Bottleneck #01: Tech Debt

In its early days, a startup searches for a great product-market match. When
it finds one it seems to be to develop quickly, a section often called a scaleup. At this
time it is rising quickly alongside many dimensions: revenues, buyer,
headcount. At Thoughtworks, we have labored with many such scaleups, and our
work has targeted on the best way to assist them overcome varied bottlenecks that
impede this progress.

As we have performed this work, we have observed widespread
bottlenecks, and discovered approaches to take care of them. This text is the
first in a sequence that examines these bottlenecks. In every article we’ll look
at how startups get into the bottleneck, normally by doing the fitting
issues which are wanted early in a startup’s life, however are now not proper as
progress adjustments the context for tactics of working. We’ll spotlight key indicators
that the startup is approaching or caught within the bottleneck. We’ll then discuss
about the best way to break by the bottleneck, describing the adjustments we have seen
that permit scaleups to achieve their correct potential.

We begin this sequence by technical debt: how the instruments and
practices that facilitate speedy experimentation of the product/market match
want to vary as soon as progress kicks in.

How did you get into the bottleneck?

The most typical scaling bottleneck we encounter is technical debt —
startups commonly state that tech debt is their predominant obstacle to
progress. The time period “tech debt” tends for use as a catch-all time period,
usually indicating that the technical platform and stack wants
enchancment. They’ve seen function improvement decelerate, high quality points, or
engineering frustration. The startup crew attributes it to technical debt
incurred as a result of an absence of technical funding throughout their progress section.
An evaluation is required to determine the sort and scale of the tech debt.
It may very well be that the code high quality is dangerous, an older language or framework
is used, or the deployment and operation of the product isn’t totally
automated. The answer technique could be slight adjustments to the groups’
course of or beginning an initiative to rebuild components of the appliance.

It’s necessary to say that prudent technical debt is wholesome and desired,
particularly within the preliminary phases of a startup’s journey. Startups ought to
commerce technical facets resembling high quality or robustness for product supply
pace. It will get the startup to its first aim – a viable enterprise
mannequin, a confirmed product and prospects that love the product. However because the
firm seems to be to scale up, we’ve to handle the shortcuts taken, or it
will in a short time have an effect on the enterprise.

Let’s study a few examples we’ve encountered.

Firm A – A startup has constructed an MVP that has proven sufficient
proof (consumer visitors, consumer sentiment, income) for traders and secured
the subsequent spherical of funding. Like most MVPs, it was constructed to generate consumer
suggestions reasonably than high-quality technical structure. After the
funding, as a substitute of rebuilding that pilot, they construct upon it, retaining the
traction by specializing in options. This is probably not an instantaneous drawback
for the reason that startup has a small senior crew that is aware of the sharp edges and
can put in bandaid options to maintain the corporate afloat.

The problems begin to come up when the crew continues to give attention to function
improvement and the debt isn’t getting paid down. Over time, the
low-quality MVP turns into core parts, with no clear path to enhance or
substitute them. There may be friction to be taught, work, and assist the code. It
turns into more and more troublesome to increase the crew or the function set
successfully. The engineering leaders are additionally very nervous concerning the
attrition of the unique engineers and dropping the data they’ve.

Ultimately, the shortage of technical funding involves a head. The crew
turns into paralyzed, measured in decrease velocity and crew frustration. The
startup has to rebuild considerably, that means function improvement has to
decelerate, permitting opponents to catch up.

Firm B – The corporate was based by ex-engineers and so they
wished to do every thing “proper.” It was constructed to scale out of the field.
They used the newest libraries and programming languages. It has a finely
grained structure, permitting every a part of the appliance to be
applied with completely different applied sciences, every optimized to scale
completely. In consequence, it can simply be capable to deal with hyper progress when
the corporate will get there.

The difficulty with this instance is that it took a very long time to create,
function improvement was sluggish, and plenty of engineers hung out engaged on the
platform reasonably than the product. It was additionally laborious to experiment — the
finely grained structure meant concepts that didn’t match into an current
service structure have been difficult to do. The corporate didn’t understand
the worth of the extremely scalable structure as a result of it was not in a position to
discover a product-market match to achieve that scale of buyer base.

