Entytle · Yamparala Rahul

What Entytle is

Installed base intelligence for industrial OEM teams: the aftermarket side of heavy equipment, where the money is in parts, service contracts and renewals rather than the original sale.

Only designer, 2022 to 2024: Just over two years, remote, for a US based company.
Three solutions designed as products: Quote Agent, Asset Lifecycle Management, Installed Base Health Check.
A fourth come to a module at a time: Installed Base Intelligence was already shipping: contacts, dynamic pricing, download visibility, automated email.
Retention up 7 percent: Across the three feature products.

Entytle's own marketing site at entytle.com, with the Solution menu open and listing Installed Base Intelligence, Quote Agent, Asset Lifecycle Management and Installed Base Health Check in that order — Entytle's own Solution menu, which is where the list below comes from and the order it runs in. I worked across all four.

Executive Dashboard

The analytics home across all four solutions rather than one product's screen: territory scoped sales, opportunities, pipeline and customer insight, where every summary line is a filter you can act on.

The dashboard

Insyghts, the installed base intelligence product for industrial OEM teams. Everything on the page is scoped by the two controls in the top right, one territory and one date range, so a regional lead and a global one read the same layout about a different book of business.
The date control leads with the four ranges people actually ask for and keeps the calendar for the fifth. Apply is its own step, so a half set range never redraws the page under the reader.
The whole scroll, five bands deep: Sales Overview, Opportunities, Pipeline, Sales Trend, Customer Insights. Click it to read at size. Each band answers a question about the same territory and date range, so nobody has to carry a filter in their head between sections.

1 / 3

The summary is the query

The strongest idea in the product is that the summary line is the query. Top 30 high value opportunities is not a heading, it is a control: it cuts 965 records to those 30, checks them, and points Bulk Send Email at the result. The number stops being something you read and becomes something you act on.
Picking the other summary, 43 opportunities unassigned, swaps the list to those 43 and the actions to Assign To and Change Stage. One salesperson, one save, forty three records that had nobody on them.
Bulk send keeps the merge fields on show as tokens, so a rep can see that contact full name and part detail get filled in before the send rather than after it. The breadcrumb stays behind them, because someone who arrived here from a filter needs to know what the send is about to hit.

1 / 3

The lists it scopes

965 contacts, and the product admits it does not trust all of them. Status reads Unknown, Bad or Verified and is set from the row itself, while 20 contacts are unknown is another filter rather than another warning. The person acting on a contact is the person who can grade it.
Every pipeline carries its own funnel with a count, a share and a value at each stage, so one sitting at a hundred percent Open reads as untouched without opening it. The summary above counts the drafts, the completed, the in progress, and the twelve nobody has updated in ninety days.

1 / 2

The AI panel

Newton, an AI side panel, presenting a thirty thousand dollar upgrade opportunity as named fields for contact, action required, service contract details and amount, with two follow up actions and a notice that mistakes are possible — Newton, 2024. A thirty thousand dollar upgrade opportunity as named fields instead of a paragraph: who to call, what to propose and by when, the contract that justifies it, the amount. Two things to do next, and a line at the bottom saying the model can be wrong. Built before an AI panel was a standard product feature. The Platform tab has the rest of it, including what it was called before and after this.

How it got there

- Iteration 1
- Iteration 2
- Iteration 3, executive view
- Iteration 4
- Iteration 6
- Iteration 7
Six directions before the one that shipped. No single frame here is the point; the point is that the dashboard was argued about rather than styled.
- Hackathon explorations
- Hackathon, further
An internal hackathon took another run at the same problem. Two sheets, laid out side by side because the comparison is the content.

1 / 2

Handed to engineering

The component sheet underneath those directions, which is what made them comparable: when every exploration is built from the same cards, charts and controls, a review argues about the layout instead of the styling.
My own comments on the file, which is the part a finished screen cannot show: numbering the bulk send flow from step zero, asking what high value actually means and who sets the condition, phasing a customisable top X into a second release, and a note to go settle grouping with the engineer. The last frame is the email copy coming back to be edited, approved or sent.

1 / 2

Installed Base Intelligence

Entytle's oldest and broadest solution, already shipping when I arrived, so I came to it a module at a time: who to call and how far to trust the record, what to charge, where a slow export went, and how outreach starts from the machine rather than the address book.

The account page

An account overview in Entytle Insyghts with tabs for overview, equipment, parts, service, service contracts, warranties, contacts, leads, transaction history and audit log, showing account segment and last transaction, cards for locations, five year recurring revenue and open lead value, and revenue and equipment count charts by category and location — The account page, and the reason the solution is called intelligence rather than reporting. Ten tabs of one customer's installed base, and above them the three numbers a rep opens with: how many locations, what it recurs at, what is open. The 88 percent fall on recurring revenue is the kind of thing nobody goes looking for and everybody needs told.

