Silmari
95% of Enterprise AI Projects fail because Institutional Memory is locked inside company systems
Silmari is the key
THE PROBLEM
Companies can't explain how their oldest software systems work, because the people who made the decisions have left.
When a company runs the same system for ten years or more, the reasons behind it end up buried in old email. The people who made those decisions retire or move on, and what they knew goes with them.
ONE EXAMPLE: A LEGACY SYSTEM AT THE CITY OF CHICAGO, LIVE SINCE 2013
- There are over 200,000 emails about it.
- Its project manager of 14 years leaves at the end of this year. About 100,000 of her emails go with her.
- Every week someone asks a question only those emails can answer:
A new CIO asks: why does this run on our own servers instead of AWS?
Someone asks: what year did we upgrade to Oracle 19c? Nobody remembers.
Someone asks: can we turn off server 15? Nobody is sure what depends on it.
GUESSING WRONG IS EXPENSIVE
Someone once switched off a forgotten server in a lab, eight years after the person who built it had left. That server was sending every text message for an application used by tens of thousands of people.
WHY NOW
Systems installed in 2010–2014 are now 12–16 years old, and much of the software under them has lost vendor support.
- Windows Server 2008 R2 and Python 2.7 lost support in January 2020, SQL Server 2012 in July 2022, and Windows Server 2012 in October 2023.
- Microsoft, AWS, IBM and Google now sell AI tools that upgrade these older applications. Those tools can rewrite the code. They can't know the edge cases that make each company, department and project different: the exception finance added in 2016, the workaround one client needed, the rule nobody wrote down.
- Every AI upgrade has to carry that institutional memory into the new system. Silmari is where it comes from.
VALUE PROPOSITION
Silmari turns days of digging through old email into an answer with sources, in minutes.
WHEN THIS HAPPENS
TODAY
WITH SILMARI
A new CIO asks why the system runs on the company's own servers
Someone spends days searching email and asking whoever is left
Silmari shows the decision, its date, and the emails it came from
An auditor asks why a transaction was recorded a certain way
Ten people get asked; inboxes get searched
Silmari shows who decided and when, with sources
A long-time employee resigns
What they knew leaves with them
Their emails are already searchable
The company plans to replace the system
Months of research before work can start
The original requirements and trade-offs, dated and sourced
What it's worth: one system's memory pays back Silmari's $60,000 a year many times over.
WHERE THE VALUE COMES FROM
OUR ESTIMATE
HOW WE GOT IT
The system itself
$100M invested in a 15-year system
Customer interviews. Silmari costs 0.06% of that a year to protect it.
Everyday questions
~$62K a year
One question a week × 8 hours of digging through email × $150/hour.
Audit season
~$60K a year
20 auditor questions × 10 people asked × 2 hours × $150/hour.
A key person leaving
~$100K, each time
A replacement needs 6–12 months to rebuild 14 years of context; about half a $200K salary is lost while they do.
Replacing the system
$2M+ saved, once
Replacement costs "tens of millions" (video 2). If 20% of a $20M project is spent rediscovering why the old system works the way it does, and Silmari halves that, it saves $2M.
Questions and audits alone are worth about $120K a year, twice the license. One replacement project is worth more than 30 years of Silmari.
UNDERLYING MAGIC
Silmari reads a system's history and turns it into facts, each with a date and a source.
1 · LOAD
email, tickets,
change requests,
documents
2 · EXTRACT
who decided what, when, and why
each fact linked to the message it came from
3 · ASK
"Who upgraded server 7 in 2018?"
the answer, plus the emails it came from
Dates come from the record, not from memory.
People misremember years. Email timestamps don't.
It keeps the reason, not just the event:
why the servers were upgraded, how, and by whom.
It runs on databases companies already trust:
Postgres today, Oracle in progress. A 14-year history can't live on one person's laptop.
BUSINESS MODEL
Companies pay Silmari a yearly license for each system it covers, plus a one-time setup fee.
LICENSE
$60,000 per system per year ($5,000 a month).
SETUP
A one-time fee of about $10,800 to load the system's history, scoped during discovery. (Estimate: 18% of the first-year license, the top of the 12–18% services band in the revops pricing playbook.)
EXTRA COMPUTING
If a large archive needs it, it is billed to the customer.
CUSTOMER'S OWN SERVERS
Silmari can run on the customer's own servers, for companies that can't send email outside.
