enriched with partners’ public theses. Features include
keywords and emotional markers from the Twitter
stream: a recent sample of 994,969 tweets from 822
investors enabled the authors to identify clusters such
as infrastructure AI or climate-tech that directly
correlate with subsequent checks (Kaiser & Kuckertz,
2024).
Finally, activity signals impart temporal sensitivity to the
model. The average fund seeks to close only seven deals
per year
—
a low throughput through the funnel
—
and
the probability of a response increases sharply if scoring
marks the partner as having recently closed a round
(Venture Capital Association, 2023). Hiring pace is added
to the dynamic features: combined Crunchbase-
LinkedIn analyses show that headcount growth and
Series A probability are statistically linked, thus a
monthly sampling of team profiles enters the dataset as
a proxy traction metric (Te et al., 2023).
Such a multilayer corpus
—
deal graph, thematic
embeddings, public theses, and live activity indicators
—
provides the model with sufficient information to rank
investors by current relevance with high stability rather
The American Journal of Management and Economics Innovations
49
https://www.theamericanjournals.com/index.php/tajmei
than by past-experience stereotypes.
As noted in the Introduction, the average business email
is opened by only 17
–
28% of recipients and clicked by
approximately 4.48% (Send Grid, n.d.). Against this
background, every structural improvement of the email
becomes critical: in the author’s internal campaign
sample, the template redesign increased the open rate
while maintaining the reply rate. The key element of this
dynamic is the first text block, limited to fifteen words:
it simultaneously states the essence of the product and
demonstrates a concrete result. Due to its high
information density, the recipient decides to open the
attachment before reading the rest of the email.
This is followed by a single bullet highlighting the key
growth metric: conversions drop sharply if the list is
expanded beyond one figure because attention shifts
from the call to action to details. Brevity is offset by a
hyperlink to a ten-page Pitch Deck: email analysis
showed that the metric + document combination nearly
doubles the probability of entering the data room
compared to the metric alone, since the investor
receives both the incentive and a convenient format for
hypothesis verification. The first interaction finishes
with an explicit connection to Calendly, giving the choice
to pick a slot in the present week; a Focus Digital study
says that having such a tool increases total deal
conversion to 0.21% (McGee, 2024). All three
—
the
brief intro, one strong figure, and two clickable items
—
the Deck and the calendar
—
make up a simple but
enough info package that changes the balance.
B2B campaign experience shows that a single email
seldom closes the investor: Pipeful analytics on a sample
of 11 million cold emails records a reply-rate increase of
almost 50% after the first reminder, meaning that it is
the follow-up that transforms a formal contact into a
dialogue (What Are B2B Cold Email Response Rates?,
2024). Longer sequences amplify the effect: a Calendly
study based on 300,000 emails demonstrated that a
three-touch series raises the cumulative reply rate from
1% to 9%, and with seven touches the metric climbs to
27% (Batrawy & Cottle, 2024). At the same time, nearly
half (48%) of salespeople do not make repeat calls;
however, 93% of converted leads are often achieved
only after the sixth cold-call attempt, as shown in Fig. 2.
Fig. 2. Percentage of sales personnel who give up after several rejections (Batrawy & Cottle, 2024)
Let us consider an example of an algorithm
implemented with TaskInfinity. First, the team
formulates a brief description of the startup
—
one or
two sentences about the sector, stage, and key product
features, so that the system can correctly understand
the target filters for investor search. Then, on the
investor-search platform, mandatory filters are set for
industry (Industry), round stage (Stage), and market
(Market), with additional optional filters such as
geography, investor type, and diversity criteria.
The American Journal of Management and Economics Innovations
50
https://www.theamericanjournals.com/index.php/tajmei
Next, an initial shortlist of investors is formed, and each
participant is assigned a weight based on four metrics:
the number of deals in selected industries and related to
LLM (weight 0.1), the number of deals at selected stages
(0.4), the number of deals at selected markets (0.4), and
the number of deals by diversity (0.1). If any metrics are
missing, the system retains proportions and reallocates
100% of the weight among the remaining filters. After
calculating the weights, investors are categorized into
three groups: the "green" category for a total weight of
45% and above, the "yellow" category for 15
–
44%, and
the "red" category for 1
–
14%.
Within each category, a second prioritization is carried
out based on three indicators with equal coefficients
(0.33 each): the ratio of recent investments to total
investments, leading participation in selected stages,
and the share of total investments from filters. By
summing the percentages from the first and second
prioritization, the system generates the final ranking of
investors within the "green," "yellow," and "red" groups
in descending order of the total score.