These are two excessive examples, based mostly on an amalgamation of assorted
purchasers with whom the startup groups at Thoughtworks have labored. Firm A
obtained itself right into a technical debt bottleneck that paralyzed the corporate.
Firm B over-engineered an answer that slowed down improvement and
crippled its means to pivot shortly because it learnt extra.

The theme with each is an incapability to seek out the fitting stability of technical
funding vs. product supply. Ideally we need to leverage using prudent technical debt to energy
speedy function improvement and experimentation. When the concepts are discovered to
be invaluable, we should always pay down that technical debt. Whereas that is very simply
acknowledged, it may be a problem to place into follow.

To discover the best way to create the fitting stability, we’re going to study the
various kinds of technical debt:

Typical kinds of debt:

Technical debt is an ambiguous time period, typically considered purely
code-related. For this dialogue, we’re going to make use of technical debt to imply
any technical shortcut, the place we’re buying and selling long-term funding right into a
technical platform for short-term function improvement.

Code high quality
Code that’s brittle, laborious to check, laborious to know, or poorly
documented will make all improvement and upkeep duties slower and can
degrade the “enjoyment” of writing code whereas demotivating engineers.
One other instance is a site mannequin and related information mannequin that doesn’t
match the present enterprise mannequin, leading to workarounds.

An absence of unit, integration, or E2E exams, or the flawed distribution
(see take a look at pyramid). The developer can’t shortly get confidence that
their code won’t break current performance and dependencies. This leads
to builders batching adjustments and a discount of deployment frequency.
Bigger increments are more durable to check and can typically end in extra bugs.
Between modules (typically occurs in a monolith), groups probably
block one another, thus decreasing the deployment frequency and
growing lead time for adjustments. One answer is to drag out companies
into microservices, which comes with it’s personal
— there might be extra easy methods of setting
clear boundaries throughout the monolith.

Unused or low worth options
Not usually regarded as technical debt, however one of many signs of
tech debt is code that’s laborious to work with. Extra options creates
extra situations, extra edge instances that builders need to design
round. This erodes the supply pace. A startup is experimenting. We
ought to all the time ensure that to return and re-evaluate if the experiment
(the function) is working, and if not, delete it. Emotionally, it may be very
troublesome for groups to make a judgment name, however it turns into a lot simpler
when you have got goal information quantifying the function worth.

Old-fashioned libraries or frameworks
The crew shall be unable to reap the benefits of new enhancements and
stay susceptible to safety issues. It would end in a abilities
drawback, slowing down the onboarding of recent hires and irritating
present builders who’re pressured to work with older variations. Moreover, these
legacy frameworks are likely to restrict additional upgrades and innovation.

Sub-optimum third-party merchandise or instruments that require a variety of
upkeep. The panorama is ever-changing, and extra environment friendly
tooling could have entered the market. Builders additionally naturally need to
work with probably the most environment friendly instruments. The stability between shopping for vs.
constructing is complicated and wishes reassessment with the remaining debt in

Reliability and efficiency engineering issues
This may have an effect on the shopper expertise and the flexibility to scale. We
need to watch out, as we’ve seen wasted effort in untimely
optimization when scaling for a hypothetical future scenario. It’s higher to
have a product confirmed to be invaluable with customers than an unproven product
that may scale. We’ll describe this in additional element within the piece on
“Scaling Bottleneck: Constructed with out reliability and observability in thoughts”.

Handbook processes
A part of the product supply workflow isn’t automated. This might
be steps within the developer workflow or issues associated to managing the
manufacturing system. A warning: this may additionally go the opposite manner while you
spend a variety of time automating one thing that isn’t used sufficient to be
well worth the funding.

Automated deployments
Early stage startups can get away with a easy setup, however this could
be addressed very quickly — small incremental deployments energy experimental
software program supply. Use the 4 key metrics as your information submit. It’s best to
have the flexibility to deploy at will, normally at the very least as soon as a day.

Data sharing
Lack of helpful data is a type of technical debt. It makes
it troublesome for brand spanking new staff and dependent groups to rise up to hurry.
As commonplace follow, improvement groups ought to produce concisely
written technical documentation, API Specs, and architectural
resolution information. It must also be discoverable by way of a developer
portal or search engine. An anti-pattern isn’t any moderation and
deprecation course of to make sure high quality.

Is that basically technical debt or performance?