WHO PAYS
US companies with more than $20M in revenue that run long-lived systems with a small IT team. For a company with $50M or more in revenue, $60,000 is 1–4% of a typical IT budget.
GO-TO-MARKET PLAN
Silmari sells first through firms that already maintain old systems for many companies.
01 · PARTNER WITH SUPPORT FIRMS
Firms that maintain old business software for many clients. One partner reaches many systems. Examples: Rimini Street (3,102 clients), Spinnaker Support (1,000 clients). Plan: 15 partners signed in year 1, 50 in year 2, 120 in year 3.
02 · ADD OUR OWN SALES TEAM
Once the first customers are live: 1 account executive in year 1, 4 in year 2, 6 in year 3.
03 · SELL WHEN THE NEED IS OBVIOUS
A key person resigns, a new CIO arrives, audit season starts, or a replacement project is planned.
ESTIMATED COST TO WIN A CUSTOMER
About $17,000 by year 3, earned back in under five months. A typical sale takes about one quarter.
(Model estimate, Appendix A7: sales and marketing spend ÷ new customers; partners bring most of the deals. The one-quarter sales cycle is the revops playbook's mid-market norm of 30–90 days.)
COMPETITIVE ANALYSIS
We found no product that reads a company's old email and explains why a system was built the way it was.
WHAT THE CUSTOMER NEEDS
SEARCH TOOLSGlean · Microsoft Copilot
WIKISConfluence · Notion
EMAIL ARCHIVE AND LEGAL SEARCH
PEOPLEStaff who are left · consultants
SILMARIthis is us
Answers who did what, when and why, with the date and the source
Finds the messages; you work out the why
Only if someone wrote it down
Answers with sources, built for legal cases
If they remember; people forget years
Yes, with dates and sources
Works from the records we already have: email, tickets, change requests
Only email in their cloud or live apps
No; someone has to write it
Email yes; tickets and change requests no
Yes, read by hand
Yes; full 200,000-email load is next
Keeps what people knew after they leave
Inconvenient = zero use
Yes, for what was written down
Yes; the archive keeps it
No; it leaves with them
Yes; their email stays searchable
Lasts 10+ years on our own approved database and servers
Copilot: no. Glean: unknown
Mostly cloud
Mostly cloud
No; only while they stay
Postgres today, Oracle in progress
Records what AI agents did, so it can be checked later
Logs its own chats, not other agents' work
Only if someone writes it up
Only what agents send by email
No
Yes; agents save to it today (prototype)
Answers in minutes, so a routine question is not a project
Minutes; finds messages, not answers
Minutes, if the page exists
Minutes to hours
Days to weeks
Minutes, with sources (pilot)
MANAGEMENT TEAM
Silmari is built by people who have lived with this problem.
MJ
Maceo Jourdan
FOUNDER & CEO
2023–PRESENT · AI GROWTH ENGINEER & SYSTEMS ARCHITECTBuilds AI automation pipelines and production systems for clients in SaaS, e-commerce and services.
2018–2023 · CO-FOUNDER & CEO, CANEXXIA HEALTHCAREScaled the company from $0 to a $1.2B capital raise, on financial models he built.
2018–2023 · CANEXXIA HEALTHCAREDeployed $36M in M&A. Moved legacy patient intake and billing to digital-first, automated systems.
2017–2018 · CHIEF MARKETING OFFICER, TRUDOGManaged a $120K-a-month ad budget across Meta, Instagram and email.
ADVISORS
LA
Landon Allen
ADVISOR · TALENT AND RECRUITING
- At Adyen. Former Head of Recruiting at Splunk (acquired by Cisco for about $28B); earlier leadership at PayPal and early-stage Venmo.
- 1M+ professional connections.
MR
Matt Richter, PhD
TECHNICAL ADVISOR · ARCHITECTURE
- Stanford Professor of Physics.
- 30 years in machine learning, from pre-LLM methods through CNNs and transformers.
FINANCIAL PROJECTIONS & KEY METRICS
The plan reaches 211 paying systems and $12.7M in yearly recurring revenue by the end of year three.