Advanced search

Contacts

29,791 contacts, each carrying a status a person set: verified, bad, or unknown. The tooltip gives the date it was graded, which is the part that makes the grade worth anything a year later. The dashboard shows a 965 record slice of this same list; this is the module it comes from.
The same three states on a phone, scoped to one account and reduced to a dot. This is the view for someone standing in front of the equipment rather than sitting in front of the pipeline, and the grading has to survive the trip.

1 / 2

Pricing intelligence

Every item the model thinks is mispriced, with its suggestion next to the price book rather than instead of it. A gear box at 56 dollars against a book price of 50, a kit at 300 against 320: the direction is not always up, which is what stops this reading as a machine for raising prices. The action on each row is Set Price, so the recommendation ends in a person's decision.
What sits behind a single suggestion: seven years of what this part actually sold for, six of how much of it people wanted, and the four accounts that buy it with their segment and last transaction. A number a person is asked to accept has to be arguable, and this is the page that makes it arguable.
Parts get the same treatment, priced against the open market rather than their own history. The popup is the part that matters: the market price and the average delivery are not asserted, they are shown as the four supplier quotes they were averaged from, each with its own lead time. A rep can disagree with the number and still use the page.

1 / 3

Download visibility

Exporting 893 locations is slow enough that the answer cannot be a spinner. The confirmation says three things: an email is coming, it will take a while, and here is where to watch it. The third is the one that stops people re-requesting.
Every state a long job can be in, written out rather than colour coded and left there. Queued explains why nothing is happening yet, in progress says what it is doing at the moment, failed offers the retry, completed names the address it went to. The tooltip is the rule the list runs on: 24 hours, plus anything still running.
The empty state teaches the rule instead of apologising for the emptiness. A miniature of a filled log with a 24 hour badge on it, so a reader who expected to see yesterday's export learns where it went rather than filing a bug.

1 / 3

Automated email

Outreach starts on the map rather than in a contact list, so the reason for the email is on screen before the email is: this location needs attention at 20 percent, and these are the two people at it worth writing to. Ticking a contact is what builds the send.
The draft arrives knowing what the customer owns: the chiller, when they bought it, the part due next month, and what happens if they skip it. That is the automated part, and it is why the rep is editing a specific letter rather than filling in a blank one. Draft for Me is there for the case where the template is wrong. The dashboard's bulk send is the same idea at scale, with merge fields shown as tokens.
The confirmation says where to go next rather than only that it worked. A small thing, and the same instinct as the download activity log: a send that disappears is a send someone repeats.

1 / 3

Quote Agent

Quoting for parts and service, with the model drafting and a person deciding.

Before and after

Before the product: a customer emails the manufacturer asking what four part numbers cost, and waits somewhere between one and three days for a person to read it, price it and reply.
The same loop with the agent in the middle. The claim on the slide is twelve hours, and the rest of the panel is where those hours actually go.

1 / 2

The queue

Every inbound request becomes a row with a status: received, processing, verified, processed, failed. A failed row says why on hover rather than leaving someone to guess. This is also where the product calls itself Order Agent, though the menu that sells it says Quote Agent.
The email the request came from stays one click from the row it produced. Anyone checking the agent's reading can see the sentence it read.

1 / 2

Verifying a quote

The screen the work actually happens on. The email sits beside the four line items pulled out of it, so verifying is reading one against the other rather than trusting a total. Location carries a warning because the agent could not fill it, which is the product admitting what it does not know instead of guessing.
The draft the model produces, and three ways out of it. CSV, XML and JSON exist because the quote has to land in whatever the OEM already runs, which is the unglamorous half of shipping AI into a business that has systems.
The override. When the scan misses an item or reads it wrong, a person searches the catalogue and adds it, so a bad extraction costs a search rather than the whole quote.

1 / 3

Sending it back

A slide showing the Process modal with a generated Quote.pdf attached, a subject line and a message, and a Send Email action returning the quote to the requester — Processing sends the quote back to the address it came from, as a PDF with a message on it. The loop closes where it opened, in the requester's inbox.

Asset Lifecycle Management

Where every machine in a customer's installed base sits in its service life, and what the next service on it is worth.

Before the product

Configuring a lifecycle

Step two, defining equipment lifetime in years and naming service events as major, annual or operational with a trigger time in months, with an option to add service events from the bill of materials — The form that replaced the spreadsheet. Pick the asset, define its lifetime and the service events that hang off it, allocate parts and checklist items to each event, attach the manuals. The summary rail on the right fills in as you go, so a long form is never a black box about what it has so far.