After this, investors with high and medium relevance are
selected from the list. The database is enriched with
contact details through the "Get Investor Contacts"
interface, which pulls email, LinkedIn profile links, and
other required fields. Investors in the "green" and
"yellow" categories are marked, and a ready CSV file is
exported with columns such as "First Name," "Last
Name," "Company," "Email," "LinkedIn," "Website," and
other details for subsequent mass mailing.
At the same time, the infrastructure for cold email
campaigns is set up: additional domains are registered,
mailboxes are created via Google Workspace, 301
redirects and DNS records (SPF, DMARC, DKIM, CNAME
tracking) are configured. Over one to two weeks, a
"warm-up" is conducted by sending small test emails to
improve the reputation of the new domains. LinkedIn
accounts are prepared separately: profiles are
completed, the contact network is grown, and basic
activity is maintained to avoid blocks. During cold
outreach, limits are followed
—
no more than 20
invitations per day and 100 messages per week.
Finally, sequential messaging scripts are created. In
LinkedIn, an invitation to connect is first sent mentioning
mutual contacts and requesting advice, followed by a
short proposal for a call after acceptance; two days later,
if no reply is received, a follow-up with an expanded bio
is sent; another two days later, a polite apology message
is sent asking for expert feedback. In email campaigns,
the first message contains an introduction to the team,
key metrics (ARR, YoY, GM, LTV/CAC), information about
past investments, the round status, and a link to
schedule a call; after three days, the first follow-up is
sent with “Hope you are having a great week”; another
three days later, the second follow-up is sent with a brief
repetition of metrics and a link to the Pitch Deck. Once
templates and contact lists are prepared, they are
uploaded to the chosen mailing tool, and automatic
sequences are activated with set limits on outbound
messages and follow-up ratios. During the campaign,
key metrics
—
open rate, response rate, and scheduled
meetings
—
are monitored, and adjustments are made
to messaging, segmentation, or sending frequency. A/B
tests on subject lines and offers are conducted to
optimize performance. After collecting the data,
strategy and scripts are adjusted, and outreach
continues with the updated database. This systematic
approach allows for continuously finding the right
investors, warming up domains and accounts, sending
personalized messages, and quickly optimizing tactics
based on actual results.
It was this algorithm that enabled the author to run a
few successful campaigns - data from one of them is
shown in Fig. 3.
The American Journal of Management and Economics Innovations
51
https://www.theamericanjournals.com/index.php/tajmei
Fig. 3. Outcomes of one of the author’s campaigns (compiled by the author)
The author of the article first formulated a brief
description of the startup and configured filters by
industry, stage, geography, and investor type, after
which a list of relevant contacts was obtained and new
domains were warmed up by test mailings. In the period
from 19 to 25 May, a total of 936 emails were sent, of
which 465 were opened (49,7%), 153 were clicked
(32,9%), and 34 generated a reply (7,3%). A follow-up
three days later in the same week, on 687 sent
messages, yielded 169 opens (24,6%), 25 clicks (14,8%),
and 3 replies (1,8%). Two subsequent reminders in small
batches of 59 messages each confirmed the validity of
the scenario by adding one or two additional responses.
After analysing the initial results, the author adjusted
the subject line and div text of the first email,
enhanced personalization, and adapted sending times.
In the second week, of 1331 emails sent, 63,2% were
opened, 43,6% were clicked, and 5,8% generated
replies. In the first follow-up of 604 sent messages, there
were 253 opens, 137 clicks, and 12 replies. The third
wave of 653 emails produced 335 opens, 79 clicks, and
13 replies, while the fourth reminder resulted in 115
opens and 5 replies. This yielded 17 scheduled calls,
exceeding the first week’s results by one third.
In the third week, the author continued fine-tuning
follow-ups and segmenting by time zone. Of 900 emails
sent, 584 were opened, 283 clicked, and 40 replied. The
second message achieved 400 opens, 171 clicks, and 17
replies. The third wave of 818 emails recorded 509
opens, 231 clicks, and 20 replies, while the fourth
reminder generated 243 opens and 7 replies. As a result,
21 calls were scheduled in the third week, marking a
record outcome.
This case demonstrates that a systematic approach
—
thoughtful filtering of investors, domain warming, clear
sequences comprising an initial email and three follow-
ups with continuous metric analysis
—
enables stable
increases in open rate, click rate, and conversion into
meetings. The growth in scheduled calls from 14 to 21
over three weeks and the progressive improvement in
engagement confirm the effectiveness of the described
algorithm.
The American Journal of Management and Economics Innovations
52
https://www.theamericanjournals.com/index.php/tajmei
Conclusion
As a result of the conducted research, it has been
demonstrated that an algorithmic approach to
identifying relevant investors based on machine learning
methods fundamentally alters the efficiency of the cold
communication channel. In contrast to traditional
manual research, which fails to achieve acceptable open
and reply rates without warm introductions, the
automated model sustains an open rate of 74
–
80% and
achieves a reply rate of 10
–
17%, more than twice the
average industry benchmarks. Meanwhile, the multistep
follow-up strategy and rapid updating of investor data
create a controlled feedback loop that enables iterative
improvement
of
recommendation
quality
and
conversion into calls: in the author’s campai
gns, 96
meetings were scheduled over one quarter without a
single warm introduction.