Paying systems at year end
9
63
211
New systems from our sales team
6
36
75
New systems from partners
3 (+1 City of Chicago pilot)
21
86
Win rate: our sales team · partner pilots
25% · 35%
25% · 40%
25% · 45%
Yearly recurring revenue
$0.55M
$3.79M
$12.66M
Throughput (revenue minus truly variable costs)
$0.27M
$2.07M
$8.23M
Operating expense (salaries and overhead)
$1.43M
$4.63M
$8.96M
Net profit
−$1.16M
−$2.56M
−$0.73M
STATUS, TIMELINE & USE OF FUNDS
We're raising a pre-seed round to prove Silmari on real customer systems.
TODAY
A working prototype for personal and institutional memory, production-ready.
NEXT
Load the full 13-year history of the City of Chicago legacy system (over 200,000 emails) and answer the real questions that come in each week, within the first year.
WHAT THE MONEY PAYS FOR
Handling 200,000-email histories reliably
12%
Running on Oracle and SQL Server
5%
Sales and partners: reach 63 paying systems by year two
77%
Security reviews for running on customers' own servers
1%
Operations and everything else
5%
Every time someone leaves, a company forgets part of how its systems work. Silmari keeps that knowledge.
Slide 1 · Title
Open with one sentence
"A legacy system at the City of Chicago went live in 2013. There are over 200,000 emails about it, and every week someone asks a question only those emails can answer."
Slide 2 · Problem / Opportunity
Why now
In 2014 about 65% of enterprise workloads still ran in company data centers (Uptime Institute). Cloud began earlier — Salesforce sold on-demand CRM in 2000, AWS S3 launched in 2006, Azure became generally available in 2010 — but the 2010–2014 systems are the ones now aging out of support. Not every package from that era is unsupported (Java 8 and .NET 3.5 SP1 still are), so say "much of", never "all". Source: research/2026-10-04_why-now-2010-2014-legacy-wave.md.
The server story
Tell the server story out loud; it is the pulse-rate moment.
Slide 3 · Value Proposition
What it's worth
These are estimates, built from the City of Chicago legacy system (13 years of history) and from the two videos. The reasoning for each:
- $100M system — video 2 (5VdmZIm6UfU): "You've got a system that you put together 15 years ago. You're $100 million into this system and you can't remember who built the stored procedures or why this module exists."
- One question a week — video 2: questions come up "more often than you would think", and each one means deriving the answer "from all these other sources, like the emails, the documentation, the change requests". 8 hours and $150/hour (loaded cost of a senior IT person or project manager) are our assumptions; 52 × 8 × $150 = $62,400.
- Audit — video 1 (ZJVbErOClpM): "a huge amount of money is spent every year during the audit process when the auditors ask a question and you have to go ask 10 people or search through emails." 20 questions and 2 hours each are our assumptions; 20 × 10 × 2 × $150 = $60,000.
- Key person — video 2: the project manager of 14 years is leaving with about 100,000 emails. The 6–12 month ramp and $200K loaded salary are our assumptions; half a year of $200K ≈ $100,000.
- Replacement — video 2: a city system that "costs the city millions of dollars every year to run … it would cost tens of millions dollars to replace it, and that's because there's no institutional knowledge." $20M (low end of "tens of millions"), 20% spent on rediscovery and a 50% reduction are our assumptions; $20M × 20% × 50% = $2M, and $2M ÷ $60K ≈ 33 years of license.
- The big picture — video 1: "companies lose billions of dollars because of loss of institutional memory"; video 2: "a multi-hundred million-dollar problem for companies". Use these as the founder's view, not as market-size figures (A1 has the bottom-up market).
Slide 4 · Underlying Magic
No speaker note.
Slide 5 · Business Model
No speaker note in the v15 draft.
Slide 6 · Go-to-Market Plan
No speaker note in the v15 draft.
Slide 7 · Competitive Analysis
Where the rows come from
Each row is a need the customer states in the two videos (video A: 5VdmZIm6UfU, video B: ZJVbErOClpM). Full list, ranking and evidence per cell: research/2026-10-05_customer-requirements-from-transcripts.md.
- Who, when, why — A: "Who did the upgrade on server seven in 2018?"; "What year did you upgrade to Oracle 19c?"
- Records we already have — A: "the only time we can find the change request is from the email that we sent to them at the time it was done."
- After people leave — A: the project manager of 14 years is leaving with "100,000 emails of her own".
- 10+ years, own database — A: "my laptop will not last 14 years"; "Postgres is not an allowable a database for several of my clients".
- AI agents — A: "without memory, you can't ask the AI agent what it did."