Startups typically inform us about being swamped with technical debt, however
underneath examination they’re actually referring to the restricted performance
of the technical platform, which wants its personal correct therapy with
planning, requirement gathering, and devoted assets.

For instance, Thoughtworks’ startup groups typically work with purchasers on
automating buyer onboarding. They could have a single-tenant answer
with little automation. This begins off nicely sufficient — the builders can
manually arrange the accounts and observe the variations between installs.
However, as you add extra purchasers, it turns into too time-consuming for the
builders. So the startup may rent devoted operations workers to set
up the shopper accounts. Because the consumer base and performance grows, it
turns into more and more troublesome to handle the completely different installs —
buyer onboarding time will increase, and high quality issues improve. At
this level automating the deployment and configuration or transferring to a
multi-tenant setup will immediately impression KPIs — that is

Different types of technical debt are more durable to identify and more durable to level
to a direct impression, resembling code that’s troublesome to work with or brief
repeated handbook processes. One of the simplest ways to determine them is with
suggestions from the groups that have them day-to-day. A crew’s
steady enchancment course of can deal with it and shouldn’t require a
devoted initiative to repair it.

How do you get out of the bottleneck?

The method that groups are taking to technical debt ought to come from
its technical technique, set by its leaders. It must be intentional,
clear, and re-evaluated over time. Sadly, we frequently see groups
working off historic instructions, creating future issues with out
realizing it. For a corporation on this circumstance, a number of alternatives
generally set off when to re-evaluate their present technique:

  • New funding means extra options and extra assets — this can compound
    present issues. Addressing present technical debt must be a part of the
    funding plan.
  • New product path can invalidate earlier assumptions and put
    stress on new components of the programs.
  • governance course of includes reevaluating the state of the
    know-how on an everyday cadence.
  • New opinions may also help keep away from “boiling frog” issues. Exterior assist, crew
    rotations and new staff will convey a contemporary perspective.

The slippery slope

How did you find yourself with a variety of technical debt? It may be very laborious to
pinpoint. Usually it isn’t as a result of only one occasion or resolution, however
reasonably a sequence of choices and trade-offs made underneath strain.

Paradoxically, looking back, if one considers every resolution on the level
in time at which it was made, based mostly on what was identified on the
time, it’s unlikely to be thought-about a mistake. Nevertheless, one
concession results in one other and so forth, till you have got a major problem
with high quality. There may be generally a tipping level at which resolving the
tech debt takes extra time than growing incremental worth.

It’s laborious to get well and the scenario tends to snowball. It’s
pure for builders to make use of the present state as an indicator of what
is suitable. In these situations, growing the brand new options will
end in much more debt. That is the slippery slope, a vicious cycle
that sadly results in a cliff as the hassle to implement the subsequent
function will increase non-linearly.

Set a high quality bar

Many organizations discover it useful to have a set of requirements and
practices to which the corporate is dedicated that information technical
evolution. Take into account that some technical practices are fairly
troublesome to attain, for instance steady supply; deploying
commonly with out affecting customers is technically difficult. Groups
typically have preliminary issues, and in response management could deprioritize
the follow. As a substitute we suggest the other, do it extra typically and
your groups will grasp the practices and kind sturdy habits. When the
robust time comes, reasonably than dropping the follow, use the suggestions to
information future funding in crew functionality.

Blast Radius

We settle for that taking shortcuts is a mandatory a part of scaling the
enterprise. How can we restrict the blast radius, figuring out that these shortcuts
will must be resolved, and even completely rebuilt? Clearly, we want a
technique that limits the impression to the enterprise. A technique is to decouple
groups and programs, which permits a crew to introduce tech debt that’s
remoted and received’t essentially snowball as described above.

Prime quality literature about decoupling is plentiful, so we received’t
try to clarify right here. We suggest focusing consideration on
microservices and area pushed design methods. Nevertheless, watch out
doing an excessive amount of too early, decoupling provides latency and complexity to your
programs, and selecting poor area boundaries between groups can add
communication friction. We shall be writing about anti-patterns associated
to overcomplicated distributed architectures in future articles.

Product and Engineering Collaboration

If commerce off conversations aren’t balanced between enterprise technique,
product and engineering, technical high quality mostly degrades first,
and because of this product high quality ultimately suffers as nicely. Once you
search for the foundation explanation for this bottleneck, it almost all the time comes down
to the stability throughout the firm between enterprise, product and
engineering targets. Lack of collaboration usually results in brief
sighted selections made in a vacuum. This may go each methods, chopping
corners in vital areas or gold plating one thing that isn’t invaluable
are equally possible.