What it tells you

The same question, answered for 28 machines at once: what stage each is at, what service is due, and what that service is worth. The Competitor column is the quiet one, because the installed base counts machines the customer did not sell.
The slider is the idea: recommendations for the next Today to twenty years. Scrub it forward and the recommended service and its value move with it, so a maintenance schedule reads as a revenue forecast. Underneath, the manual's checks are structured data instead of a PDF.
Every asset configured, and the last column is the one that matters. The machine sold for fifty thousand dollars; the service life around it is worth 1.2 million. That number is what the wizard exists to produce, and it is the whole aftermarket argument in one column.

1 / 3

Keeping it current

The Asset Lifecycle Management list with a row's actions revealed: a download menu offering the manual or the checklist, and a view lifecycle control — Managing what was configured. Service events carry a repeat flag, so a one time commissioning inspection models differently from a recurring overhaul, and a part can belong to several events at once. The spreadsheet's caveat is a Remark field here, which is how it survives all the way out to the rep.

A Service Events Info drawer listing initial inspection at six months, periodic inspection at two years and overhaul at five years, each with whether it repeats — Managing what was configured. Service events carry a repeat flag, so a one time commissioning inspection models differently from a recurring overhaul, and a part can belong to several events at once. The spreadsheet's caveat is a Remark field here, which is how it survives all the way out to the rep.

The alternative

A competitor product, myOEM, showing its asset page for a compressor: asset image and information, an asset status panel with availability and last connected, an asset score gauge, and a sensors tab charting live package outlet and discharge pressure over a date range — myOEM, the competitor, and not my work. Theirs is telemetry first: live pressure curves, last connected, an asset score, a dynamic warranty. Entytle reads the same machine from ship date, service history and the manufacturer's intervals instead, which is the difference between needing the asset connected and not.

Installed Base Health Check

Inspections as a product, and the other half of Asset Lifecycle Management: the service ALM says is due is the inspection this one authors, schedules and hands back filled in.

Authoring a checklist

The manufacturer's inspection table, authored as a form instead of read off a PDF. Sections and questions, an answer type chosen per question, and a reference image so the person in the field can see the part they are being asked about.
Authoring and previewing are the same screen, and the preview defaults to the device the work happens on. Question 1 of 26 as the person in the plant will meet it: one question, the machine above it, two answers. Everything the desk version carries is gone, because none of it helps someone standing in front of a conveyor.

1 / 2

Scheduling it

An inspection is assembled from checklists that already exist, each one bound to a business unit and then narrowed to a product line and item number. The panel on the left says what will appear there rather than sitting blank, which matters because it is empty for the whole first half of the task.
Scheduling binds a checklist to one machine: business unit, product line, serial. The same inspection can be reused across a product line without ever being ambiguous about which asset it was run on.

1 / 2

Tracking it

Where the scheduled ones live. Five states rather than done and not done: completed, in progress, planned, not scheduled, overdue. Overdue is its own state instead of a date a reader has to compare against today, which is the difference between a list you scan and a list you act on.
Opened from a row rather than by navigating away, and it carries all three checklists on that inspection. Checking what you are about to send someone into a plant with should cost a click, not a round trip.

1 / 2

What comes back

An inspection report in Entytle Insyghts showing date, account, location, who scheduled it and who inspected, an introduction paragraph, and a table of every question with its answer, an attached file, and the inspector's comment — The review page: every question with its answer, the file the inspector attached, and the note they left, all still tied to the serial it was scheduled against. The loop that opened in a spreadsheet closes on the same machine.

Platform

The parts that belong to no single solution: an assistant that reaches every surface, and the account security page underneath all of them.

The assistant

The assistant opens the conversation rather than waiting to be asked. Three opportunities, already ranked, and one button to see them. A blank chat box on a dashboard is a question nobody knows how to answer; this is the version that has done the work before you arrive.
Each opportunity arrives as fields a rep already uses: who to contact, what is required and by when, the contract that justifies it, the amount. Not prose to be read and re-typed. The follow ups underneath are the next questions written out, because knowing what to ask an assistant is most of the difficulty.
- EnvisionAI, on a blue banner
- Entytle Newton, on a white one
The same screen and the same suggestion, under two different names. Across these captures the assistant is called Einstein AI, EnvisionAI, Entytle Newton and Installed Base GPT; here two of them sit side by side. Naming an AI feature inside a product that already has a name is its own design problem, and this one went several rounds.
Opened, it shows its working: what it is proposing, why (their service history and similar equipment's tickets), the equipment details it read, and who to call. Then Create Opportunity, so the answer ends in a record rather than in a rep's memory. Every field is one a person can check and disagree with.
The third altitude: ask it outright. A rep pastes the location they are visiting tomorrow and gets the same structure back, opportunity, equipment, who to call, rather than a different personality for the chat surface. The controls under the reply are the honest part, since copy, regenerate and a thumbs down are what a person needs when a model is confidently wrong.

1 / 5

Account security