- Minutes — A: questions come "more often than you would think"; B: auditors mean you "go ask 10 people or search through emails".
Columns and scores
"Consultants" became "People" (staff who are left, plus consultants): asking whoever is left is what most companies do today, and consultants do the same thing for a fee. AI code-modernization tools (Mechanical Orchard, Moderne, CAST, and the cloud vendors' tools) read code, not decisions, so they are possible partners, not a column (Appendix A2). Our assessment, not a cited fact: search tools on staff who left and on AI agents; wikis on speed and AI agents; email archives on tickets and AI agents; all of the People column except "Weeks".
Source: research/2026-10-04_tam-sam-som-institutional-memory.md, section 4. Cited fact: Microsoft Learn (Exchange hybrid and Copilot, updated 2026-08-18) says Copilot is cloud-based and cannot ground on on-premises mailboxes. Glean's on-prem support is unknown, and no vendor with Oracle support was found. That is an absence of evidence, not proof about roadmaps.
Silmari's own cells are honest: Postgres works today, the Oracle port is in progress, loading the full 200,000-email history is the next milestone, the agent memory is a prototype, and "minutes, with sources" is a pilot claim not yet independently verified. Re-check Sentra and Copilot-for-Exchange-Server status right before the meeting.
Detail: Appendix A2.
Slide 8 · Management Team
Founder receipts come from Maceo's resume (v8c). Advisor lines come from the v12/v14 team slide. Landon's current employer is Adyen (confirmed by the founder, 2026-10-04).
Slide 9 · Financial Projections & Key Metrics
Assumptions
Every input is an assumption until paying customers exist. The $60K per system per year price is founder-set. Direct quota ($859K to $1.2M of new license ARR per account executive) is founder-set. Partner pilot-to-paid (35%, 40%, 45%) and direct win rate (25% for account executives, 30% for the founder) are assumptions. Model counts are fractional and rounded on the slide.
Market math
TAM: 137,522 US firms with 75 to 499 employees × 40% thin or outsourced IT = 55,009 firms × 2 systems = 110,018 systems × $60,000 = $6.60B a year. SAM: 55,009 × 50% with a system live 10+ years × 50% with a retained, exportable archive reachable by a partner = 13,752 firms × 2 = 27,504 systems × $60,000 = $1.65B a year. Year-3 systems: 211 / 27,504 = 0.77% of SAM systems. The 40%, 50%, 50% and 2-systems shares are assumed, not sourced.
Throughput view
Throughput = revenue − truly variable costs (partner share, hosting, commissions). Operating expense = all salaries and other operating spend. Net profit = throughput − operating expense. Year 3 is nearly break-even (throughput / operating expense = 0.92). The base case never runs out of cash: lowest cash is $726,684 in month 25. At a $36K price the model runs out of cash in month 23. Detail: Appendix A1, A4–A7 (A6 financial model, A7 throughput and unit economics).
Slide 10 · Status, Timeline & Use of Funds
Use of funds comes from the financial model (Appendix A6): of the funding modeled, about 11.7% goes to the email-history engine, 5.0% to the Oracle and SQL Server ports, 77.6% to sales and partner delivery, 0.5% to security reviews and 5.2% to everything else, rounded here to 12 / 5 / 77 / 1 / 5. The funding lasts about 25 months (lowest cash in month 25), which is why the sales line targets the model's 63 paying systems by the end of year two.
Slide 1 — Title
Silmari. 95% of Enterprise AI Projects fail because Institutional Memory is locked inside company systems.
Silmari is the key.
Maceo Jourdan, Founder & CEO. hello@silmari.ai.
Slide 2 — Problem / Opportunity
Companies can't explain how their oldest software systems work, because the people who made the decisions have left.
When a company runs the same system for ten years or more, the reasons behind it end up buried in old email. The people who made those decisions retire or move on, and what they knew goes with them.
One example: a legacy system at the City of Chicago, live since 2013.
- There are over 200,000 emails about it.
- Its project manager of 14 years leaves at the end of this year. About 100,000 of her emails go with her.
- Every week someone asks a question only those emails can answer: a new CIO asks why this runs on our own servers instead of AWS; someone asks what year we upgraded to Oracle 19c, and nobody remembers; someone asks whether we can turn off server 15, and nobody is sure what depends on it.