  • The enterprise technique at any cut-off date must be clear and clear.
  • We empower crew leaders to make selections which profit the enterprise.
  • Product and Engineering ought to have an equal footing, belief in one another, and
    be prepared to make commerce off selections based mostly on lengthy and brief time period impression to the enterprise.
  • Choices are made with information – e.g. the present state of the technical platform,
    estimates, evaluation of anticipated worth and KPI enchancment, consumer analysis, A/B take a look at outcomes.
  • Choices are revisited when information is refined or new learnings are found.

A tech technique to restrict technical debt impression

When considering of methods for a startup, and the way it scales, we like
to make use of a four-phase mannequin to know the completely different phases of a
startup’s improvement.

Part 1


Prototypes – semi-functional software program to exhibit product,
transferring to purposeful with growing curiosity

Part 2

Getting Traction

Ecosystem selections – cloud vendor, language decisions, service
integration fashion

Change prototype software program for core programs

Setup preliminary foundations – experimentation, CI/CD, API,
observability, analytics

Set up the broad domains, set preliminary tender boundaries (in

Part 3

(Hyper) Development

Create decoupled product groups managing their very own companies

Set up SLAs and high quality bar, linked to indicators round buyer
expertise of product

Set up platform groups targeted on the effectiveness of product

Part 4


Reassess SLA and high quality bar targeted on long run productiveness
and upkeep

Audit state of technical platform, sponsor initiatives in product
groups and create non permanent tiger groups to repair greatest technical debt

Rebuild or purchase capabilities for improved effectivity

Practice groups on good technical high quality practices

How do you handle the tech debt

It begins with clear data sharing how the
enterprise is doing, the present product path, metrics on the present
scaling capability, what prospects are saying concerning the product and what
buyer assist and ops are seeing. This data will permit
technologists to make knowledgeable selections. Sharing the info of the
present problem helps technologists to know why issues are being
addressed and measure their success.

There must be clear end-to-end possession of all merchandise and
their associated programs. As groups develop and take duty for his or her
respective areas, there may be typically no clear possession for an end-to-end
journey, which leaves technical gaps that always develop into stuffed with
technical debt. As groups develop and tackle new duties, it turns into
more and more troublesome to seek out an proprietor for older code. Moreover,
with out possession, groups are much less incentivized to repair issues.

We’ve got to empower groups to repair issues — resolving technical debt ought to
be a part of the pure movement of product improvement. Engineers and product
managers want to barter the wholesome stability between tech debt vs.
performance with the fitting pragmatic mentality. It’s a part of a product
crew’s job to keep up and maintain technically wholesome merchandise, not one thing
performed as an after-thought. There must be an agreed course of to sort out and
monitor technical debt regularly. This requires laborious trade-offs amongst
engineering and product leaders to maintain a steady stability.

Designing your crew topology the fitting
manner may also be an element. For instance, suppose we regularly see
technical debt created in sure areas. In that case, it’d point out
that the crew design is flawed, and there could be a platform or enterprise
functionality that wants sturdy possession and a spotlight.

Some metrics are highly effective — for instance, scanning for widespread
errors or measuring construct and deployment instances. The engineering
group ought to present self-service tooling into which groups
can shortly combine their programs. Metrics must be used as guides
for the crew to make selections about tech-debt reasonably than for managers
to watch or incentivize. Skilled builders present worth by
decoding the accessible information and grounding their intution in fact-based
qualitative data.

Whereas we imagine in autonomous groups, an excessive amount of autonomy is usually a drawback
and may end up in a chaotic technical panorama. There must be light-weight checks and balances such
as automated checks or architectural peer overview, which may also help implement
insurance policies and help builders.

How your group chooses to handle its tech debt is dependent upon your
context. One widespread theme we’ve seen throughout many organizations is the need
to “simply do one thing,” typically leading to a band-aid which quickly creates its
personal set of frictions. As a substitute, we’ve discovered that taking an iterative method
and letting the metrics mixed with present improvement exercise information the funding in resolving tech debt leads to
higher outcomes.

Leave a Comment