The key success factor is the multilayer feature
architecture combining structured deal registries,
thematic embeddings, soft signals from public sources,
and dynamic activity indicators. Such a combination
enables the model to account for both historical
connections within the investment graph and current
shifts in fund focus, which is especially critical in a highly
competitive environment for partner attention. Regular
replenishment and updating of the training dataset,
together with discipline in adhering to the cadence
strategy, ensure algorithmic resilience to drift and
preserve recommendation efficacy over the long term.
Practical implementation of the described algorithm
encompasses not only the development of an investor
ranking model but also the construction of a
comprehensive infrastructure for large-scale mailings:
from mail domain configuration and A/B testing of email
templates to integration with meeting-scheduling tools
and monitoring of key metrics. Taken together, this
transforms fundraising from a lottery into a reproducible
business process in which each stage, from initial
contact selection to subsequent result analysis, is
grounded in reliable data and continuous improvement.
The author’s campaign results, clearly presented across
three mailing waves, confirm that a systematic approach
enables not only rapid growth in investor open and
engagement rates but also steady increases in
scheduled calls: from 14 to 21 over three weeks with
fine-tuned scenarios and segmentation. This indicates
that success in capital raising under modern conditions
is determined not by the breadth of personal networks
but by the quality of the algorithm and the rigor of
operational discipline.
Thus, algorithmic investor identification via machine
learning and disciplined multistep outreach transform
the fundraising process into a controlled, repeatable
procedure with clear metrics and opportunities for
constant refinement. In an era of accessible data and
cloud models, the primary determinant of outcome
becomes not an elite network but a startup’s ability to
build and maintain a high-precision algorithm that
places the most relevant investors at the top of the
funnel.
References
1.
Batrawy, B., & Cottle, C. (2024).
The Art of Cold
Outreach
.
Calendly.
https://pages.calendly.com/rs/482-NMZ-
854/images/The-Art-of-Cold-Outreach-EB.pdf
2.
Campaign Monitor. (n.d.).
What are good open
rates, CTRs, & CTORs for email campaigns?
Campaign Monitor. Retrieved May 15, 2025, from
https://www.campaignmonitor.com/resources/kno
wledge-base/what-are-good-email-metrics/
3.
Chatelaine, J. (2024, March 14).
The Complete 2024
Guide to Cold Email Metrics
. Quick Mail.
https://quickmail.com/cold-email/metrics
4.
Kaiser, M., & Kuckertz, A. (2024). Emotions and
entrepreneurial finance: Analysis of venture
capitalists’ and business angels’ digital footprints on
Twitter.
International
Entrepreneurship
and
Management
Journal
,
21
(1).
https://doi.org/10.1007/s11365-024-01018-9
5.
McGee, C. (2024, August 29).
Average Cold Email
Conversion Rate by Industry
. Focus Digital.
https://focus-digital.co/cold-email-conversion-
rate/
6.
Piloterr. (n.d.).
Crunchbase Company Dataset
.
Piloterr.
Retrieved
May
18,
2025,
from
https://www.piloterr.com/datasets/crunchbase-
company
7.
Send Grid. (n.d.).
2023 Email Marketing Benchmark
Report
. Send Grid. Retrieved May 23, 2025, from
https://sendgrid.com/en-us/resource/2023-Email-
Marketing-Benchmark-
Report?utm_source=chatgpt.com
8.
Te, Y.-F., Wieland, M., Frey, M., Pyatigorskaya, A.,
Schiffer, P., & Grabner, H. (2023). Making it into a
The American Journal of Management and Economics Innovations
53
https://www.theamericanjournals.com/index.php/tajmei
successful series a funding: An analysis of
Crunchbase and LinkedIn data.
The Journal of
Finance
and
Data
Science
,
9
,
100099.
https://doi.org/10.1016/j.jfds.2023.100099
9.
Thornhill, J. (2025, June 12).
Are machines smarter
than
venture
capitalists?
Financial
Times.
https://www.ft.com/content/1cd7af37-2583-429e-
b4d4-6d6a21108d75
10.
Venture Capital Association. (2023).
Venture capital
market study 2023
. Venture Capital Association.
11.
What are B2B Cold Email Response Rates?
(2024,
November
23).
LinkedIn.
https://www.linkedin.com/pulse/what-b2b-cold-
email-response-rates-2024-study-pipeful-3tuke