Guessing wrong is expensive. Someone once switched off a forgotten server in a lab, eight years after the person who built it had left. That server was sending every text message for an application used by tens of thousands of people.
Why now: systems installed in 2010–2014 are now 12–16 years old, and much of the software under them has lost vendor support.
- Windows Server 2008 R2 and Python 2.7 lost support in January 2020, SQL Server 2012 in July 2022, and Windows Server 2012 in October 2023.
- Microsoft, AWS, IBM and Google now sell AI tools that upgrade these older applications. Those tools can rewrite the code. They can't know the edge cases that make each company, department and project different: the exception finance added in 2016, the workaround one client needed, the rule nobody wrote down.
- Every AI upgrade has to carry that institutional memory into the new system. Silmari is where it comes from.
Slide 3 — Value Proposition
Silmari turns days of digging through old email into an answer with sources, in minutes.
| When this happens | Today | With Silmari |
|---|
| A new CIO asks why the system runs on the company's own servers | Someone spends days searching email and asking whoever is left | Silmari shows the decision, its date, and the emails it came from |
| An auditor asks why a transaction was recorded a certain way | Ten people get asked; inboxes get searched | Silmari shows who decided and when, with sources |
| A long-time employee resigns | What they knew leaves with them | Their emails are already searchable |
| The company plans to replace the system | Months of research before work can start | The original requirements and trade-offs, dated and sourced |
What it's worth: one system's memory pays back Silmari's $60,000 a year many times over.
| Where the value comes from | Our estimate | How we got it |
|---|
| The system itself | $100M invested in a 15-year system | Founder's experience (video 2). Silmari costs 0.06% of that a year to protect it. |
| Everyday questions | ~$62K a year | One question a week × 8 hours of digging through email × $150/hour. |
| Audit season | ~$60K a year | 20 auditor questions × 10 people asked × 2 hours × $150/hour. |
| A key person leaving | ~$100K, each time | A replacement needs 6–12 months to rebuild 14 years of context; about half a $200K salary is lost while they do. |
| Replacing the system | $2M+ saved, once | Replacement costs "tens of millions" (video 2). If 20% of a $20M project is spent rediscovering why the old system works the way it does, and Silmari halves that, it saves $2M. |
Questions and audits alone are worth about $120K a year, twice the license. One replacement project is worth more than 30 years of Silmari.
Slide 4 — Underlying Magic
Silmari reads a system's history and turns it into facts, each with a date and a source.
1. Load: email, tickets, change requests, documents. 2. Extract: who decided what, when, and why, each fact linked to the message it came from. 3. Ask: "Who upgraded server 7 in 2018?" and get the answer, plus the emails it came from.
- Dates come from the record, not from memory. People misremember years. Email timestamps don't.
- It keeps the reason, not just the event: why the servers were upgraded, how, and by whom.
- It runs on databases companies already trust: Postgres today, Oracle in progress. A 14-year history can't live on one person's laptop.
Slide 5 — Business Model
Companies pay Silmari a yearly license for each system it covers, plus a one-time setup fee.
- License: $60,000 per system per year ($5,000 a month).
- Setup: a one-time fee of about $10,800 to load the system's history, scoped during discovery. (Estimate: 18% of the first-year license, the top of the 12–18% services band in the revops pricing playbook.)
- Extra computing, if a large archive needs it, is billed to the customer.
- Can run on the customer's own servers for companies that can't send email outside.
Who pays: US companies with more than $20M in revenue that run long-lived systems with a small IT team. For a company with $50M or more in revenue, $60,000 is 1–4% of a typical IT budget.
Slide 6 — Go-to-Market Plan
Silmari sells first through firms that already maintain old systems for many companies.
- Partner with support firms that maintain old business software for many clients. One partner reaches many systems. Examples: Rimini Street (3,102 clients), Spinnaker Support (1,000 clients). Plan: 15 partners signed in year 1, 50 in year 2, 120 in year 3.
- Add our own sales team once the first customers are live: 1 account executive in year 1, 4 in year 2, 6 in year 3.
- Sell when the need is obvious: a key person resigns, a new CIO arrives, audit season starts, or a replacement project is planned.
Estimated cost to win a customer: about $17,000 by year 3, earned back in under five months. A typical sale takes about one quarter. (Model estimate, Appendix A7: sales and marketing spend ÷ new customers; partners bring most of the deals. The one-quarter sales cycle is the revops playbook's mid-market norm of 30–90 days.)
Slide 7 — Competitive Analysis
We found no product that reads a company's old email and explains why a system was built the way it was.
| What the customer needs | Search tools (Glean, Microsoft Copilot) | Wikis (Confluence, Notion) | Email archive and legal search | People (staff who are left, consultants) | Silmari |
|---|
| Answers who did what, when and why, with the date and the source | Finds the messages; you work out the why | Only if someone wrote it down | Answers with sources, built for legal cases | If they remember; people forget years | Yes, with dates and sources |
| Works from the records we already have: email, tickets, change requests | Only email in their cloud or live apps | No; someone has to write it | Email yes; tickets and change requests no | Yes, read by hand | Yes; full 200,000-email load is next |
| Keeps what people knew after they leave | Only if their mailbox is kept | Yes, for what was written down | Yes; the archive keeps it | No; it leaves with them | Yes; their email stays searchable |
| Lasts 10+ years on our own approved database and servers | Copilot: no. Glean: unknown | Mostly cloud | Mostly cloud | No; only while they stay | Postgres today, Oracle in progress |
| Records what AI agents did, so it can be checked later | Logs its own chats, not other agents' work | Only if someone writes it up | Only what agents send by email | No | Yes; agents save to it today (prototype) |
| Answers in minutes, so a routine question is not a project | Minutes; finds messages, not answers | Minutes, if the page exists | Minutes to hours | Days to weeks | Minutes, with sources (pilot) |
Microsoft's own documentation says Copilot can't read email stored on a company's own servers. Closest competitors: Glean, Microsoft 365 Copilot, Sentra. Detail: Appendix A2 (appendix/a2-competitive-landscape.html).
Slide 8 — Management Team
Silmari is built by people who have lived with this problem.
- Maceo Jourdan, Founder & CEO. 2023–present: AI growth engineer and systems architect, building AI automation pipelines and production systems for clients. 2018–2023: co-founder and CEO of Canexxia Healthcare, scaled from $0 to a $1.2B capital raise on financial models he built; deployed $36M in M&A; moved legacy patient intake and billing to digital-first, automated systems. 2017–2018: Chief Marketing Officer of TruDog, managing a $120K-a-month ad budget.
- Advisor: Landon Allen, talent and recruiting — at Adyen; former Head of Recruiting at Splunk (acquired by Cisco for about $28B); earlier leadership at PayPal and early-stage Venmo; 1M+ professional connections.
- Advisor: Matt Richter, PhD, technical advisor for architecture — Stanford Professor of Physics; 30 years in machine learning, from pre-LLM methods through CNNs and transformers.
Slide 9 — Financial Projections & Key Metrics
The plan reaches 211 paying systems and $12.7M in yearly recurring revenue by the end of year three. Year 1 / Year 2 / Year 3:
- Paying systems at year end: 9 / 63 / 211
- New systems from our sales team: 6 / 36 / 75
- New systems from partners: 3 (plus 1 City of Chicago pilot) / 21 / 86
- Win rate, our sales team · partner pilots: 25% · 35% / 25% · 40% / 25% · 45%
- Yearly recurring revenue: $0.55M / $3.79M / $12.66M
- Throughput (revenue minus truly variable costs): $0.27M / $2.07M / $8.23M
- Operating expense (salaries and overhead): $1.43M / $4.63M / $8.96M
- Net profit: −$1.16M / −$2.56M / −$0.73M
The market: about 55,000 US companies fit our profile. At two systems each and $60,000 per system, that's $6.6B a year. About $1.65B of it is reachable through partners. Year three is under 1% of that. Cash: the plan never runs out. Cash is lowest in month 25, at $0.73M. Detail: Appendix A1, A4–A7.
Slide 10 — Status, Timeline & Use of Funds
We're raising a pre-seed round to prove Silmari on real customer systems.
Today: a working prototype for personal and institutional memory, production-ready.
Next: load the full 13-year history of the City of Chicago legacy system (over 200,000 emails) and answer the real questions that come in each week, within the first year.
- Handling 200,000-email histories reliably — 12%
- Running on Oracle and SQL Server — 5%
- Sales and partners: reach 63 paying systems by year two — 77%
- Security reviews for running on customers' own servers — 1%
- Operations and everything else — 5%
Every time someone leaves, a company forgets part of how its systems work. Silmari keeps that knowledge.